New Updates


Table of contents
I Thanks 7
II Preface 8
1  Introduction 9
1.1  Definition 9
1.2  Comparison to maps, GIS, aerial photography / Photogrammetry, SONAR   10
1.2.1  Satellite Images Vs Maps 10
1.2.2  Remote Sensing Vs GIS 10
1.2.3  Remote Sensing Vs Aerial Photography / Photogrammetry 10
1.2.4  Remote Sensing Vs SONAR 11
1.3  Applications in general 12
1.3.1 Agriculture 12
1.3.2 Forestry 12
1.3.3 Geology 13
1.3.4 Hydrology 13
1.3.5 Sea Ice 14
1.3.6 Land Cover & Land Use 14
1.3.7  Mapping 14
1.3.8  Oceans & Coastal Monitoring 15
2 Electromagnetic radiation 17
2.1 Electromagnetic energy 17
2.2 Interaction mechanisms 17
2.3 Laws regarding the amount of energy radiated from an object 18
2.3.1 Planck Radiation Law 18
2.3.2 Wien’s displacement law 19
2.3.3 Black body concept, Emissivity and Radiant Temperature 20
2.4 Electromagnetic Spectrum 21
2.4.1 Wavelength bands 22
2.4.2 Atmosphere effects 22 Scattering 22 Absorption 23
2.4.3 Reflectance spectra 25 Mixtures 27 Grain Size Effects 28 The Continuum and Band Depth 30 Continuum-Removed Spectral Feature Comparison 31  Viewing Geometry 33
3 Sensors 34
3.1 History 34
3.2 Satellite Characteristics: Orbits and Swaths 34
3.3 Scanner Sensor Systems 37
3.4 Spatial Resolution, Pixel Size, and Scale 38
3.5 Spectral / Radiometric resolution 40
3.5.1 Spectral charaecteristics 40 Spectral range 41 Spectral bandwidth 42 Spectral sampling 44 Signal-to-noise ratio 45
3.6 Temporal Resolution 45
3.7 Overview of different sensors – satellites and airborne 46
3.7.1 Comparison Table 46 Weather Satellites/Sensors 47 GOES 47 National Oceanic and Atmospheric Administration's Advanced
Very High Resolution Radiometer 483 Land Observation Satellites/Sensors 48 Landsat Multispectral Scanner 48 Landsat Thematic Mapper 49 Systeme Probatoire d'Observation de la Terra (SPOT) High
 Resolution Visible Sensor 49  Indian Remote Sensing 50  IKONOS 51 Marine Observation Satellites/Sensors 52 Coastal Zone Colour Scanner (CZCS) 52 MOS 53 SeaWiFS 53 Laser fluorosensor - another kind of sensor 53 Hyperspectral sensors 53 Compact Airborne Spectrographic Imager CASI 54 Digital Airborne Imaging Spectrometer DAIS 7915 54 AVIRIS Airborne Visible InfraRed Imaging Spectrometer 54 Synthetic Aperture Radar Sensors 55
4 Corrections 56
4.1  Radiometric calibration 56
4.1.1  Main elements of sensor calibration 56  Absolute radiometric calibration – from radiance to DN and back 56  Uniformity calibration 57  Spectral calibration 57  Geometric calibration 58
4.1.2  Calibration approaches 58  Prelaunch calibration 58  Onboard calibration 59  Vicarious calibration 59
4.2  Atmospheric - from radiance to reflectance or to temperatute\emissivity       60
4.2.1  Calibrating images from different dates to like-values 62
4.2.2  Internal Average Realtive Reflectance (IARR) 63
4.2.3  Flat Field 63
4.2.4  Empirical line 63
4.2.5  Atmospheric modelling 64  Band transmittance computer models 66  Line-by-line models 67  MODTRAN 67  2 nd  simulation of satellite signal in the solar spectrum – 6s code 69  ATmospheric REMoval program (ATREM) 70  ATCOR 72
4.2.6  Temperature calibration of images 73
4.2.7  Thermal properties of materials 73
4.2.8  Retrieval of temperature and emissivity from radiance in thermal images      77
4.3 Geometric corrections 79
4.3.1 Geometric registration 80  Plane transformations 81  Polynomial transformations 83  Triangulation 83  Ground Control Points 84  Resampling 85  Relief displacement 86
4.3.2  LANDSAT – geometric characteristics 90  TM geometric accuracy 90  TM data processing levels 90  Raw data 90  System corrected products 90  Geocoded products 91  Level A – without ground control points 91  Level B – with ground control points 914  Level C – with ground control points plus DTM 91
4.3.3 SPOT – processing levels 92  Level 1A – no geometric correction 92  Level 1B - compensating internal geometric distortions 92  Level 1AP - Photographic product for photogrametric use with
analogue devices 92  Level 2A – entry level for cartographic products – projected,
no control points 93  Level 2B – geocoded product, projected, with ground control points 93  Level ORTHO - The ultimate level of preprocessing for the best
cartographic accuracy: correction of the residual parallax errors
brought by the relief 93
4.3.4 Parametric Geocoding (based on the PARGE alogrithm) 94
5 Image processing 98
5.1 Storage formats 98
5.2 Image enhancment: 100
5.2.1 Histogram, stretching, colour palettes 101 Contrast enhancement 101 Linear Stretch 101 Histogram equalization 102 Piece-wise linear stretch 102 RGB false colour composit 104 Colour Definition Models, fusion of TM and SPOT 104 Colour Definition Models 104 Red, Green, and Blue (RGB) 104 Hue, Saturation, and Lightness (HSL) 105          CMYK Model 105          Fusion by HSV Transformation 105 Spatial filters – noise reduction (low-pass), edge enhancment (high-pass)     106 The basic principles of a filter 106 Low pass filters 107 High pass filters 110
5.3 Multi-band operations 113
5.3.1 Image ratios: Brightness variations 113
5.3.2 Normalized Difference Vegetation Index 114
5.3.3 Principal components analysis 114
5.3.4 Image classification 117 Density slicing 117 Multi-spectral image classification 118 Supervised Classification 119 Sampling 119 Classification 119 Accuracy assessment 121 Unsupervised classification (clustering) 126
5.3.5 Unmixing 126 Modeling mixed spectra 126 Practical unmixing methods 127
6 Active Remote Sensing 129
6.1 Side Looking Airborne Radar (SLAR): 129
6.1.1 Frequencies 132
6.1.2 Polarization 133
6.1.3 Viewing Geometry and Spatial Resolution  133
6.1.4 Radar image distortions 136
6.1.5 Target interaction and image appearance 138
6.1.6 Radar image properties 143
6.1.7 Advanced Radar applications 1455
7 Remote Sensing Applications for the sea
(passive and SLAR) 148
7.1  Sea Surface Temperature 148
7.2  Oil Spill Detection 150
7.2.1  Case study 1 – oil slick 151
7.2.2  Case study 2 – oil seep 151
7.3  Ice motion and monitoring 152
7.3.1 Case study (example): Operational ice monitoring with ERS-1 SAR 153
7.4  Mapping the sea floor (bathymetry) 155
7.4.1 Case study – SHOM’s nautical space chart (La spatiocarte marine) 155
7.5  Vessel detection 158
8  Digital video Remote Sensing 161
8.1  A little history 161
8.2  General advantages of using video 161
8.3  The video image 161
8.4  Charge-Coupled Devices 163
8.5  The geometric resolution of video 164
8.6  Airborne video data acquisition 165
8.7  Types of airborne video systems 166
8.7.1  Multiband information 166
8.7.2  Multiplex systems 166
8.7.3  Single camera systems 167
8.8 The Silvacam camera 167
8.8  Applications 169
8.8.1  Coastal monitoring in the Netherlands 169
8.8.2  The Southwest Washington Coastal Erosion Study,
and the ARGUS program 170
9  Altimetry 172
9.1  Laser altimetry 173
9.2  Radar altimetry 173
9.3 Radar altimetry over the oceans 174
9.3.1 Measuring Ocean Topography for Understanding and Predicting
Climate Change 174
9.3.2 Data sources 175  Geosat follow-on 175  TOPEX/Poseidon 175  Mission requirements 175  Sensors on board the TOPEX/POSEIDON 176  Orbit of the TOPEX/POSEIDON 177  Data retrieval 177  Jason 177  ERS-2 177  NRL layered ocean model 179  Modular ocean data assimilation system (MODAS) 179
9.3.3  Data processing 179  Initial Data Processing 179  Interpolation 180  Tide removal 180  Orbit error removal 181  Referencing to a consistent mean 181
9.3.4 Altimetry products and derived products 182  Dynamic Sea Surface Topography - (from Altitude) 183  Sea Surface Variability 184  Wind Speed - (from backscatter coefficient) 184 Scatterometry 186  Ocean circulation 187  Significant Wave Height - (from Wave Form Leading Edge) 1886  Watervapor 188  Marine gravity and sea-floor topography 189 Global Bathymetric Prediction for Ocean Modelling and
Marine Geophysics 190
9.4 Airborne laser scanning (ALS) 195
9.4.1  Introduction – laser principles 195  Pulse lasers 196  Wavelength 196  Scanning 197  Position and orientation system 198  Typical processing steps 198  Some extended laser capabilities 198  A short overview of applications 199  Airborne Laser Scanning vs Photogrammetry – a comparison 199
9.4.2  Laser remote sensing of forest structure 201 Multi-Beam Laser Altimeter (MBLA) - The Lidar Instrument
for the Vegetation Canopy Lidar (VCL) 204
9.5 Scanning laser mapping of the coastal zone 205
9.5.1  Historic development 205
9.5.2  Working principles 206
9.5.3  Benefits 208
9.5.4  Case study: The ALACE project - Airborne LIDAR Assessment
 of Coastal Erosion Project 209
9.5.5  IHO and FEMA standards, and their relation to ALS technology 213  IHO S-44 213  FEMA 215  AIRBORNE LIGHT DETECTION AND RANGING SYSTEMS -
General Guidelines for Use 215  Performance Standards 215  GPS Control 216  Post-Processing of Data 216  Quality Control/Quality Assurance 217  Deliverables 218
10 Metadata 219
10.1 GIS metadata 219
10.1.1 GIS metadata worked example – Bathymetry of the Gulf of
Carpentaria and the Arafura Sea 220
10.2  Remote Sensing metadata 222
11 References and links 224
11.1  Basic books about remote sensing 224
11.2  Remote Sensing journals 224
11.3  On-line tutorials 224
11.4  Remote Sensing softwares 225
11.5  Other Remote Sensing links 2257
First, I would like to express my thanks to the following people, for making this work
•   The staff of IMO-IMA, for supplying me with the appropriate working conditions
(free access to the Internet, the academy’s library) and helping whenever needed,
•   the  staff  of  the  Earth  Sciences  library  of  the  Trieste  university,  for  their  help  in
lending books and making photocopies of articles,
•   all  the  people  and  institutions  that  maintain  internet  sites,  and  publish  on-line
articles,  tutorials,  technical  documentations,  etc  (references  mentioned  along  the
text, the principal sources at the end of the text),
•   Dr.  Eyal  Ben  Dor  from  the  geography  department  of  Tel  Aviv’s  university  for
introducing me to the field of Remote Sensing, and
•   to my family and friends.8
§     Craig  J.M.  (1998),  The  application  of  satellite  imagery  in  support  of  nautical  charting;  past
experience  and  future  possibilities  -  a  practical  view,  International  Hydrographic  Review,
LXXV(1), no. 142, pp. 95-105
The field of Remote Sensing is very wide, both in the data acquisition methods, data
processing procedures and techniques and the applications it is used for; it is also a
fast developing field, in all the above themes. This text is therefore intended to give
only a general overview about several subjects, yet I hope, an extensive one, covering
all the important topics regarding Remote Sensing of the surface of the Earth. The text
also attempts to give the reader an understanding of the capabilities and limitations of
Remote  Sensing.  Very  few  equations  and  formulas  will  be  given  in  the  text,  as  the
focus will be on understanding the basic ideas. The main subjects covered are:
•   The role Remote Sensing plays in our understanding of the Earth and the natural
and human processes affecting it,
•   The radiometric and geometric principles of Remote Sensing,
•   The  principal  sensors  used,  and  their  characteristics,  in  passive  and  active,
imaging and non-imaging Remote Sensing, on airborne or on satellite platforms,
from monochromatic to hyperspectral,
•   The pre processing phase of data: radiometric, atmospheric, geometric and noises
corrections, and
•   The image processing phase of data: visualisation, enhancement and classification.
Due to the wide scope covered, the subjects could not be covered in details and the
interested reader should turn to the relevant literature.
Detailed examples of Remote Sensing applications will be given in areas which have
direct  importance  for  either  hydrography  or  oceanography.  The  advantages  of
acquiring  information  by  Remote  Sensing  apply,  irrespective  of  platform  or  sensor,
also for hydrography:
1.  It is cheaper than conventional surveying;
2.  It is safer than hydrographic surveying in shoal areas such as coral reefs;
3.  It  is  capable  of  change  detection  in  rapidly  developing  ports  and  regular
monitoring of mobile  areas such as deltas and sandbanks;
4.  World-wide  coverage  is  comercially  available,  without  security,  political,  or
copyright restrictions, enabling dataa acquisition from remote areas;
5.  The inherent geometry and therefore the relative positioning of features within a
single scene is generally very good.
Through out the work all the references from which I have taken material are given,
though not in the formal academic way.9
1. Introduction:
1.1 Definition
1.  The  Definition  of  Remote  Sensing  In  the  broadest  sense,  the  measurement  or
acquisition  of  information  of  some  property  of  an  object  or  phenomenon,  by  a
recording  device  that  is  not  in  physical  or  intimate  contact  with  the  object  or
phenomenon  under  study;  e.g.,  the  utilization  at  a  distance  (as  from  aircraft,
spacecraft,  or  ship)  of  any  device  and  its  attendant  display  for  gathering
information  pertinent  to  the  environment,  such  as  measurements  of  force  fields,
electromagnetic  radiation,  or  acoustic  energy.  The  technique  employs  such
devices as the camera, lasers, and radio frequency receivers, radar systems, sonar,
seismographs, gravimeters, magnetometers, and scintillation counters.
2.  The  practice  of  data  collection  in  the  wavelengths  from  ultraviolet  to  radio
regions.   This   restricted   sense   is   the   practical   outgrowth   from   airborne
photography. Sense (1) is preferred and thus includes regions of the EM spectrum
as  well  as  techniques  traditionally  considered  as  belonging  to  conventional
As humans, we are intimately familiar with remote sensing in that we rely on visual
perception  to  provide  us  with  much  of  the  information  about  our  surroundings.  As
sensors,  however,  our  eyes  are  greatly  limited  by  1)  sensitivity  to  only  the  visible
range of electromagnetic energy; 2) viewing perspectives dictated by the location of
our bodies; and 3) the inability to form a lasting record of what we view. Because of
these  limitations,  humans  have  continuously  sought  to  develop  the  technological
means  to  increase  our  ability  to  see  and  record  the  physical  properties  of  our
Beginning  with  the  early  use  of  aerial  photography,  remote  sensing  has  been
recognized  as  a  valuable  tool  for  viewing,  analyzing,  characterizing,  and  making
decisions about our environment. In the past few decades, remote sensing technology
has  advanced  on  three  fronts:  1)  from  predominantly  military  uses  to  a  variety  of
environmental analysis applications that relate to land, ocean, and atmosphere issues;
2) from (analog) photographic systems to sensors that convert energy from many parts
of the electromagnetic spectrum to electronic signals; and 3) from aircraft to satellite
Today, we define satellite remote sensing as the use of satellite-borne sensors to
observe, measure, and record the electromagnetic radiation reflected or emitted by the
Earth and its environment for subsequent analysis and extraction of information.10
1.2 Comparison to maps, GIS, aerial photography / Photogrammetry, SONAR
Here will be given the main points of similarity and difference between the field of
Remote Sensing (analysis and images) and the fields/products mentioned above.
1.2.1 Satellite Images Vs Maps
§     Anson R.W. and Ormeling F.J., (1993), Basic Cartography – for students and technicians, Vol 1.,
§     Portugaly Y. (1996), “The maps hidden from the eye”, Mishkafayim, 27, pp. 44-47 (in Hebrew)
According  to  the  International  Cartographic  Union,  a  map  is  “a  conventionalised
image   representing   selected   features   or   characteristics   of   geographical   reality,
designed  for  use  when  spatial  relationships  are  of  primary  importance”.  This
definition does declare that in every map there’s a process of selection present (and in
addition  - symbolization, abstraction and generalization), but also keeps the aura of
scientific  accuracy  of  a  map.  But,  we  should  remember,  that  “a  map  shows  us  the
world as we know it, and what we know, is a very complex subject, that is comprised
•   The limits of matter, technology and our measurement tools,
•   what we believe that exists,
•   what we think to be important,
•   and what we want and aspire to”
Thus,  a  map  is  a  subjective,  for  we  always  decide  what  to  put  on  it,  and  how  to
represent  it.  A  Remote  Sensing  image  in  contrast,  is  an  objective  recording  of  the
Electromagnetic reaching the sensor.
Another  important  difference,  is  that  a  map  is  a  projection  of  the  earth  on  paper,
without  any  relief  displacements,  while  in  a  Remote  Sensing  image  both  relief
displacements and geometrical distortions.
1.2.2 Remote Sensing Vs GIS
GIS (Geographic Information System) is a kind of software that enables:
•   The collection of spatial data from different sources (Remote Sensing being one of
•   Relating spatial and tabular data.
•   Performing tabular and spatial analysis.
•   Symbolize and design the layout of a map.
A  GIS  software  can  handle  both  vector  and  raster  data  (some  handle  only  one  of
them).  Remote  Sensing  data  belongs  to  the  raster  type,  and  usually  requires  special
data  manipulation  procedures  that  regular  GIS  does  not  offer.  However,  after  a
Remote Sensing analysis has been done, its results are usually combined within a GIS
or into database of an area, for further analysis (overlaying with other layers, etc). In
the last years, more and more vector capabilities are being added to Remote Sensing
softwares, and some Remote Sensing functions are inserted into GIS modules.
1.2.3 Remote Sensing Vs Aerial Photography / Photogrammetry
Both  systems  gather  data  about  the  upper  surface  of  the  Earth,  by  measuring  the
Electromagnetic  radiation,  from  airborne  systems.  The  following  major  differences
can be given:
•   Aerial  photos  are  taken  by  an  analog  instrument:  a  film  of  a  (photogrammetric)
camera, then scanned to be transformed to digital media. Remote Sensing data is
usually gathered by a digital CCD camera.11
•   The advantage of a film is its high resolution (granularity), while the advantage of
the  CCD  is  that  we  measure  quantitatively  the  radiation  reaching  the  sensor
(radiance  values,  instead  of  a  gray-value  scale  bar).  Thus,  Remote  Sensing  data
can be integrated into physical equations of energy-balance for example.
•   An Aerial photograph is a central projection, with the whole picture taken at one
instance.  A  Remote  Sensing  image  is  created  line  after  line;  therefor,  the
geometrical  correction  is  much  more  complex,  with  each  line  (or  even  pixel)
needing to be treated as a central projection.
•   Aerial  photos  usually  gather  data  only  in  the  visible  spectrum  (there  are  also
special films sensitive to near infrared radiation), while Remote Sensing sensors
can be designed to measure radiation all along the Electromagnetic spectrum.
•   Aerial  photos  are  usually  taken  from  planes,  Remote  Sensing  images  also  from
•   Both systems are affected by atmospheric disturbances. Aerial photos mainly from
haze  (that  is,  the  scattering  of  light  –  the  process  which  makes  the  sky  blue),
Remote   Sensing   images   also   from   processes   of   absorption.   Atmospheric
corrections to Aerial photos can be made while taking the picture (using a filter),
or  in  post-processing,  as  in  done  Remote  Sensing.  Thermal  Remote  Sensing
sensors  can  operate  also  at  nighttime,  and  Radar  data  is  almost  weather
•   In Photogrammetry the main efforts are dedicated for the accurate creation of a 3d
model, in order to plot with high accuracy the location and boundaries of objects,
and  to  create  a  Digital  Elevation  Model,  by  applying  sophisticated  geometric
corrections. In Remote Sensing the main efforts are dedicated for the analysis of
the    incoming    Electromagnetic    spectrum,    using    atmospheric    corrections,
sophisticated  statistical  methods  for  classification  of  the  pixels  to  different
categories,  and  analysing  the  data  according  to  known  physical  processes  that
affect the light as it moves in space and interacts with objects.
•   Remote  Sensing  images  are  very  useful  for  tracking  phenomena  on  regional,
continental and even global scale, using the fact that satellites cover in each image
a wide area, and taking images all the time (whether fixed above a certain point, or
“revisiting” the same place every 15 days (for example).
•   Remote  Sensing  images  are  available  since  the  early  1970’s.  Aerial  photos,
provide a longer time span for landscape change detection (the regular coverage of
Israel by Aerial photos started in 1944/5, for example, with many Aerial photos
taken also during World War 1).
•   Remote  Sensing  images  are  more  difficult  to  process,  and  require  trained
personnel, while aerial photographs can be interpreted more easily.
1.2.4 Remote Sensing Vs SONAR
The  SONAR  can  also  be  considered  as  Remote  Sensing  –  that  is,  studying  the
surfaces of the sea (bathymetry and sea bed features) from a distance. The SONAR is
an active type of Remote Sensing (like Radar; Not depending on an external source of
waves, measuring the time between the transmission and reception of waves produced
by   our   instruments,   and   their   intensity),   but   using   sound   waves,   and   not
Electromagnetic radiation.
Both  systems  transmit  waves  through  an  interfering  medium  (water,  air),  that  adds
noise to the data we are looking for, and there for corrections must be applied to the12
raw  data  collected.  In  Remote  Sensing  however,  Radar  is  considered  to  be  almost
weather  independent,  and  atmospheric  disturbances  affect  mainly  passive  Remote
Sensing). To make these necessary corrections, both systems depend on callibration
from field data (be it salinity, temperature and pressure measured by the ship while
surveying,    or    measurements    of    the    atmospheric    profile    parameters    by    a
meteorological radiosonde for example).
Sonar’s are mainly used to produce the bathymetry of the sea, while Remote Sensing
techniques are focusing more on identification of the material’s properties than on its
Echo-sounders (single or multi-beam) can be compared to Airborne Laser Scanning –
both of them create point (vector) data containing X,Y,Z, that needs to be further post
processed in order to remove noise (spikes). An added complexity when dealing with
bathymetry (as opposed to topography) is the need for tide corrections.
Side Scan SONAR can be compared to Side Looking Aperture RADAR, both of them
creating images (raster) analyzing the surface.
Another major difference is that in Remote Sensing the results of the analysis can be
compared  easily  to  the  field  (aerial  photos,  maps,  field  measurments),  while  in
SONAR the underlying bottom of the sea is hidden from us, and we depend totally on
the data gathered.
1.3 Applications in general
As  will  be  learned  in  the  section  on  sensors,  each  one  is  designed  with  a  specific
purpose.  With  optical  sensors,  the  design  focuses  on  the  spectral  bands  to  be
collected. With radar imaging, the incidence angle and microwave band used plays an
important role in defining which applications the sensor is best suited for.
Each application itself has specific demands, for spectral resolution, spatial resolution,
and  temporal  resolution.  There  can  be  many  applications  for  Remote  Sensing,  in
different  fields,  as  described  below.  In  the  body  of  this  tutorial  of  Remote  Sensing,
some applications relevant for hydrography and oceanography will be given in more
1.3.1 Agriculture
Agriculture plays a dominant role in economies of both developed and undeveloped
countries. Satellite and airborne images are used as mapping tools to classify crops,
examine  their  health  and  viability,  and  monitor  farming  practices.  Agricultural
applications of remote sensing include the following:
•   crop type classification
•   crop condition assessment
•   crop yield estimation
•   mapping of soil characteristics
•   mapping of soil management practices
•   compliance monitoring (farming practices)
1.3.2 Forestry
Forests are a valuable resource providing food, shelter, wildlife habitat, fuel, and daily
supplies  such  as  medicinal  ingredients  and  paper.  Forests  play  an  important  role  in
balancing  the  Earth's  CO2  supply  and  exchange,  acting  as  a  key  link  between  the
atmosphere, geosphere, and hydrosphere.13
Forestry applications of remote sensing include the following:
•   reconnaissance mapping:
Objectives to be met by national forest/environment agencies include forest cover
updating, depletion monitoring, and measuring biophysical properties of forest
•   Commercial forestry:
Of importance to commercial forestry companies and to resource management
agencies are inventory and mapping applications: collecting harvest information,
updating of inventory information for timber supply, broad forest type, vegetation
density, and biomass measurements.
•   Environmental monitoring:
Conservation authorities are concerned with monitoring the quantity, health, and
diversity of the Earth's forests.
1.3.3 Geology
Geology involves the study of landforms, structures, and the subsurface, to understand
physical  processes  creating  and  modifying  the  earth's  crust.  It  is  most  commonly
understood as the exploration and exploitation of mineral and hydrocarbon resources,
generally to improve the conditions and standard of living in society.
Geological applications of remote sensing include the following:
•   surficial deposit / bedrock mapping
•   lithological mapping
•   structural mapping
•   sand and gravel (aggregate) exploration/ exploitation
•   mineral exploration
•   hydrocarbon exploration
•   environmental geology
•   geobotany
•   baseline infrastructure
•   sedimentation mapping and monitoring
•   event mapping and monitoring
•   geo-hazard mapping
•   planetary mapping
1.3.4 Hydrology
Hydrology is the study of water on the Earth's surface, whether flowing above ground,
frozen in ice or snow, or retained by soil
Examples of hydrological applications include:
•   wetlands mapping and monitoring,
•   soil moisture estimation,
•   snow pack monitoring / delineation of extent,
•   measuring snow thickness,
•   determining snow-water equivalent,
•   river and lake ice monitoring,
•   flood mapping and monitoring,
•   glacier dynamics monitoring (surges, ablation)
•   river /delta change detection
•   drainage basin mapping and watershed modelling14
•   irrigation canal leakage detection
•   irrigation scheduling
1.3.5 Sea Ice
Ice covers a substantial part of the Earth's surface and is a major factor in commercial
shipping and fishing industries, Coast Guard and construction operations, and global
climate change studies.
Examples of sea ice information and applications include:
•   ice concentration
•   ice type / age /motion
•   iceberg detection and tracking
•   surface topography
•   tactical identification of leads: navigation: safe shipping routes/rescue
•   ice condition (state of decay)
•   historical ice and iceberg conditions and dynamics for planning purposes
•   wildlife habitat
•   pollution monitoring
•   meteorological / global change research
1.3.6 Land Cover & Land Use
Although  the  terms  land  cover  and  land  use  are  often  used  interchangeably,  their
actual  meanings  are  quite  distinct.  Land  cover  refers  to  the  surface  cover  on  the
ground, while Land use refers to the purpose the land serves. The properties measured
with  remote  sensing  techniques  relate  to  land  cover,  from  which  land  use  can  be
inferred, particularly with ancillary data or a priori knowledge.
Land use applications of remote sensing include the following:
•   natural resource management
•   wildlife habitat protection
•   baseline mapping for GIS input
•   urban expansion / encroachment
•   routing  and  logistics  planning  for  seismic  /  exploration  /  resource  extraction
•   damage delineation (tornadoes, flooding, volcanic, seismic, fire)
•   legal boundaries for tax and property evaluation
•   target  detection  -  identification  of  landing  strips,  roads,  clearings,  bridges,
land/water interface
1.3.7 Mapping
Mapping  constitutes  an  integral  component  of  the  process  of  managing  land
resources,  and  mapped  information  is  the  common  product  of  analysis  of  remotely
sensed data.
Mapping applications of remote sensing include the following:
•   Planimetry:
Land surveying techniques accompanied by the use of a GPS can be used to meet
high  accuracy  requirements,  but  limitations  include  cost  effectiveness,  and
difficulties in attempting to map large, or remote areas. Remote sensing provides a
means  of  identifying  and  presenting  planimetric  data  in  convenient  media  and
efficient manner. Imagery is available in varying scales to meet the requirements
of  many  different  users.  Defence  applications  typify  the  scope  of  planimetry15
applications  -  extracting  transportation  route  information,  building  and  facilities
locations, urban infrastructure, and general land cover.
•   digital elevation models (DEM's):
Generating DEMs from remotely sensed data can be cost effective and efficient. A
variety  of  sensors  and  methodologies  to  generate  such  models  are  available  and
proven  for  mapping  applications.  Two  primary  methods  if  generating  elevation
data  are  1.  Stereogrammetry  techniques  using  airphotos  (photogrammetry),  VIR
imagery, or radar data (radargrammetry), and 2. Radar interferometry.
•   Baseline thematic mapping / topographic mapping:
As   a   base   map,   imagery   provides   ancillary   information   to   the   extracted
planimetric  or  thematic  detail.  Sensitivity  to  surface  expression  makes  radar  a
useful  tool  for  creating  base  maps  and  providing  reconnaissance  abilities  for
hydrocarbon and mineralogical companies involved in exploration activities. This
is  particularly  true  in  remote  northern  regions,  where  vegetation  cover  does  not
mask   the   microtopography   and   generally,   information   may   be   sparse.
Multispectral imagery is excellent for providing ancillary land cover information,
such  as  forest  cover.  Supplementing  the  optical  data  with  the  topographic  relief
and  textural  nuance  inherent  in  radar  imagery  can  create  an  extremely  useful
image composite product for interpretation.
1.3.8 Oceans & Coastal Monitoring
The oceans not only provide valuable food and biophysical resources, they also serve
as transportation routes, are crucially important in weather system formation and CO2
storage, and are an important link in the earth's hydrological balance. Coastlines are
environmentally  sensitive  interfaces  between  the  ocean  and  land  and  respond  to
changes  brought  about  by  economic  development  and  changing  land-use  patterns.
Often coastlines are also biologically diverse inter-tidal zones, and can also be highly
urbanized .
Ocean applications of remote sensing include the following:
•   Ocean pattern identification:
•   currents, regional circulation patterns, shears
•   frontal zones, internal waves, gravity waves, eddies, upwelling zones, shallow
water bathymetry ,
•   Storm forecasting
•   wind and wave retrieval
•   Fish stock and marine mammal assessment
•   water temperature monitoring
•   water quality
•   ocean productivity, phytoplankton concentration and drift
•   aquaculture inventory and monitoring
•   Oil spill
•   mapping and predicting oilspill extent and drift
•   strategic support for oil spill emergency response decisions
•   identification of natural oil seepage areas for exploration
•   Shipping
•   navigation routing
•   traffic density studies
•   operational fisheries surveillance
•   near-shore bathymetry mapping16
•   Intertidal zone
•   tidal and storm effects
•   delineation of the land /water interface
•   mapping shoreline features / beach dynamics
•   coastal vegetation mapping
•   human activity / impact17
2. Electromagnetic radiation:
Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
2.1 Electromagnetic energy
Electromagnetic energy refers to all energy that moves with the velocity of light  in a
harmonic  wave  pattern.  The  word  harmonic  implies  that  the  component  waves  are
equally and repetitively spaced in time. The wave concept explains the propagation of
Electromagnetic energy, but this energy is detectable only in terms of its interaction
with matter. In this interaction, Electromagnetic energy behaves as though it consists
of  many  individual  bodies  called  photons  that  have  such  particle-like  properties  as
energy and momentum.
Electromagnetic waves can be described in terms of their:
•   Velocity: The speed of light, c=3*10 8  m*sec -1 ).
•   Wavelength: l, the distance from any position in a cycle to the same position in
the  next  cycle,  measured  in  the  standard  metric  system.  Two  units  are  usually
used: the micrometer (mm, 10 -6 m) and the nanometer (nm, 10 -9 m).
•   Frequency: n, the number of wave crests passing a given point in specific unit of
time, with one hertz being the unit for a frequency of one cycle per second.
Wavelength and frequency are related by the following formula:
c = l* n
Electro-Magnetic  radiation    consists  of  an  electrical  field  (E)  which  varies  in
magnitude  in  a  direction  perpendicular  to  the  direction  in  which  the  radiation  is
traveling, and a magnetic field (M) oriented at right angles to the electrical field. Both
these fields travel at the speed of light (c).
Figure: Electro-Magnetic radiation Figure: wavelength and frequency
2.2 Interaction mechanisms
A  number  of  interactions  are  possible  when  Electromagnetic  energy  encounters
matter, whether solid, liquid or gas. The interactions that take place at the surface of a
substance  are  called  surface  phenomena.  Penetration  of  Electromagnetic  radiation
beneath  the  surface  of  a  substance  results  in  interactions  called  volume  phenomena.
The surface and volume interactions with matter can produce a number of changes in
the  incident  Electromagnetic  radiation;  primarily  changes  of  magnitude,  direction,
wavelength,  polarization  and  phase.  The  science  of  Remote  Sensing  detects  and
records  these  changes.  The  resulting  images  and  data  are  interpreted  to  identify
remotely  the  characteristics  of  the  matter  that  produced  the  changes  in  the  recorded
Electromagnetic radiation.
The following interactions may occur:18
•   Radiation may be transmitted, that is, passed through the substance. The velocity
of  Electromagnetic  radiation  changes  as  it  is  transmitted  from  air,  or  a  vacuum,
into other substances.
•   Radiation  may  be  absorbed  by  a  substance  and  give  up  its  energy  largely  to
heating the substance.
•   Radiation  may  be  emitted  by  a  substance  as  a  function  of  its  structure  and
temperature. All matter at temperatures above absolute zero, 0°K, emits energy.
•   Radiation may be scattered, that is, deflected in all directions and lost ultimately
to absorption or further scattering (as light is scattered in the atmosphere).
•   Radiation  may  be  reflected.  If  it  is  returned  unchanged  from  the  surface  of  a
substance with the angle equal and opposite to the angle of incidence, it is termed
specular  reflectance  (as  in  a  mirror).  If  radiation  is  reflected  equally  in  all
directions, it is termed diffuse. Real materials lie somewhere in between.
The  interactions  with  any  particular  form  of  matter  are  selective  with  regard  to  the
Electromagnetic  radiation  and  are  specific  for  that  form  of  matter,  depending
primarily upon its surface properties and its atomic and molecular structure.
2.3 Laws regarding the amount of energy radiated from an object
2.3.1 Planck Radiation Law
The primary law governing blackbody radiation is the Planck Radiation Law, which
governs the intensity of radiation emitted by unit surface area into a fixed direction
(solid angle) from the blackbody as a function of wavelength for a fixed temperature.
The Planck Law can be expressed through the following equation.
(through water)
(through the
Every  object  with  a  temperature  above  the  absolute  zero  radiates  energy.  The
relationship  between  wavelength  and  the  amount  of  energy  radiated  at  different
wavelengths, is shown in the following figure, and formulated above.
The figure shows the emitted radiance of the Earth and the Sun as Black Bodies (each given on a
different Y scale), according to Plank Law, and also the effective radiance of the sun reaching the
Earth. It can be seen, that from about 2-3 mm the radiance emitted from the Earth is greater than
reaching us from the Sun (both of them presented on the same scale).
2.3.2 Wien’s displacement law
For  an  object  at  a  constant  temperature  the  radiant  power  peak  refers  to  the
wavelength at which the maximum amount of energy is radiated, which is expressed
as l max . The sun, with a surface temperature of almost 6000°K, has its peak at 0.48mm
(wavelength  of  yellow).  The  average  surface  temperature  of  the  earth  is  290°K
(17°C),  which  is  also  called  the  ambient  temperature;  the  peak  concentration  of
energy emitted from the earth is at 9.7mm.
This shift to longer wavelengths with decreasing temperature is described by Wien’s
displacement law, which states:
 l max  = 2,897mm°K / T rad °K  .20
2.3.3 Black body concept, Emissivity and Radiant Temperature
Temperature is a measure of the concentration of heat. The concentration of kinetic
heat of a body of material may be called the kinetic temperature, T kin  and is measured
by a thermometer placed in direct contact with the material.
By definition a black body is a material that absorbs all the radiant energy that strikes
it. A black body also radiates the maximum amount of energy, which is dependent on
the kinetic temperature. According to the Stefan-Boltzman  law  the  radiant  flux  of  a
black body, F b , at a kinetic temperature, T kin , is F b  = s* T kin
4   where s is the Stefan-
Boltzman constant, 5.67*10 -12  W*cm -2 *°K -4 .
A black body is a physical abstraction, for no material has an absolute absorptivity,
and no material radiates the full amount of energy as defined in the equation above.
For real materials a property called emissivity, e, has been defined, as e=F r /F b , where
F r  is radiant flux from a real material. For a black body e=1, but for all real materials
e<1.  Emissivity  is  wavelength  dependent,  which  means  that  the  emissivity  of  a
material  is  different  when  is  measured  at  different  wavelengths  of  radiant  energy
(each material has both a reflectance spectrum and an emissivity spectrum – see later).
In  the  next  table  are  given  the  average  emissivities  of  various  materials  in  the  8  to
12mm wavelength region (thermal), which is used in Remote Sensing.
Material                                                               Emissivity, e
Polished metal surface 0.006
Granite 0.815
Quartz sand, large grains 0.914
Dolomite, polished 0.929
Basalt, rough 0.934
Asphalt paving 0.959
Concrete walkway 0.966
A coat of flat black paint 0.970
Water, with a thin film of petroleum 0.972
Water, pure 0.993
Thus,  the  radiant  flux  of  a  real  material  may  be  expressed  as  F r   =  e*s*  T kin
4   .
Emissivity is a measure of the ability of a material to both radiate and absorb energy.
Materials with a high emissivity absorb and radiate large proportions of incident and
kinetic energy, respectively (and vice-versa). The result, is that two surfaces, with the
same kinetic temperature but with a different emissivity will have a different radiant
temperature. As Remote Sensing devices measure the radiant temperature, in order to
derive the kinetic temperature of the object, we need to know its emissivity, that is, w
need  to  be  able  to  identify  the  material  (see  later,  in  the  chapter  about  atmospheric
corrections). Then we can apply the following equation: T rad =e 1/4 T kin  .21
2.4 Electromagnetic Spectrum
The Electromagnetic spectrum is the continuum of energy ranging from kilometers to
nanometers  in  wavelength.  This  continuum  is  commonly  divided  into  the  following
ranges, called spectral bands, the boundaries between them being gradational.
A passive Remote Sensing system records the energy naturally radiated or reflected
from an object. An active Remote Sensing system supplies its own source of energy,
which  is  directed  at  the  object  in  order  to  measure  the  returned  energy.  Flash
photography  is  active  Remote  Sensing  in  contrast  to  available  light  photography,
which is passive. An other common form of active Remote Sensing is radar, which
provides its own source of Electromagnetic energy in the microwave region. Airborne
laser scanning is a relatively new form of active Remote Sensing, operating the in the
visible and Near Infra Red wavelength bands.
Active system Passive system22
2.4.1 Wavelength bands
Band Wavelength Remarks
Gamma ray <0.03 nm Incoming radiation from the sun is completely absorbed by the
upper atmosphere, and is not available for Remote Sensing.
Gamma radiation from radioactive minerals is detected by low-
flying aircraft as a prospecting method.
X-ray 0.03 to 0.3 nm       Incoming radiation is completely absorbed by atmosphere. Not
employed in Remote Sensing.
Ultraviolet, UV      3 nm to 0.4 mm      Incoming UV radiation atmosphere wavelengths <0.3 mm is
completely absorbed by ozone in the upper atmosphere.
0.3 to 0.4 mm    Transmitted through the atmosphere. Detectable with film and
photodetectors, but atmospheric scattering is severe.
Visible 0.4 to 0.7 mm         Detected with film and photodetectors. Includes earth reflectance
peak at about 0.5 mm.
Infrared, IR 0.7 to 300 mm        Interaction with matter varies with wavelength. Atmospheric
transmission windows are separated by absorption bands.
Reflected IR       0.7 to 3 mm This is primarily reflected solar radiation and contains no
information about thermal properties of materials. Commonly
divided into the following regions:
•     Near Infra Red (NIR) between 0.7 to 1.1 mm.
•     Middle Infra Red (MIR) between 1.3 to 1.6 mm.
•     Short Wave Infra Red (SWIR) between 2 to 2.5 mm.
Radiation from 0.7 to 0.9 mm is detectable with film and is called
photographic IR radiation.
Thermal IR         3 to 5 mm
8 to 14 mm
These are the principal atmospheric windows in the thermal
region. Imagery at these wavelengths is acquired through the use
of optical-mechanical scanners, not by film.
Microwave 0.3 to 300 cm        These longer wavelengths can penetrate clouds and fog. Imagery
may be acquired in the active or passive mode.
Radar 0.3 to 300 cm        Active mode of microwave Remote Sensing.
2.4.2 Atmosphere effects
Our eyes inform us that the atmosphere is essentially transparent to light, and we tend
to assume that this condition exists for all Electromagnetic radiation. In fact, however,
the  gases  of  the  atmosphere  selectively  scatter  light  of  different  wavelengths.  The
gases  also  absorb  Electromagnetic  energy  at  specific  wavelength  intervals  called
absorption  bands.  The  intervening  regions  of  high  energy  transmittance  are  called
atmospheric transmission bands, or windows. The transmission and absorption bands
are  shown  in  the  following  figure,  together  with  the  gases  responsible  for  the
absorption bands.
Particles  and  gases  in  the  atmosphere  can  affect  the  incoming  light  and  radiation.
These effects are caused by the mechanisms of scattering and absorption. Scattering
Scattering  occurs  when  particles  or  large  gas  molecules  present  in  the  atmosphere
interact with and cause the electromagnetic radiation to be redirected from its original
path.  How  much  scattering  takes  place  depends  on  several  factors  including  the
wavelength of the radiation, the abundance of particles or gases, and the distance the
radiation travels through the atmosphere. There are three (3) types of scattering which
take place.23
Rayleigh scattering occurs when particles are very small compared to the wavelength
of the radiation. These could be particles such as small specks of dust or nitrogen and
oxygen  molecules.  Rayleigh  scattering  causes  shorter  wavelengths  of  energy  to  be
scattered  much  more  than  longer  wavelengths.  Rayleigh  scattering  is  the  dominant
scattering mechanism in the upper atmosphere. The fact that the sky appears "blue"
during  the  day  is  because  of  this  phenomenon.  As  sunlight  passes  through  the
atmosphere,  the  shorter  wavelengths  (i.e.  blue)  of  the  visible  spectrum  are  scattered
more than the other (longer) visible wavelengths. At sunrise and sunset the light has to
travel farther through the atmosphere than at midday and the scattering of the shorter
wavelengths  is  more  complete;  this  leaves  a  greater  proportion  of  the  longer
wavelengths to penetrate the atmosphere (thus the sky is “painted” in red).
Mie  scattering  occurs  when  the  particles  are  just  about  the  same  size  as  the
wavelength  of  the  radiation.  Dust,  pollen,  smoke  and  water  vapour  are  common
causes of Mie scattering which tends to affect longer wavelengths than those affected
by  Rayleigh  scattering.  Mie  scattering  occurs  mostly  in  the  lower  portions  of  the
atmosphere where larger particles are more abundant, and dominates when cloud
conditions are overcast.
The final scattering mechanism of importance is called nonselective scattering. This
occurs when the particles are much larger than the wavelength of the radiation. Water
droplets  and  large  dust  particles  can  cause  this  type  of  scattering.  Nonselective
scattering gets its name from the fact that all wavelengths are scattered about equally.
This  type  of  scattering  causes  fog  and  clouds  to  appear  white  to  our  eyes  because
blue,  green,  and  red  light  are  all  scattered  in  approximately  equal  quantities
(blue+green+red light = white light). Absorption
Absorption  is  the  other  main  mechanism  at  work  when  electromagnetic  radiation
interacts  with  the  atmosphere.  In  contrast  to  scattering,  this  phenomenon  causes
molecules in the atmosphere to absorb energy at various wavelengths. Ozone, carbon
dioxide, and water vapour are the three main atmospheric constituents which absorb
Any effort to measure the spectral properties of a material through a planetary
atmosphere, must consider where the atmosphere absorbs. For example, the Earth's
atmospheric transmittance is shown in next Figure. The drop toward the ultraviolet is
due to scattering and strong ozone absorption at wavelengths short of 0.35 µm. Ozone
also displays an absorption at 9.6 µm. Oxygen absorbs at 0.76 µm in a narrow feature.
CO 2  absorbs at 2.01, 2.06, and a weak doublet near 1.6 µm. Water causes most of the
rest of the absorption throughout the spectrum and hides additional (weaker)
absorptions from other gases. The mid-IR spectrum in following Figure shows the
effect of doubling CO 2 , which in this case is small compared to the absorption due to
water. While we will see that the spectral region near 1.4 and 3 µm can be diagnostic24
of OH-bearing minerals, we can't usually use these wavelengths when remotely
measuring spectra through the Earth's atmosphere (it has been done from high
elevation observatories during dry weather conditions). Those areas of the spectrum
which are not severely influenced by atmospheric absorption and thus, are useful to
remote sensors, are called atmospheric windows.
 However, these spectral regions can be used in the laboratory where the atmospheric
path  lengths  are  thousands  of  times  smaller,  or  when  measuring  spectra  of  other
planets from orbiting spacecraft.
Figure: Modtran (Berk et al., 1989) modeled atmospheric transmittance, visible to near-infrared. Most
of the absorptions are due to water. Oxygen occurs at 0.76 µm, carbon dioxide at 2.0 and 2.06 µm.25
Figure: Atmospheric transmittance, mid-infrared is compared to scaled grey-body spectra. Most of the
absorption  is  due  to  water.  Carbon  dioxide  has  a  strong  15-µm  band,  and  the  dotted  line  shows  the
increased  absorption  due  to  doubling  CO 2 .  Also  shown  is  the  black-body  emission  at  288  K  and  the
grey-body emission from water and a sandstone scaled to fit on this transmittance scale. The water and
sandstone curves were computed from reflectance data using: 1 - reflectance times a black-body at 288
Figure: wavelengths we can use most effectively
2.4.3 Reflectance spectra
Spectroscopy of Rocks and Minerals, and Principles of Spectroscopy by Roger N. Clark
Albedo, is defined as the ratio of the amount of electromagnetic energy reflected by a
surface to the amount of energy incident upon it. It differs from spectral reflectance,
since  usually  albedo  is  averaged  over  the  visible  range  of  the  Electro-Magnetic
spectrum,  while  the  term  reflectance  relates  to  a  specific  wavelength  (or  a  specific
band of a satellite).
The complex interaction of light with matter involves reflection and refraction from
index  of  refraction  boundaries,  a  process  we  call  scattering,  and  absorption  by  the
medium  as  light  passes  through  the  medium.  The  amount  of  scattering  versus
absorption  controls  the  amount  of  photons  we  receive  from  a  surface.  As  each
material has its own unique interaction with light, we can find the unique reflectance
spectra of each material. Absorption bands in the spectra of materials are caused by
two general processes: electronic and vibrational (however, these processes will not
be explained here). In addition, the reflectance spectra of a material is also dependent
upon the mixture ratio of the material with other, and the material’s grain size.
As  an  example  for  the  above  dicussion,  are  given  below  the  reflectance  spectra  of
some organic materials26
Figure:  Reflectance  spectra  of  montmorillonite,  and  montmorillonite  mixed  with  super  unleaded
gasoline, benzene, toluene, and trichlorethylene. Montmorillonite has an absorption feature at 2.2 µm,
whereas the organics have a CH combination band near 2.3 µm.
Spectra of vegetation come in two general forms: green and wet (photosynthetic), and
dry non-photosynthetic but there is a seemingly continuous range between these two
end members. The spectra of these two forms are compared to a soil spectrum in the
next Figure. Because all plants are made of the same basic components, their spectra
appear  generally  similar.  However,  in  the  spectral  analysis  section  we  will  see
methods for distinguishing subtle spectral details. The near-infrared spectra of green27
vegetation are dominated by liquid water vibrational absorptions. The water bands are
shifted to slightly shorter wavelengths than in liquid water, due to hydrogen bonding.
The absorption in the visible is due to chlorophyll, and is discussed in more detail in
Chapter  XX  of  Manual  of  Remote  Sensing.  The  dry  non-photosynthetic  vegetation
spectrum  shows  absorptions  due  to  cellulose,  lignin,  and  nitrogen.  Some  of  these
absorptions  can  be  confused  with  mineral  absorptions,  unless  a  careful  spectral
analysis is done.
Figure : Reflectance spectra of photosynthetic (green) vegetation, non-photosynthetic (dry) vegetation,
and  a  soil.  The  green  vegetation  has  absorptions  short  of  1  µm  due  to  chlorophyll.  Those  at
wavelengths greater than 0.9 µm are dominated by liquid water. The dry vegetation shows absorptions
dominated  by  cellulose,  but  also  lignin  and  nitrogen.  These  absorptions  must  also  be  present  in  the
green vegetation, but can be detected only weakly in the presence the stronger water bands. The soil
spectrum shows a weak signature at 2.2 µm due to montmorillonite. Mixtures
The real world (and for that matter, the universe) is a complex mixture of materials, at
just about any scale we view it. In general, there are 4 types of mixtures:
1) Linear Mixture. The materials in the field of view are optically separated so there is
no multiple scattering between components. The combined signal is simply the sum of
the  fractional  area  times  the  spectrum  of  each  component.  This  is  also  called  areal
2)  Intimate  Mixture.  An  intimate  mixture  occurs  when  different  materials  are  in
intimate contact in a scattering surface, such as the mineral grains in a soil or rock.
Depending  on  the  optical  properties  of  each  component,  the  resulting  signal  is  a
highly non-linear combination of the end-member spectra.
3)  Coatings.  Coatings  occur  when  one  material  coats  another.  Each  coating  is  a
scattering/transmitting  layer  whose  optical  thickness  varies  with  material  properties
and wavelength.
4) Molecular Mixtures. Molecular mixtures occur on a molecular level, such as two
liquids,  or  a  liquid  and  a  solid  mixed  together.  Examples:  water  adsorbed  onto  a
mineral; gasoline spilled onto a soil. The close contact of the mixture components can28
cause band shifts in the adsorbate, such as the interlayer water in montmorillonite, or
the water in plants.
An example mixture comparison is shown in the next Figure for alunite and jarosite.
Note in the intimate mixture how the jarosite dominates and the 0.4 to 1.3-µm region.
The reason is because in an intimate mixture, the darker material dominates because
photons  are  absorbed  when  they  encounter  a  dark  grain.  In  the  areal  mixture,  the
brighter material dominates.
Figure. Reflectance spectra of alunite, jarosite and mixtures of the two. Two mixture types are shown:
intimate  and  areal.  In  the  intimate  mixture,  the  darker  of  the  two  spectral  components  tends  to
dominate, and in and areal mixture, the brighter component dominates. The areal mixture is a strictly
linear  combination  and  was  computed  from  the  end-members,  whereas  the  intimate  mixture  is  non-
linear and the spectrum of the physical mixture was measured in the laboratory. The jarosite dominates
the 0.3-1.4 µm wavelength region in the intimate mixture because of the strong absorption in jarosite at
those wavelengths and because the jarosite is finer grained than the alunite and tends to coat the larger
alunite grains. Grain Size Effects
The  amount  of  light  scattered  and  absorbed  by  a  grain  is  dependent  on  grain  (see
figures of ice spectra for example). A larger grain has a greater internal path where
photons  may  be  absorbed  according  to  Beers  Law.  It  is  the  reflection  from  the
surfaces and internal imperfections that control scattering. In a smaller grain there are
proportionally more surface reflections compared to internal photon path lengths, or
in  other  words,  the  surface-to-volume  ratio  is  a  function  of  grain  size.  If  multiple
scattering  dominates,  as  is  usually  the  case  in  the  visible  and  near-  infrared,  the
reflectance decreases as the grain size increases. However, in the mid-infrared, where
absorption coefficients are much higher and the index of refraction varies strongly at
the  Christensen  frequencies,  first  surface  reflection  is  a  larger  or  even  dominant
component of the scattered signal. In these cases, the grain size effects are much more
complex, even reversing trends commonly seen at shorter wavelengths.29
Figure  22a.  The  near-infrared  spectral  reflectance  of  A)  a  fine  grained  (~50  µm)  water  frost,  B)
medium  grained  (~200  µm)  frost,  C)  coarse  grained  (400-2000  µm)  frost  and  D)  an  ice  block
containing abundant microbubbles. The larger the effective grain size, the greater the mean photon path
that  photons  travel  in  the  ice,  and  the  deeper  the  absorptions  become.  Curve  D  is  very  low  in
reflectance because of the large path length in ice. The ice temperatures for these spectra are 112-140
K. From Clark et al. (1986).
Figure 22b. A series of reflectance spectra of melting snow. The top curve (a) is at 0 o  C and has only a
small amount of liquid water, whereas the lowest spectrum (j) is of a puddle of about 3 cm of water on
top of the snow. Note in the top spectrum, there is no 1.65-µm band as in the ice spectra in figure 22a
because  of  the  higher  temperature..  The  1.65-µm  feature  is  temperature  dependent  and  decreases  in
strength  with  increasing  temperature  (see  Clark,  1981a  and  references  therein).  Note  the  increasing
absorption at about 0.75 µm and in the short side of the 1-µm ice band, as more liquid water forms. The
liquid  water  becomes  spectrally  detectable  at  about  spectrum  e,  when  the  UV  absorption  increases.
Spectra from Clark, King and Swayze, in preparation.30 The Continuum and Band Depth.
Absorptions in a spectrum have two components: continuum and individual features.
The continuum is the "background absorption" onto which other absorption features
are superimposed (e.g. see Clark and Roush, 1984). It may be due to the wing of a
larger  absorption  feature.  For  example,  in  the  pyroxene  spectra  in  Figure  21a,  the
weak  feature  at  2.3  µm  is  due  to  a  trace  amount  of  tremolite  in  the  sample  and  the
absorption  is  superimposed  on  the  broader  2-µm  pyroxene  band.  The  broader
pyroxene  absorption  is  the  continuum  to  the  narrow  2.3-µm  feature.  The  pyroxene
1.0-µm  band  is  superimposed  on  the  wing  of  a  stronger  absorption  centered  in  the
The depth of an absorption band, D, is usually defined relative to the continuum, R c :
D = 1 - R b  / R c  (2)
where  R b   is  the  reflectance  at  the  band  bottom,  and  R c   is  the  reflectance  of  the
continuum at the same wavelength as R b  (Clark and Roush, 1984).
The depth of an absorption is related to the abundance of the absorber and the grain
size  of  the  material.  Consider  a  particulate  surface  with  two  minerals,  one  whose
spectrum  has  an  absorption  band.  As  the  abundance  of  the  second  mineral  is
increased, the band depth, D, of the absorption in the first mineral will decrease. Next
consider  the  visual  and  near-infrared  reflectance  spectrum  of  a  pure  powdered
mineral. As the grain size is increased from a small value, the absorption-band depth,
D, will first increase, reach a maximum, and then decrease. This can be seen with the
pyroxene  spectra  in  Figure  21a  and  more  so  in  the  ice  spectra  in  Figure  22.  If  the
particle size were made larger and larger, the reflectance spectrum would eventually
consist only of first surface reflection, like at most wavelengths beyond 1.45 µm in
the  ice  spectra  in  Figure  22.  The  reflectance  can  never  go  to  zero  because  of  this
reflection, unless the index of refraction of the material is 1.0. These concepts, called
"band saturation" are explored further in Clark and Lucey (1984) and Lucey and Clark
A sloping continuum causes an apparent shift in the reflectance minimum, as shown
in  the  above  Figure.  Continuaa  can  be  thought  of  as  an  additive  effect  of  optical
constants, but in reflectance spectra, scattering and Beers Law make the effects non-
linearly  multiplicative  (see  Clark  and  Roush,  1984  for  more  details).  So  the
continuum should be removed by division, whether you are working in reflectance or
emittance. The continuum should be removed by subtraction only when working with
absorption coefficients. In a spectrum with a sloping continuum, correction removes
the  effect  of  shifts  in  the  local  reflectance  minimum.  Note  your  perception  of
spectrum E versus A in the Figure. The spectral features do not appear to be the same,
but if you remove the continuum, it is obvious they are the same (Figure, top).31
Figure: Illustration of continua and continuum removal. In the lower set of curves, the local minimum
in the curve shifts to shorter wavelengths with increasing slope. removal of the continuum by division
isolates the spectral features so they may be compared (top). The top set of curves are offset for clarity.
In the continuum-removed spectra, we can see there is no real shift in the absorption-band center. Continuum-Removed Spectral Feature Comparison.
The  continuum-removal  process  isolates  spectral  features  and  puts  them  on  a  level
playing  field  so  they  may  be  intercompared.  Continuum  removal  and  feature
comparison,  is  in  this  author's  opinion,  the  key  to  successful  spectral  identification.
For example, compare the spectra of calcite (CaCO 3 ) and dolomite (CaMg(CO 3 ) 2 ). If
we isolate the spectral features, remove the continuum, and scale the band depth (or
band area) to be equal, we can see subtle band shifts and shapes:
Figure 24a. Comparison of calcite and dolomite continuum-removed features. The dolomite absorption
occurs at a shorter wavelength than the calcite absorption.
One  of  the  most  challenging  spectral  features  to  distinguish  between  are  those  in
spectra  of  various  plant  species.  Figure  a  shows  four  plant  spectra  (the  spectra  are32
offset for clarity). The overall shapes are quite similar. If we remove the continuum
according, we see the detailed chlorophyll absorption spectral variations for these as
well  as  other  plants  in  Figure  b.  Shape  matching  algorithms,  like  that  presented  in
Clark  et  al.  (1990b),  can  distinguish  between  these  spectra  and  accurately  compare
them to spectral libraries.
Figure a. Reflectance spectra of four types of vegetation. Each curve is offset by 0.05 reflectance unit
from the one below. From Clark et al. (1995, 1997).
Figure  b.  Continuum-removed  chlorophyll  absorptions  for  8  vegetation  types  (including  the  4  from
Figure  a)  showing  that  the  continuum  removed  features  can  show  subtle  spectral  differences.  From
Clark et al. (1995, 1997).33 Viewing Geometry
We  have  seen  tremendous  variation  in  the  spectral  properties  of  minerals  and
materials in general, due to composition, grain size, and mixture types. So far viewing
geometry  has  not  been  discussed.  Viewing  geometry,  including  the  angle  of
incidence,  angle  of  reflection,  and  the  phase  angle:  the  angle  between  the  incident
light  and  observer  (the  angle  of  reflection),  all  affect  the  intensity  of  light  received.
Varying the viewing geometry results in changes in shadowing and the proportions of
first  surface  to  multiple  scattering  (e.g.  Hapke,  1993;  Nelson,  1986;  Mustard  and
Pieters, 1989), which can affect band depths a small amount except in rare cases (like
extreme  specular  reflection  off  a  mirror  or  lake  surface).  While  measuring  precise
light levels are important for things like radiation balance studies, they are of lesser
importance in spectral analysis. The following illustrates why.
First, your eye is a crude spectrometer, able to distinguish the spectral properties of
materials  in  a  limited  wavelength  range  by  the  way  we  interpret  color.  Pick  up  any
colored object around you. Change the orientation of the local normal on the surface
of the object with respect to the angle of incident light, the angle at which you observe
it  (called  the  emission  or  scattering  angle),  and  the  angle  between  the  incident  and
scattered light (the phase angle). As you do this, note any color changes. Unless you
chose an unusual material (like a diffraction grating or very shiny object), you will see
no  significant  color  change.  Plant  leaves  appear  green  from  any  angle,  a  pile  of
hematite appears red from any angle. This tells you that the spectral features do not
change  much  with  viewing  geometry.  Your  eye/brain  combination  normalizes
intensity variations so that you see the same color, regardless of the actual brightness
of the object (the amount of light falling on the surface). The continuum removal does
a similar but more sophisticated normalization. The band depth, shape, and position
are basically constant with viewing geometry. Band depth will only change with the
proportion  of  specular  reflection  added  to  the  reflected  light.  For  surfaces  (and  at
wavelengths)  where  multiple  scattering  dominates,  that  change  in  band  depth  is
3. Sensors:
In order for a sensor to collect and record energy reflected or emitted from a target or
surface, it must reside on a stable platform removed from the target or surface being
observed. Platforms for remote sensors may be situated on the ground, on an aircraft
or balloon (or some other platform within the Earth's atmosphere), or on a spacecraft
or satellite outside of the Earth's atmosphere.
Ground-based sensors are often used to record detailed information about the surface
which  is  compared  with  information  collected  from  aircraft  or  satellite  sensors.  In
some cases, this can be used to better characterize the target which is being imaged by
these  other  sensors,  making  it  possible  to  better  understand  the  information  in  the
imagery. Sensors may be placed on a ladder, scaffolding, tall building, cherry-picker,
crane, etc. Aerial platforms are primarily stable wing aircraft, although helicopters are
occasionally used. Aircraft are often used to collect very detailed images and facilitate
the collection of data over virtually any portion of the Earth's surface at any time. In
space,  remote  sensing  is  sometimes  conducted  from  the  space  shuttle  or,  more
commonly, from satellites. Satellites are objects which revolve around another object
- in this case, the Earth. For example, the moon is a natural satellite, whereas man-
made satellites include those platforms launched for remote sensing, communication,
and  telemetry  (location  and  navigation)  purposes.  Because  of  their  orbits,  satellites
permit repetitive coverage of the Earth's surface on a continuing basis. Cost is often a
significant factor in choosing among the various platform options.
3.1 History
Since  the  early  1960s,  numerous  satellite  sensors  have  been  launched  into  orbit  to
observe  and  monitor  the  Earth  and  its  environment.  Most  early  satellite  sensors
acquired  data  for  meteorological  purposes.  The  advent  of  earth  resources  satellite
sensors  (those  with  a  primary  objective  of  mapping  and  monitoring  land  cover)
occurred when the first Landsat satellite was launched in July 1972. Currently, more
than a dozen orbiting satellites of various types provide data crucial to improving our
knowledge of the Earth's atmosphere, oceans, ice and snow, and land.
3.2 Satellite Characteristics: Orbits and Swaths
We learned in the previous section that remote sensing instruments can be placed on a
variety  of  platforms  to  view  and  image  targets.  Although  ground-based  and  aircraft
platforms may be used, satellites provide a great deal of the remote sensing imagery
commonly used today. Satellites have several unique characteristics which make them
particularly useful for remote sensing of the Earth's surface.
We can calculate the height of a satellite above the Earth, using the following laws:
Gravitation Force       F = GmM/R 2
Centrifugal Force       F = v 2  * m / R35
Since the gravitation force is equal to the centrifugal force in an orbiting satellite, the
gravitational force or mass of the planet is proportional to the velocity squared of any
of it's orbiting satellites.
Centrifugal Force = Gravitation Force
v 2  m / R = GmM/R 2
to simplify the equation, we can take the orbit of the satellite to be on a circle, thus:
v = 2pR/T
with the following,
m         Mass of the satellite (is of no importance for the calculation)
M        Mass of the Earth, 5.976e+24 kg
R         Distance between the center of the Earth and the satellite (consider the
equatorial radius of the Earth to be about 6,378.14 km)
v          Velocity of the satellite
T         Period of the orbit of the satellite around the Earth
G         Newtonian constant of gravitation , 6.6725985e-11 m 3  kg -1  s -2
the distance of a satellite from the face of the Earth can be easily calculated:
R 3  = GMT 2 /4p 2
The path followed by a satellite is referred to as its orbit. Satellite orbits are matched
to the capability and objective of the sensor(s) they carry. Orbit selection can vary in
terms  of  altitude  (their  height  above  the  Earth's  surface)  and  their  orientation  and
rotation  relative  to  the  Earth.  Satellites  at  very  high  altitudes,  which  view  the  same
portion   of   the   Earth's   surface   at   all   times   have   geostationary   orbits.   These
geostationary  satellites,  at  altitudes  of  approximately  36,000  kilometres,  revolve  at
speeds which match the rotation of the Earth so they seem stationary, relative to the
Earth's   surface.   This   allows   the   satellites   to   observe   and   collect   information
continuously  over  specific  areas.  Weather  and  communications  satellites  commonly
have  these  types  of  orbits.  Due  to  their  high  altitude,  some  geostationary  weather
satellites can monitor weather and cloud patterns covering an entire hemisphere of the
Figure: Geostationary orbit Near-polar orbit36
Many remote sensing platforms are designed to follow an orbit (basically north-south)
which, in conjunction with the Earth's rotation (west-east), allows them to cover most
of  the  Earth's  surface  over  a  certain  period  of  time.  These  are  near-polar  orbits,  so
named for the inclination of the orbit relative to a line running between the North and
South poles. Many of these satellite orbits are also sun-synchronous such that they
cover each area of the world at a constant local time of day called local sun time. At
any given latitude, the position of the sun in the sky as the satellite passes overhead
will  be  the  same  within  the  same  season.  This  ensures  consistent  illumination
conditions when acquiring images in a specific season over successive years, or over a
particular  area  over  a  series  of  days.  This  is  an  important  factor  for  monitoring
changes between images or for mosaicking adjacent images together, as they do not
have to be corrected for different illumination conditions.
Most  of  the  remote  sensing  satellite  platforms  today  are  in  near-polar  orbits,  which
means that the satellite travels northwards on one side of the Earth and then toward
the  southern  pole  on  the  second  half  of  its  orbit.  These  are  called  ascending  and
descending  passes,  respectively.  If  the  orbit  is  also  sun-synchronous,  the  ascending
pass is most likely on the shadowed side of the Earth while the descending pass is on
the sunlit side. Sensors recording reflected solar energy only image the surface on a
descending pass, when solar illumination is available. Active sensors which provide
their own illumination or passive sensors that record emitted (e.g. thermal) radiation
can also image the surface on ascending passes.
As  a  satellite  revolves  around  the  Earth,  the  sensor  "sees"  a  certain  portion  of  the
Earth's surface. The area imaged on the surface, is referred to as the swath. Imaging
swaths   for   spaceborne   sensors   generally   vary   between   tens   and   hundreds   of
kilometres  wide.  As  the  satellite  orbits  the  Earth  from  pole  to  pole,  its  east-west
position wouldn't change if the Earth didn't rotate. However, as seen from the Earth, it
seems that the satellite is shifting westward because the Earth is rotating (from west to
east)  beneath  it.  This  apparent  movement  allows  the  satellite  swath  to  cover  a  new
area  with  each  consecutive  pass.  The  satellite's  orbit  and  the  rotation  of  the  Earth
work  together  to  allow  complete  coverage  of  the  Earth's  surface,  after  it  has
completed one complete cycle of orbits.
Figure: Ascending and descending passes Swath
If we start with any randomly selected pass in a satellite's orbit, an orbit cycle will be
completed  when  the  satellite  retraces  its  path,  passing  over  the  same  point  on  the
Earth's surface directly below the satellite (called the nadir point) for a second time.
The exact length of time of the orbital cycle will vary with each satellite. The interval37
of  time  required  for  the  satellite  to  complete  its  orbit  cycle  is  not  the  same  as  the
"revisit  period".  Using  steerable  sensors,  an  satellite-borne  instrument  can  view  an
area (off-nadir) before and after the orbit passes over a target, thus making the 'revisit'
time less than the orbit cycle time. The revisit period is an important consideration for
a  number  of  monitoring  applications,  especially  when  frequent  imaging  is  required
(for example, to monitor the spread of an oil spill, or the extent of flooding). In near-
polar orbits, areas at high latitudes will be imaged more frequently than the equatorial
zone due to the increasing overlap in adjacent swaths as the orbit paths come closer
together near the poles.
3.3 Scanner Sensor Systems
Electro-optical and spectral imaging scanners produce digital images with the use of
detectors  that  measure  the  brightness  of  reflected  electromagnetic  energy.  Scanners
consist of one or more sensor detectors depending on type of sensor system used.
One type of scanner is called a whiskbroom scanner also referred to as across-track
scanners  (e.g.  on  LANDSAT).  It  uses  rotating  mirrors  to  scan  the  landscape  below
from  side  to  side  perpendicular  to  the  direction  of  the  sensor  platform,  like  a
whiskbroom. The width of the sweep is referred to as the sensor swath. The rotating
mirrors  redirect  the  reflected  light  to  a  point  where  a  single  or  just  a  few  sensor
detectors are grouped together. Whiskbroom scanners with their moving mirrors tend
to be large and complex to build. The moving mirrors create spatial distortions that
must  be  corrected  with  preprocessing  by  the  data  provider  before  image  data  is
delivered to the user. An advantage of whiskbroom scanners is that they have fewer
sensor detectors to keep calibrated as compared to other types of sensors.
Another  type  of  scanner,  which  does  not  use  rotating  mirrors,  is  the  pushbroom
scanner  also  referred  to  as  an  along-track  scanner  (e.g.  on  SPOT).  The  sensor
detectors in a pushbroom scanner are lined up in a row called a linear array. Instead of
sweeping from side to side as the sensor system moves forward, the one dimensional
sensor  array  captures  the  entire  scan  line  at  once  like  a  pushbroom  would.  Some
recent  scanners  referred  to  as  step  stare  scanners  contain  two-dimensional  arrays  in
rows  and  columns  for  each  band.  Pushbroom  scanners  are  lighter,  smaller  and  less
complex  because  of  fewer  moving  parts  than  whiskbroom  scanners.  Also  they  have
better  radiometric  and  spatial  resolution.  A  major  disadvantage  of  pushbroom
scanners is the calibration required for a large number of detectors that make up the
sensor system.38
3.4 Spatial Resolution, Pixel Size, and Scale
For some remote sensing instruments, the distance between the target being imaged
and the platform, plays a large role in determining the detail of information obtained
and  the  total  area  imaged  by  the  sensor.  Sensors  onboard  platforms  far  away  from
their  targets,  typically  view  a  larger  area,  but  cannot  provide  great  detail.  Compare
what  an  astronaut  onboard  the  space  shuttle  sees  of  the  Earth  to  what  you  can  see
from  an  airplane.  The  astronaut  might  see  your  whole  province  or  country  in  one
glance,  but  couldn't  distinguish  individual  houses.  Flying  over  a  city  or  town,  you
would be able to see individual buildings and cars, but you would be viewing a much
smaller area than the astronaut. There is a similar difference between satellite images
and airphotos.
The detail discernible in an image is dependent on the spatial resolution of the sensor
and  refers  to  the  size  of  the  smallest  possible  feature  that  can  be  detected.  Spatial
resolution  of  passive  sensors  (we  will  look  at  the  special  case  of  active  microwave
sensors  later)  depends  primarily  on  their  Instantaneous  Field  of  View  (IFOV).  The
IFOV is the angular cone of visibility of the sensor (A) and determines the area on the
Earth's surface which is "seen" from a given altitude at one particular moment in time
(B).  The  size  of  the  area  viewed  is  determined  by  multiplying  the  IFOV  by  the
distance  from  the  ground  to  the  sensor  (C).  This  area  on  the  ground  is  called  the
resolution  cell  and  determines  a  sensor's  maximum  spatial  resolution.  For  a
homogeneous feature to be detected, its size generally has to be equal to or larger than39
the resolution cell. If the feature is smaller than this, it may not be detectable as the
average  brightness  of  all  features  in  that  resolution  cell  will  be  recorded.  However,
smaller features may sometimes be detectable if their reflectance dominates within a
articular resolution cell allowing sub-pixel or resolution cell detection.
Figure: Instanteneous Field Of View
Most remote sensing images are composed of a matrix of picture elements, or pixels,
which  are  the  smallest  units  of  an  image.  Image  pixels  are  normally  square  and
represent a certain area on an image. It is important to distinguish between pixel size
and  spatial  resolution  -  they  are  not  interchangeable.  If  a  sensor  has  a  spatial
resolution of 20 metres and an image from that sensor is displayed at full resolution,
each pixel represents an area of 20m x 20m on the ground. In this case the pixel size
and resolution are the same. However, it is possible to display an image with a pixel
size different than the resolution. Many posters of satellite images of the Earth have
their pixels averaged to represent larger areas, although the original spatial resolution
of the sensor that collected the imagery remains the same.
A photograph can be represented and displayed in a digital format by subdividing the
image into small equal-sized and shaped areas, called picture elements or pixels, and
representing the brightness of each area with a numeric value or digital number.40
Images where only large features are visible are said to have coarse or low resolution.
In fine or high resolution images, small objects can be detected. Military sensors for
example, are designed to view as much detail as possible, and therefore have very fine
resolution. Commercial satellites provide imagery with resolutions varying from a few
metres to several kilometres. Generally speaking, the finer the resolution, the less total
ground area can be seen.
The ratio of distance on an image or map, to actual ground distance is referred to as
scale. If you had a map with a scale of 1:100,000, an object of 1cm length on the map
would  actually  be  an  object  100,000cm  (1km)  long  on  the  ground.  Maps  or  images
with small "map-to-ground ratios" are referred to as small scale (e.g. 1:100,000), and
those with larger ratios (e.g. 1:5,000) are called large scale.
3.5 Spectral / Radiometric resolution
§     Mosher David C. and Simpkin Peter G. (), Status and Trends of Marin e High-Resolution Seismic
Reflection Profiling: Data Acquisition
3.5.1 Spectral charaecteristics
While  the  arrangement  of  pixels  describes  the  spatial  structure  of  an  image,  the
radiometric characteristics describe the actual information content in an image. Every
time an image is acquired on film or by a sensor, its sensitivity to the magnitude of the
electromagnetic  energy  determines  the  radiometric  resolution.  The  radiometric
resolution  of  an  imaging  system  describes  its  ability  to  discriminate  very  slight
differences  in  energy  The  finer  the  radiometric  resolution  of  a  sensor,  the  more
sensitive it is to detecting small differences in reflected or emitted energy.41
Digital resolution is the number of bits comprising each digital sample. Imagery data
are  represented  by  positive  digital  numbers  which  vary  from  0  to  (one  less  than)  a
selected  power  of  2.  This  range  corresponds  to  the  number  of  bits  used  for  coding
numbers in binary format. Each bit records an exponent of power 2 (e.g. 1 bit = 2 1  =
2).  The  maximum  number  of  brightness  levels  available  depends  on  the  number  of
bits used in representing the energy recorded. Thus, if a sensor used 8 bits to record
the  data,  there  would  be  2 8   =  256  digital  values  available,  ranging  from  0  to  255  –
termed also as the dynamic range of the system. However, if only 4 bits were used,
then  only  2 4   =  16  values  ranging  from  0  to  15  would  be  available.  Thus,  the
radiometric  resolution  would  be  much  less.  Image  data  are  generally  displayed  in  a
range  of  grey  tones,  with  black  representing  a  digital  number  of  0  and  white
representing the maximum value (for example, 255 in 8-bit data). By comparing a 2-
bit image with an 8-bit image, we can see that there is a large difference in the level of
detail discernible depending on their radiometric resolutions.
The  range  of  energy  values  expected  from  a  system  must  “fit”  within  the  range  of
values  possible  of  the  data  format  type,  and  yet  the  value  must  represent  accurately
the energy value of the signal relative to others. The cost of more bits per data point is
longer  acquisition  times,  the  requirement  of  larger  storage  capacity  and  longer
processing time. Any signal outside the range is “clipped” and thus unrecoverable. On
the other hand, if the dynamic range of the signal is widened too much as to allow the
recording of extremely high or low energy values, the true variability within the signal
will be lost.
Many remote sensing systems record energy over several separate wavelength ranges
at  various  spectral  resolutions.  These  are  referred  to  as  multi-spectral  sensors  and
will  be  described  in  some  detail  in  following  sections.  Advanced  multi-spectral
sensors called hyperspectral sensors, detect hundreds of very narrow spectral bands
throughout the visible, near-infrared, and mid-infrared portions of the electromagnetic
spectrum.  Their  very  high  spectral  resolution  facilitates  fine  discrimination  between
different targets based on their spectral response in each of the narrow bands.
There  are  4  general  parameters  that  describe  the  capability  of  a  spectrometer:  1)
spectral  range,  2)  spectral  bandwidth,  3)  spectral  sampling,  and  4)  signal-to-noise
ratio (S/N). Spectral range
Spectral range is important to cover enough diagnostic spectral absorptions to solve a
desired  problem.  There  are  general  spectral  ranges  that  are  in  common  use,  each  to
first order controlled by detector technology: a) ultraviolet (UV): 0.001 to 0.4 µm, b)
visible:  0.4  to  0.7  µm,  c)  near-infrared  (NIR):  0.7  to  3.0  µm,  d)  the  mid-infrared
(MIR):  3.0  to  30  µm,  and  d)  the  far  infrared  (FIR):  30  µm  to  1  mm  (e.g.  see  The
Photonics Design and Applications Handbook, 1996 and The Handbook of Chemistry
and  Physics,  any  recent  year).  The  ~0.4  to  1.0-µm  wavelength  range  is  sometimes
referred to in the remote sensing literature as the VNIR (visible-near-infrared) and the
1.0  to  2.5-µm  range  is  sometimes  referred  to  as  the  SWIR  (short-wave  infrared).  It
should  be  noted  that  these  terms  are  not  recognized  standard  terms  in  other  fields
except remote sensing, and because the NIR in VNIR conflicts with the accepted NIR
range,  the  VNIR  and  SWIR  terms  probably  should  be  avoided.  The  mid-infrared42
covers  thermally  emitted  energy,  which  for  the  Earth  starts  at  about  2.5  to  3  µm,
peaking  near  10  µm,  decreasing  beyond  the  peak,  with  a  shape  controlled  by  grey-
body emission. Spectral bandwidth
Spectral bandwidth is the width of an individual spectral channel in the spectrometer.
The  narrower  the  spectral  bandwidth,  the  narrower  the  absorption  feature  the
spectrometer  will  accurately  measure,  if  enough  adjacent  spectral  samples  are
obtained.  Some  systems  have  a  few  broad  channels,  not  contiguously  spaced  and,
thus,  are  not  considered  spectrometers  (Figure  1a).  Examples  include  the  Landsat
Thematic    Mapper    (TM)    system    and    the    MODerate    Resolution    Imaging
Spectroradiometer (MODIS), which can't resolve narrow absorption features. Others,
like the NASA JPL Airborne Visual and Infra-Red Imaging Spectrometer (AVIRIS)
system  have  many  narrow  bandwidths,  contiguously  spaced  (Figure  1b).  Figure  1
shows  spectra  for  the  mineral  alunite  that  could  be  obtained  by  some  example
broadband  and  spectrometer  systems.  Note  the  loss  in  subtle  spectral  detail  in  the
lower  resolution  systems  compared  to  the  laboratory  spectrum.  Bandwidths  and
sampling  greater  than  25  nm  rapidly  lose  the  ability  to  resolve  important  mineral
absorption features. All the spectra in Figure 1b are sampled at half Nyquist (critical
sampling)  except  the  Near  Infrared  Mapping  Spectrometer  (NIMS),  which  is  at
Nyquist sampling (named after H. Nyquist, that in his work published in 1928, stated
that there must be at least to samplings per wavelength of the highest frequency, in
order to appropriately sample the waveform). Note, however, that the fine details of
the absorption features are lost at the ~25 nm  bandpass  of  NIMS.  For  example,  the
shoulder  in  the  2.2-µm  absorption  band  is  lost  at  25-nm  bandpass.  The  Visual  and
Infrared Mapping Spectrometer (VIMS) and NIMS systems measure out to 5 µm, thus
can see absorption bands not obtainable by the other systems.4344
The shape of the bandpass profile is also important. Ideally each spectrometer channel
rejects  all  light  except  that  from  within  a  given  narrow  wavelength  range,  but
occasionally,  due  to  optical  effects  too  complex  to  discuss  in  detail  here,  light  may
leak  in  from  out  of  the  bandpass  (e.g.  scattering  within  the  optical  system,  or
inadequate  blocking  filters).  The  most  common  bandpass  in  spectrometers  is  a
Gaussian   profile.   While   specific   spectrometer   designs   may   have   well-defined
theoretical  bandpass  profiles,  aberrations  in  the  optical  system  usually  smears  the
profile closer to a Gaussian shape. The width of the bandpass is usually defined as the
width in wavelength at the 50% response level of the function, as shown in the next
Figure, called the Full Width at Half Maximum (FWHM).
Figure: A Gaussian profile with a Full Width at Half Maximum (FWHM) of 10 nm is shown. This
profile is typical of spectrometers such as AVIRIS which has 224 such profiles spaced at about 10 nm. Spectral sampling
Spectral sampling is the distance in wavelength between the spectral bandpass profiles
for each channel in the spectrometer as a function of wavelength. Spectral sampling is
often  confused  with  bandpass,  with  the  two  lumped  together  and  called  resolution.
Information theory tells us that to resolve a two spectral features, we must have two
samples. Further, in order to not introduce sampling bias, the samples must be close
enough together to measure the peak and valley locations. The Nyquist theorem states
that  the  maximum  information  is  obtained  by  sampling  at  one-half  the  FWHM.
Spectrometer  design,  however,  sometimes  dictates  a  different  sampling,  and  many
modern spectrometers in use (e.g. AVIRIS, VIMS) sample at half-Nyquist: sampling
interval  approximately  equal  to  the  FWHM.  Note  that  the  AVIRIS  system  has  a
bandpass  ~0.01  µm  (10  nm),  a  sampling  of  ~0.01  µm,  and  thus  has  a  spectral
resolution of ~0.02 µm (20 nm). The NIMS system in the Figure above can sample at
Nyquist (shown), half-Nyquist, and lower.45 Signal-to-noise ratio
Finally,  a  spectrometer  must  measure  the  spectrum  with  enough  precision  to  record
details in the spectrum. The signal-to-noise ratio (S/N) required to solve a particular
problem will depend on the strength of the spectral features under study. The S/N is
dependant on the detector sensitivity, the spectral bandwidth, and intensity of the light
reflected  or  emitted  from  the  surface  being  measured.  A  few  spectral  features  are
quite strong and a signal to noise of only about 10 will be adequate to identify them,
while  others  are  weak,  and  a  S/N  of  several  hundred  (and  higher)  are  often  needed
(Swayze et al., 1997).
3.6 Temporal Resolution
In  addition  to  spatial,  spectral,  and  radiometric  resolution,  the  concept  of  temporal
resolution is also important to consider in a remote sensing system. The revisit period
of a satellite sensor is usually several days. Therefore the absolute temporal resolution
of a remote sensing system to image the exact same area at the same viewing angle a
second  time  is  equal  to  this  period.  However,  the  actual  temporal  resolution  of  a
sensor depends on a variety of factors, including the satellite/sensor capabilities, the
swath overlap, and latitude.
The  ability  to  collect  imagery  of  the  same  area  of  the  Earth's  surface  at  different
periods  of  time  is  one  of  the  most  important  elements  for  applying  remote  sensing
data. Spectral characteristics of features may change over time and these changes can
be  detected  by  collecting  and  comparing  multi-temporal  imagery.  For  example,
during  the  growing  season,  most  species  of  vegetation  are  in  a  continual  state  of
change  and  our  ability  to  monitor  those  subtle  changes  using  remote  sensing  is
dependent  on  when  and  how  frequently  we  collect  imagery.  By  imaging  on  a
continuing basis at different times we are able to monitor the changes that take place
on  the  Earth's  surface,  whether  they  are  naturally  occurring  (such  as  changes  in
natural  vegetation  cover  or  flooding)  or  induced  by  humans  (such  as  urban
development or deforestation). The time factor in imaging is important when:
•   persistent clouds offer limited clear views of the Earth's surface (often in the
•   short-lived phenomena (floods, oil slicks, etc.) need to be imaged
•   multi-temporal  comparisons  are  required  (e.g.  the  spread  of  a  forest  disease
from one year to the next)
•   the  changing  appearance  of  a  feature  over  time  can  be  used  to  distinguish  it
from near-similar features (wheat / maize)46
3.7 Overview of different sensors – satellites and airborne
3.7.1 Comparison Table
Bands and wavelength
Swath width        Repeat
NOAA       1 (0.58-0.68) 1.1 km 2399 daily 833
2 (0.725-1.10) 1.1 km 2399 daily 833
3 (3.55-3.93) 1.1 km 2399 daily 833
4 (10.3-11.3) 1.1 km 2399 daily 833
5 (11.5-12.5) 1.1 km 2399 daily 833
MSS 4-5     1 (0.5-0.6) 79/82m 185 16 days 705
2 (0.6-0.7) 79/82m 185 16 days 705
3 (0.7-0.8) 79/82m 185 16 days 705
4 (0.8-1.1) 79/82m 185 16 days 705
TM 4-5       1 (0.45-0.52) 30m 185 16 days 705
2 (0.52-0.60) 30m 185 16 days 705
3 (0.63-0.69) 30m 185 16 days 705
4 (0.76-0.90) 30m 185 16 days 705
5 (1.55-1.75) 30m 185 16 days 705
6 (10.40-12.50) 120m 185 16 days 705
7 (2.08-2.35) 30m 185 16 days 705
ETM 7        1 (0.45-0.515) 30m 183*170 16 days 705
2 (0.525-0.605) 30m 183*170 16 days 705
3 (0.63-0.69) 30m 183*170 16 days 705
4 (0.75-0.90) 30m 183*170 16 days 705
5 (1.55-1.75) 30m 183*170 16 days 705
6 (10.40-12.5) 60m 183*170 16 days 705
7 (2.09-2.35) 30m 183*170 16 days 705
PAN (0.52-0.90) 15m 183*170 16 days 705
SPOT 4      XS 1 (0.50-0.59) 20m 60 (oblique
scene at max 60
by 81)
26 days 822
XS 2 (0.61-0.68) 20m 60 (oblique
scene at max 60
by 81)
26 days 822
XS 3 (0.79-0.89) 20m 60 (oblique
scene at max 60
by 81)
26 days 822
XS 4 (1.58-1.75) 20m 60 (oblique
scene at max 60
by 81)
26 days 822
Monospectral red (0.61-0.68)       10m 60 (oblique
scene at max 60
by 81)
26 days 822
IRS 1C       LISS 1 (0.52-0.59) 23.6m 142 24 days 818
LISS 2 (0.62-0.68) 23.6m 142 24 days 818
LISS 3 (0.77-0.86) 23.6m 142 24 days 818
LISS 4 (1.55-1.70) 70.8m 148 24 days 818
WIFS 1 (0.62-0.68) 189m 810 24 days 818
WIFS 2 (0.77-0.86) 189m 810 24 days 818
PAN (0.5-0.75) 5.8m 70 24 days 81847
Ikonos        Multispectral (0.45-0.52) 4m 13 at nadir 2.9 days  1m
Multispectral (0.52-0.60) 4m 13 at nadir 2.9 days  1m
Multispectral (0.63-0.69) 4m 13 at nadir 2.9 days  1m
Multispectral (0.76-0.90) 4m 13 at nadir 2.9 days  1m
Panchromatic (0.45-0.90) 1m 13 at nadir 2.9 days  1m
681 Weather Satellites/Sensors
Weather  monitoring  and  forecasting  was  one  of  the  first  civilian  (as  opposed  to
military) applications of satellite remote sensing, dating back to the first true weather
satellite,  TIROS-1  (Television  and  Infrared  Observation  Satellite  -  1),  launched  in
1960  by  the  United  States.  Several  other  weather  satellites  were  launched  over  the
next five years, in near-polar orbits, providing repetitive coverage of global weather
patterns. In 1966, NASA (the U.S. National Aeronautics and Space Administration)
launched   the   geostationary   Applications   Technology   Satellite   (ATS-1)   which
provided hemispheric images of the Earth's surface and cloud cover every half hour.
For  the  first  time,  the  development  and  movement  of  weather  systems  could  be
routinely  monitored.  Today,  several  countries  operate  weather,  or  meteorological
satellites to monitor weather conditions around the globe. Generally speaking, these
satellites  use  sensors  which  have  fairly  coarse  spatial  resolution  (when  compared  to
systems  for  observing  land)  and  provide  large  areal  coverage.  Their  temporal
resolutions  are  generally  quite  high,  providing  frequent  observations  of  the  Earth's
surface,  atmospheric  moisture,  and  cloud  cover,  which  allows  for  near-continuous
monitoring of global weather conditions, and hence - forecasting. Here we review a
few of the representative satellites/sensors used for meteorological applications. GOES
The GOES (Geostationary Operational Environmental Satellite) System is the follow-
up  to  the  ATS  series.  They  were  designed  by  NASA  for  the  National  Oceanic  and
Atmospheric Administration (NOAA) to provide the United States National Weather
Service  with  frequent,  small-scale  imaging  of  the  Earth’s  surface  and  cloud  cover.
The  GOES  series  of  satellites  have  been  used  extensively  by  meteorologists  for
weather  monitoring  and  forecasting  for  over  20  years.  These  satellites  are  part  of  a
global  network  of  meteorological  satellites  spaced  at  approximately  70°  longitude
intervals  around  the  Earth  in  order  to  provide  near-global  coverage.  Two  GOES
satellites,  placed  in  geostationary  orbits  36000  km  above  the  equator,  each  view
approximately one-third of the Earth. One is situated at 75°W longitude and monitors
North  and  South  America  and  most  of  the  Atlantic  Ocean.  The  other  is  situated  at
135°W longitude and monitors North America and the Pacific Ocean basin. Together
they cover from 20°W to 165°E longitude.
Two generations of GOES satellites have been launched, each measuring emitted and
reflected radiation from which atmospheric temperature, winds, moisture, and cloud
cover can be derived. The first generation of satellites consisted of GOES-1 (launched
1975)  through  GOES-7  (launched  1992).  Due  to  their  design,  these  satellites  were
capable of viewing the Earth only a small percentage of the time (approximately five
per  cent).  The  second  generation  of  satellites  began  with  GOES-8  (launched  1994)
and  has  numerous  technological  improvements  over  the  first  series.  They  provide48
near-continuous observation of the Earth allowing more frequent imaging (as often as
every 15 minutes). This increase in temporal resolution coupled with improvements in
the spatial and radiometric resolution of the sensors provides timelier information and
improved data quality for forecasting meteorological conditions.
GOES-8 and the other second generation GOES satellites have separate imaging and
sounding instruments. The imager has five channels sensing visible (1 km resolution)
and  infrared  reflected  and  emitted  solar  radiation  (4  km  resolution).  The  infrared
capability  allows  for  day  and  night  imaging.  Sensor  pointing  and  scan  selection
capability enable imaging of an entire hemisphere, or small-scale imaging of selected
areas.  The  latter  allows  meteorologists  to  monitor  specific  weather  trouble  spots  to
assist  in  improved  short-term  forecasting.  The  imager  data  are  10-bit  radiometric
resolution,  and  can  be  transmitted  directly  to  local  user  terminals  on  the  Earth’s
The 19 channel sounder measures emitted radiation in 18 thermal infrared bands and
reflected radiation in one visible band. These data have a spatial resolution of 8 km
and  13-bit  radiometric  resolution.  Sounder  data  are  used  for  surface  and  cloud-top
temperatures, multi-level moisture profiling in the atmosphere, and ozone distribution
analysis. National Oceanic and Atmospheric Administration's Advanced Very
High Resolution Radiometer
The  Advanced  Very  High  Resolution  Radiometer  (AVHRR)  sensor  is  carried  on-
board  the  National  Oceanic  and  Atmospheric  Administration's  (NOAA)  series  of
Polar-Orbiting  Operational  Environmental  Satellites  (POES).  A  prototype  AVHRR
sensor  as  developed  and  first  launched  in  October  1978  to  acquire  meteorological
data,  including  day  and  night  cloud  mapping,  surface  water  delineation,  and  sea
surface temperatures. The next AVHRR sensor, launched on the NOAA-6 satellite in
June  1979,  included  a  redefined  visible  spectral  band.  The  Figure  of  Observation
Characteristics  illustrates  that  the  redefined  band  eliminates  an  overlap  with  a  near-
infrared  band  and  corresponds  more  closely  with  the  red  absorption  region  of
vegetation.  This  development  heralded  the  additional  use  of  the  AVHRR  as  a
vegetation mapping and analysis tool by enabling the acquisition of data suitable for
use in a computed Normalized Difference Vegetation Index (NDVI).
AVHRR  data  are  acquired  with  a  wide-field  scanning  system  that  enables  global
coverage on a daily basis with a ground resolution of 1.1 km at nadir (directly beneath
the satellite). The sensor also provides a data stream of nominal 4-km resolution that
is  achieved  by  sampling  and  averaging  the  full  resolution  1.1-km  data  on-board  the
satellite. Land Observation Satellites/Sensors Landsat Multispectral Scanner
The Multispectral Scanner (MSS), launched on-board Landsat 1 in July 1972, was the
world's first Earth observation satellite sensor. The MSS provided four spectral bands
of   Earth   surface   reflectance   in   the   visible   and   near-infrared   regions   of   the
electromagnetic  spectrum  at  a  nominal  80-m  spatial  resolution,  as  illustrated  by  the
Figure  of  Observation  Characteristics.  The  U.S.  Geological  Survey's  Global  Land
Information System (GLIS) provides information on the sensor, and acquisition and
availability of Multispectral Scanner Landsat Data.49
Landsat MSS data, collected by Landsats 1 through 5 from 1972 to 1993, provide the
longest and most extensive archive of satellite image data for monitoring the global
land  surface.  Efforts  are  ongoing  to  reconfigure  selected  portions  of  the  historical
global  Landsat  data  archive  into  the  Landsat  Pathfinder  Data  Set  to  make  it  more
readily useful for change detection.
In  February  1993,  the  Earth  Observation  Satellite  Company  (EOSAT),  commercial
operator  of  Landsats  4  and  5,  suspended  acquisition  of  Landsat  MSS  data.  Landsat
MSS data are now superseded by data from the Landsat Thematic Mapper (TM). Landsat Thematic Mapper
Landsats 4 and 5, launched in 1982 and 1984, respectively, were augmented with an
advanced  version  of  an  Earth  observation  sensor  known  as  the  Thematic  Mapper
(TM). The TM provides a significant increase in data acquisition capability over the
MSS in a number of ways, as shown in the Figure of Observation Characteristics. The
TM  sensor  has  seven  spectral  bands:  Six  acquire  Earth  reflectance  data,  and  one
acquires  Earth  temperature  data.  The  spatial  resolution  of  bands  in  the  visible  and
reflective  infrared  regions  is  30  m,  some  2  1/2  times  better  than  the  Multispectral
Scanner (MSS). The TM sensor also has greater overall radiometric sensitivity than
the MSS.
To date, the Landsat TM sensor represents the most sophisticated satellite sensor to
provide  Earth  observation  data.  The  sensor's  complement  of  seven  spectral  bands
offers the most comprehensive set of multispectral measurements for land and water
surface  mapping,  monitoring,  and  analysis.  Data  must  be  ordered  from  the  Earth
Observation Satellite Company (EOSAT), who holds the commercial rights to all TM
data less than 10 years old.
Landsat 7 was successfully launched from Vandenberg Air Force Base on April 15,
1999 at 11:32 am PDT. The earth observing instrument on Landsat 7, the Enhanced
Thematic  Mapper  Plus  (ETM+),  replicates  the  capabilities  of  the  highly  successful
Thematic  Mapper  instruments  on  Landsats  4  and  5*.  The  ETM+  also  includes  new
features  that  make  it  a  more  versatile  and  efficient  instrument  for  global  change
studies, land cover monitoring and assessment, and large area mapping than its design
The primary new features on Landsat 7 are:
•   a panchromatic band with 15m spatial resolution
•   on board, full aperture, 5% absolute radiometric calibration
•   a thermal IR channel with 60m spatial resolution Systeme Probatoire d'Observation de la Terra (SPOT) High Resolution
Visible Sensor
The  French  Systeme  Probatoire  d'Observation  de  la  Terra  (SPOT)  Earth  observing
satellite  system  carries  two  High  Resolution  Visible  (HRV)  imaging  sensors.  First
launched in 1986, the HRV provides high spatial resolution, with three spectral bands
of visible and near-infrared data acquired at 20-m resolution, and a panchromatic band
at  10-m  resolution,  as  shown  in  the  the  table  above.  The  sensors  can  be  pointed  to
either side of the orbital track (plus or minus 27 degrees), allowing the acquisition of
stereo and repeat coverage in as short a period one or four days.50
The  high  spatial  resolution  of  the  SPOT  sensors  has  proven  to  be  very  useful  for
applications  requiring  highly  detailed  information.  For  situations  where  additional
spectral  information  is  desired  at  SPOT-equivalent  resolution,  the  10-m  spatial
resolution  of  the  SPOT  panchromatic  data  can  be  fused  with  Landsat  Thematic
Mapper (TM) data to bring the advantages of both types of data into the same image. Indian Remote Sensing
Indian program to develop an indigenous capability to image Earth, particulary India.
Mission: ground water exploration, land use, forest and flood mapping, inventory of
surface water.
Having  been  the  seventh  nation  to  achieve  orbital  capability  in  July  1980,  India  is
pressing ahead with an impressive national programme aimed at developing launchers
as  well  as  nationally  produced  communications,  meteorological  and  Earth  resources
satellites.  Prof  U.R.  Rao,  who  became  the  Chairman  of  the  ISRO  (Indian  Space
Research Organization) in October 1984, said that space technology had given India
the  opportunity  to  convert  backwardness  into  an  asset;  developing  countries  could
bypass the intermediate technology stage and leapfrog into the high technology area.
Like France, India has benefited from simultaneous co-operation with the CIS/USSR,
the US and ESA.
Currently   ISRO   has   more   than   a   half-dozen   remote   sensing   satellites   under
development, intended for launch in the next decade. The entire fleet will be a mix of
niche-role and multi-role satellites which will join India's existing constellation of five
polar   orbiters:   IRS-1A,   IRS-1B,   IRS-1C,   IRS-1D,   IRS-P2,   IRS-P3.   While
development of the US Landsat remote sensing satellites series has stumbled along,
Space Imaging EOSAT plans to market imagery and data from the 14 satellites that
India  expects  to  fly  by  the  year  2005.  Over  1988  to  2000  there  will  have  been  12
satellites  launched  towards  polar  orbits.  They  are  not  only  placing  India  in  a  strong
position to satisfy its own needs, but also dominate a significant portion of the global
commercial  market.  The  IRS  series  in  now  moving  from  land  applications  to
environmental monitoring, with a first emphasis on oceanography with the IRS-P4.
The  IRS-1C  and  1D  offer  improved  spatial  and  spectral  resolution,  on-board
recording, stereo viewing capability and more frequent revisits over the orbital repeat
cycle of 24 days. The IRS-1C is paving the way for future satellites in the 1990s. It
carries three separate imaging sensor that fulfill the traditional data needs of existing
customers,  while  also  attracting  new  users  with  high-resolution  data  acquisition
capabilities. Its Wide Field Sensor (WiFS) provides regional imagery acquiring data
with  800-kilometer  swaths  at  a  coarse  188-meter  resolution  in  two  spectral  bands,
visible  (620-680nm)  and  near-infrared  (770-860nm),  and  will  mainly  be  used  for
vegetation index mapping. The WiFS offers a rapid revisit time of 3 days. The Linear
Imaging Self-Scanning Sensor 3 (LISS-3) serves the needs of multi-spectral imagery
clients, possibly the largest of all current data user groups. LISS-3 acquires four bands
(520-590, 620-680, 770-860, and 1550-1750 nm) with a 23.7 meter spatial resolution,
which  makes  it  an  ideal  complement  to  data  form  the  aging  Landsat  5  Thematic
Mapper  sensor.  Of  course,  the  most  talked  about  of  the  three  sensors  is  the
panchromatic camera with a resolution of 5.8 meter, giving the IRS-1C (and 1D) the
highest resolution of any civilian remote sensing satellite currently in orbit. With its
5.8-meter resolution the IRS-1C is carving a market niche for itself and its twin the51
IRS-1D,  among  GIS  users  whose  applications  require  spatial  detail  and  scene  size
between the 10-meter SPOT satellites and the upcoming 1-meter systems. The PAN
sensor is steerable up to plus or minus 26 degrees and thus offers stereo capabilities
and  a  possible  frequent  revisit  of  about  5  days,  depending  on  the  latitude.  Working
together, the IRS-1C and 1D will also cater to users who need a rapid revisiting rate.
The combination of the Wide Field Sensor with the high-resolution LISS has proven
to  be  very  useful  by  offering  an  overview  of  regional  phenomena,  a  capability
traditionally  offered  by  sensors  like  the  AVHRR,  and  at  the  same  time  offering  the
possibility to zoom in to specific areas to pinpoint very local phenomena. The spectral
coverage of the two WiFS bands are equal to the bands 2 and 3 of the LISS, which
makes  comparison  on  different  scales  particularly  easy.  And,  let's  face  it,  the  188
meter resolution is far sufficient for some applications. Like, for instance, in forestry
over  large  areas,  where  the  old  cliche  about  seeing  the  forest  for  the  trees  really
applies. The 188 meter resolution is sufficient for foresters to do an inventory of large
land holdings, taking note of general species variations, and diseases and mortality in
the  tree  stands.  Some  remote  sensing  experts  even  credit  the  IRS-1C  WiFS  with
generating a renewed enthusiasm over low-resolution imagery.
For collecting data outside the visibility region of the Indian earth station, the IRS-1C
and 1D satellites have an on-board tape recorder (OTBR) which is able to record and
store data collected for a duration of 24 minutes. This data will be downlinked once
the satellite comes into the visibility region of the Indian data receiving station. Data
recording on OBTR is based on user request; the programming of the OBTR occurs at
NDC. The recorded data is played back over Shadnagar during night passes.
The IRS-1D was launched on 28 September 1997, and has similar capabilities to the
IRS-1C satellite. Unfortunately, initially, it entered a wrong highly elliptical orbit due
to  a  problem  with  the  Indian  PSLV  rocket.  Fuel  for  precise  orbit  control  had  to  be
used for raising the orbit's perigee. Satellite life because of fuel consumption will be
in  the  3-5  year  range.  It  could  have  been  more  if  not  as  much  fuel  had  been  use  to
adjust the orbit. Apart from this, the spacecraft seems to be in good health, and data
quality  of  the  images  seem  to  be  very  good.  At  SI-EOSAT  the  IRS  images  are  the
preferred data sets, especially when delivered in USGS Digital Ortho Quad format. IKONOS
IKONOS,  the  world’s  first  one-meter  resolution,  commercial  imaging  satellite,  was
launched  in  September  1999.  The  company,  Space  Imaging,  was  founded  in  late
1994,  at  Denver,  Colorado,  U.S.A.  The  satellite  has  four  mutlispectral  bands  in  the
visible  and  Near  Infrared  at  a  spatial  resolution  of  4  meters,  and  a  Panchromatic  1
meter band covering the same spectral range.
The two following images were taken be IKONOS. Notice the high spatial resolution
of the images.52
The red square in Beijing, China, 22.10.1999, four meter resolution colour image
Jefferson Memorial, Washington D.C., 30.9.1999, first public image, one meter resolution B&W image Marine Observation Satellites/Sensors
The  Earth's  oceans  cover  more  than  two-thirds  of  the  Earth's  surface  and  play  an
important role in the global climate system. They also contain an abundance of living
organisms  and  natural  resources  which  are  susceptible  to  pollution  and  other  man-
induced  hazards.  The  meteorological  and  land  observations  satellites/sensors  we
discussed in the previous two sections can be used for monitoring the oceans of the
planet,  but  there  are  other  satellite/sensor  systems  which  have  been  designed
specifically for this purpose.
These  ocean-observing  satellite  systems  are  important  for  global  and  regional  scale
monitoring  of  ocean  pollution  and  health,  and  assist  scientists  in  understanding  the
influence and impact of the oceans on the global climate system. Coastal Zone Colour Scanner (CZCS)
The Nimbus-7 satellite, launched in 1978, carried the first sensor, the Coastal Zone
Colour Scanner (CZCS), specifically intended for monitoring the Earth's oceans and
water bodies. The primary objective of this sensor was to observe ocean colour and
temperature,   particularly   in   coastal   zones,   with   sufficient   spatial   and   spectral
resolution  to  detect  pollutants  in  the  upper  levels  of  the  ocean  and  to  determine  the53
nature of materials suspended in the water column. The Nimbus satellite was placed
in  a  sun-synchronous,  near-polar  orbit  at  an  altitude  of  955  km.  Equator  crossing
times were local noon for ascending passes and local midnight for descending passes.
The repeat cycle of the satellite allowed for global coverage every six days, or every
83 orbits. The CZCS sensor consisted of six spectral bands in the visible, near-IR, and
thermal portions of the spectrum each collecting data at a spatial resolution of 825 m
at nadir over a 1566 km swath width. MOS
The first Marine Observation Satellite (MOS-1) was launched by Japan in February,
1987  and  was  followed  by  its  successor,  MOS-1b,  in  February  of  1990.  These
satellites  carry  three  different  sensors:  a  four-channel  Multispectral  Electronic  Self-
Scanning Radiometer (MESSR – 50m resolution), a four-channel Visible and Thermal
Infrared   Radiometer   (VTIR   –   900   to   2700m   resolution),   and   a   two-channel
Microwave Scanning Radiometer (MSR), in the microwave portion of the spectrum.
The MESSR bands are quite similar in spectral range to the Landsat MSS sensor and
are   thus   useful   for   land   applications   in   addition   to   observations   of   marine
environments.  The  MOS  systems  orbit  at  altitudes  around  900  km  and  have  revisit
periods of 17 days. SeaWiFS
The  SeaWiFS  (Sea-viewing  Wide-Field-of  View  Sensor)  on  board  the  SeaStar
spacecraft is an advanced sensor designed for ocean monitoring. It consists of eight
spectral  bands  of  very  narrow  wavelength  ranges  (see  accompanying  table)  tailored
for  very  specific  detection  and  monitoring  of  various  ocean  phenomena  including:
ocean primary production and phytoplankton processes, ocean influences on climate
processes  (heat  storage  and  aerosol  formation),  and  monitoring  of  the  cycles  of
carbon,  sulfur,  and  nitrogen.  The  orbit  altitude  is  705  km  with  a  local  equatorial
crossing time of 12 PM. Two combinations of spatial resolution and swath width are
available for each band: a higher resolution mode of 1.1 km (at nadir) over a swath of
2800 km, and a lower resolution mode of 4.5 km (at nadir) over a swath of 1500 km. Laser fluorosensor - another kind of sensor
Some targets fluoresce, or emit energy, upon receiving incident energy. This is not a
simple  reflection  of  the  incident  radiation,  but  rather  an  absorption  of  the  initial
energy, excitation of the molecular components of the target materials, and emission
of  longer  wavelength  radiation  which  is  then  measured  by  the  sensor.  Laser
fluorosensors  illuminate  the  target  with  a  specific  wavelength  of  radiation  and  are
capable  of  detecting  multiple  wavelengths  of  fluoresced  radiation.  This  technology
has  been  proven  for  ocean  applications,  such  as  chlorophyll  mapping,  and  pollutant
detection, particularly for naturally occurring and accidental oil slicks. Hyperspectral sensors Compact Airborne Spectrographic Imager CASI
CASI,  the  Compact  Airborne  Spectrographic  Imager,  is  a  leader  in  airborne
imaging, being the first commercial imaging spectrometer. This hyperspectral sensor
detects a vast array of narrow spectral bands in the visible and infrared wavelengths,
using  along-track  scanning.  The  spectral  range  covered  by  the  288  channels  is
between 0.4 and 0.9 mm. Each band covers a wavelength range of 0.018 mm. While54
spatial resolution depends on the altitude of the aircraft, the spectral bands measured
and  the  bandwidths  used  are  all  programmable  to  meet  the  user's  specifications  and
requirements.  Hyperspectral  sensors  such  as  this  can  be  important  sources  of
diagnostic information about specific targets’ absorption and reflection characteristics,
in  effect  providing  a  spectral  'fingerprint'.  Experimentation  with  CASI  and  other
airborne  imaging  spectrometers  has  helped  guide  the  development  of  hyperspectral
sensor systems for advanced satellite systems. Digital Airborne Imaging Spectrometer DAIS 7915
The European Union and DLR are funding a
new  79-channel  Digital  Airborne  Imaging  Spectrometer  (DAIS  7915),  which  was
built      by      the      Geophysical      Environmental      Research      corp.      (GER).
This  new  sensor  covers  the  spectral  range  from  the  visible  to  the  thermal  infrared
wavelengths  at  variable  spatial  resolution  from  3  to  20  m  depending  on  the  carrier
aircraft flight altitude. The DAIS 7915 is used since spring 1995 for remote sensing
applications  such  as  environmental  monitoring  of  land  and  marine  ecosystems,
vegetation status and stress investigations, agriculture and forestry resource mapping,
geological  mapping,  mineral  exploration  as  well  as  for  the  supply  of  data  for
geographic information systems ( GIS ).
Six spectral channels in the 8000 - 12000 nm region could be used for the retrieval of
temperature  and  emissivity  of  land  surface  objects.  These  and  72  narrow  band
channels in the atmospheric windows between 450 and 2450 nm allow to investigate
land  surface  processes  with  a  special  emphasis  on  vegetation  /  soil  interactions.
Based on the requirements for on-ground calibration of the DAIS 7915 the Laboratory
Calibration Facility (LCF) has been developed at DLR's Institute of Optoelectronics.
The     MCF     covers     the     spectral     range     from     400     to     14500     nm.
The DAIS 7915 has been flown several times on a Do 228 at DLR Oberpfaffenhofen
since autumn 1994.
Spectrometer Characteristics
(Wavelength range: 400nm - 12.6µm, 4 Spectrometers, 79 bands)
1)  400 - 1000 nm : 32 Bands, Bandwidth = 15-30 nm  Detector:   Si
2) 1500 - 1800 nm :  8 Bands, Bandwidth = 45    nm  Detector: InSb
3) 2000 - 2500 nm : 32 Bands, Bandwidth = 20    nm  Detector: InSb
   3000 - 5000 nm :  1 Band , Bandwidth = 2.0   µm  Detector: InSb
4) 8000 -12600 nm :  6 Bands, Bandwidth = 0.9   µm  Detector:  MCT AVIRIS Airborne Visible InfraRed Imaging Spectrometer
AVIRIS  is  a  world  class  instrument  in  the  realm  of  Earth  Remote  Sensing.  It  is  a
unique  optical  sensor  that  delivers  calibrated  images  of  the  upwelling  spectral
radiance  in  224  contiguous  spectral  channels  (also  called  bands)  with  wavelengths
from  400  to  2500  nanometers  (nm).  The  instrument  flies  aboard  a  NASA  ER-2
airplane  (a  U2  plane  modified  for  increased  performance)  at  approximately  20  km
above sea level, at about 730 km/hr. AVIRIS has flown all across the US, plus Canada
and Europe.
The  AVIRIS  instrument  contains  224  different  detectors,  each  with  a  wavelength
sensitive range (also known as spectral bandwidth) of approximately 10 nanometers
(nm), allowing it to cover the entire range between 380 nm and 2500 nm. When the
data  from  each  detector  is  plotted  on  a  graph,  it  yields  a  spectrum.  Comparing  the55
resulting  spectrum  with  those  of  known  substances  reveals  information  about  the
composition of the area being viewed by the instrument.
AVIRIS  uses  a  scanning  mirror  to  sweep  back  and  forth  ("whisk  broom"  fashion),
producing  614  pixels  for  the  224  detectors  each  scan.  Each  pixel  produced  by  the
instrument covers an approximately 20 meter square area on the ground (with some
overlap between pixels), thus yielding a ground swath about 11 kilometers wide. Synthetic Aperture Radar Sensors
Synthetic Aperture Radar (SAR) image data provide information different from that
of optical sensors operating in the visible and infrared regions of the electromagnetic
spectrum.  SAR  data  consist  of  high-resolution  reflected  returns  of  radar-frequency
energy from terrain that has been illuminated by a directed beam of pulses generated
by  the  sensor.  The  radar  returns  from  the  terrain  are  mainly  determined  by  the
physical characteristics of the surface features (such as surface roughness, geometric
structure, and orientation), the electrical characteristics (dielectric constant, moisture
content,  and  conductivity),  and  the  radar  frequency  of  the  sensor.  By  supplying  its
own  source  of  illumination,  the  SAR  sensor  can  acquire  data  day  or  night  without
regard  to  cloud  cover.  Elachi  (1988)  provides  a  technical  overview  of  radar  wave-
surface  interactions  and  their  applications  to  land,  water,  and  ice  phenomena  in
Chapter 2 of Spaceborne Radar Remote Sensing. Most other remote sensing textbooks
also  provide  introductory  material  on  SAR  system  properties  and  image  data
Synthetic  aperture  radar  (SAR)  satellite  systems  currently  in  operation  include  the
European  Space  Agency's  (ESA)  European  Remote  Sensing  Satellite  1  (ERS-1),
launched  July  1991,  and  the  Japanese  Earth  Resources  satellite  (JERS-1),  launched
February 1992. Contacts are provided for ERS-1  data  and  JERS-1  data.  The  ERS-1
sensor  operates  in  the  C-band  frequency  (approx.  5.6  cm  wavelength)  and  JERS-1
operates  in  the  L-band  frequency  (approx.  23  cm  wavelength).  Both  sensors  have  a
nominal spatial resolution of approximately 30 m. The Canadian Space Agency plans
to launch its RADARSAT in 1995.
The  SAR  systems  are  now  beginning  to  provide  SAR  image  data  on  a  long-term,
sustained basis. The ERS-1 satellite, with a projected lifespan of three years, will be
followed by an ERS-2 satellite planned to continue SAR data acquisition into the late
1990s, when advanced SAR sensors are expected to become operational as part of the
Earth Observing System (EOS).
The  current  level  of  experience  in  operational  use  of  SAR  data  is  very  limited
compared to the use of visible and infrared data acquired by the multispectral satellite
sensors.  Several  major  characteristics  of  SAR  data  taken  together,  however,  may
promote more extensive evaluation and use of SAR data for land-use and land-cover
information.  These  characteristics  include  1)  the  unique  information  of  surface
roughness, physical structure, and electrical conduction properties; 2) the high spatial
resolution;  3)  the  24-hour,  all-weather  data-acquisition  capability;  and  4)  the  now-
realizable long-term continuity of the data that enables repetitive (seasonal) coverage
of major global land regions.56
4. Corrections:
4.1 Radiometric calibration
Pixel values in commercially available satellite imagery represent the radiance of the
surface in the form of Digital Numbers (DN) which are calibrated to fit a certain range
of values. Sometimes the DN are referred to as the brightness values. Conversion of
DN to absolute radiance values is a necessary procedure for comparative analysis of
several images taken by different sensors (for example, Landsat-2 versus Landsat-5).
Since each sensor has its own calibration parameters used in recording the DN values,
the same DN values in two images taken by two different sensors may represent two
different radiance values.
4.1.1 Main elements of sensor calibration Absolute Radiometric Calibration – from radiance to DN and back
The  following  figure  depicts  a  hypothetical  response  curve  to  a  known  calibration
signal.  There  are  several  common  curve  characteristics  to  note.  The  first  is  that  the
response  curve  does  not  pass  through  the  origin,  i.e.  the  sensor  does  not  give  zero
output even in the dark. This so-called dark signal is caused primarily by electronic
noise in the sensor. The optimum operating region is the linear region indicated in the
figure.   Remote   sensing   systems   are   designed   to   measure   the   radiometric
characteristics of the targets of primary interest in this linear region, where the output
signal is linearly proportional to the input signal.
Hypothetical calibration response curve, with linear region of slope m and intercept b
This linear function is described by three parameters: the range of DN values in the
image, and the lowest (Lmin) and highest (Lmax) radiances measured by a detector
over the spectral bandwidth of the channel. Most commonly, the data are distributed
in  8-bit  format  corresponding  to  256  DN  levels.  Lmin  is  the  spectral  radiance57
corresponding to the minimum DN value (usually, a value of 0). Lmax is the radiance
corresponding to the maximum DN (usually, the value of 255).
For each spectral band, the output of the sensor, the grey level or DN, is related to the
input signal, the radiance L in most cases, by the following equation:
DN = G · L + D
The quantity G is the gain of the system and D is the dark current. All of the quantities
in the equation are a function of wavelength but, for simplicity, this dependence is not
indicated  in  the  equations  presented  here.  The  image  DN  recorded  by  the  sensor
system  can  be  converted  to  the  physical  quantity  L  (expressed  in  Wm -2   sr -1 )  by
inverting equation:
L = (DN - D) / G = (Lmax-Lmin)/255*DN + Lmin
The information about sensor calibration parameters (Lmin and Lmax) is usually
supplied  with  the  data  or  is  available  elsewhere  (e.g.,  Landsat  Data  User's
If the input signal exceeds the amount for which the sensor was designed, the system
response  will  become  non-linear  or  reach  the  saturation  level.  This  is  a  common
occurrence  in  land  remote  sensing  systems  when  they  image  bright  clouds  and/or
snow cover, for example. Uniformity Calibration
Many  remote  sensing  sensors  in  use  today  produce  images.  During  the  calibration
process it is important to determine how the system responds to signals coming from
different parts of the scene. The optics of a sensor system will be a determining factor.
A phenomenon know as vignetting will reduce the light reaching the detector system
towards the edges of the lens. Similarly, many sensor systems use arrays of detectors
to image scenes more quickly. The detectors involved always have some differences
in response to the same input signal level, which can lead to image striping. Thus, to
achieve  uniformity  in  response  across  the  field  of  view  of  the  sensor,  relative
radiometric calibration is also necessary. Spectral Calibration
To determine a sensor's response to light at different wavelengths, an input signal of a
known  wavelength  and  intensity  is  scanned  across  the  wavelength  range  and  the
response  as  a  function  of  wavelength  measured  and  characterized  for  each  spectral
band  prior  to  launch  (see  the  figure).  If  operational  applications  are  not  to  be
compromised, it is important to know the position and width of each spectral band, as
well as to understand out-of-band contributions from other spectral bands.58
Relative spectral response profile for a broad-band sensor Geometric Calibration
Due to optical aberrations or misalignment of discrete multiple detectors, images from
different spectral bands or detectors can be misregistered. In a framing sensor, straight
lines may appear curved. In this context the known input would be a test pattern of
lines and geometrical shapes. The resulting output image would be compared to the
input image and the appropriate geometric calibration equations implemented.
4.1.2 Calibration approaches Prelaunch Calibration
Much  of  the  time  spent  on  calibration  will  take  place  before  the  sensor  is  launched
into orbit or used operationally. Under controlled conditions in the laboratory, sensors
can  be  characterized  with  respect  to  their  radiometric,  spectral,  polarimetric  and
geometric  properties.  The  equipment,  time  and  effort  devoted  to  such  an  effort  are
significant, but they provide a sound understanding of a given sensor's performance
and facilitate stable operations on orbit over the lifetime of the mission.
Nevertheless,  a  satellite  sensor  may  be  calibrated  months  or  years  before  being  put
into  space.  By  the  time  the  sensor  is  in  orbit,  the  characteristics  of  the  sensor's
detectors  and  filters  have  often  changed  and  thus  the  preflight  calibrations  will  no
longer  be  optimum.  Despite  thermal  vacuum  testing  in  the  laboratory,  the  actual
launch  places  the  sensor  in  a  new  environment  which,  coupled  with  the  effects  of
aging before and after launch, lead to degradations in sensor responsivity  over  time
for all Earth observation sensor systems.
The following table shows as an example the change in the sensor gain coefficients as
a  function  of  time  for  the  popular  Advanced  Very  High  Resolution  Radiometer
(AVHRR)  sensors  carried  on  board  the  National  Oceanographic  and  Atmospheric
Administration (NOAA) series of satellites.59
Degradation of AVHRR sensor gain coefficients
Platform Degradation in % per year
Channel 1          Channel 2
NOAA-7 3.6 4.3
NOAA-9 5.9 3.5
NOAA-11 1.2 2.0 Onboard Calibration
It would be gratifying to conduct as vigorous a calibration onboard the spacecraft as
was  done  preflight.  For  many  reasons  this  is  never  possible.  However,  various
methods  have  been  developed  to  obtain  some  radiometric  calibration  measurements
onboard  the  spacecraft.  Few  if  any  sensor  systems  have  provisions  for  post-launch
checks  of  spectral,  polarimetric  and  geometric  sensor  characteristics.  Radiometric
reference targets that are sometimes used include onboard standard lamps and/or solar
diffuser reflectance panels, deep space, the Sun and the Moon. Vicarious Calibration
Vicarious calibration refers to techniques that make use of natural or artificial sites on
the surface of the Earth for the post-launch calibration of sensors. These targets are
imaged in near-coincident fashion by the sensor to be calibrated and by one or more
well-calibrated sensors from satellite or aircraft platforms or on the ground.
For  example,  dry  lake  beds  or  playas  present  bright  homogeneous  targets  of
significant  size  that  lend  themselves  well  to  vicarious  calibration.  They  can  be
characterized  on  the  ground  or  from  another  remote  sensing  platform  using  well-
calibrated sensors. With allowance for atmospheric and directional reflectance effects,
as well as any differences in spectral band characteristics, a comparison between data
of the same site from the reference sensor and the sensor under study can be used to
update the calibration gain coefficients of the sensor under study.60
4.2 Atmospheric correction - from radiance to reflectance or to
§     Erich   Hernandez-Baquero   (1999),   Proposal   of   Atmospheric   Compensation   for   Surface
Temperature and Emissivity Separation, A dissertation proposal submitted in partial ful_llment of
the  requirements  for  the  degree  of  Doctor  of  Philosophy  in  the  Chester  F.  Carlson  Center  for
Imaging Science of the College of Science Rochester Institute of Technology
§     ATmosphere  REMoval  Program  (ATREM),  User's  Guide,  Version  3.1,  Center  for  the  Study  of
Earth From Space (CSES), Cooperative Institute for Research in Environmental Sciences (CIRES),
University of Colorado, Boulder
§     Hook  Simon  J.  (1992),  A  comparison  of  techniques  for  extracting  emissivity  information  from
thermal infrared data for geologic studies, Remote Sensing of the Environment, 42; 123-135
§     Richter Rudolf (1991), Derivation of temperature and emittance for airborne multispectral thermal
infrared scanner data, Infrared Physical Technology, Vol. 35, No. 6, 817-826
§     Schowengerdt Robert A., Remote Sensing and Methods for Image Processing, 2 nd  ed., Academic
Press, San Diego, 520p
After preforming the radiometric correction, image values are expressed in radiance
values.  As  materials  can  be  characterized  by  their  reflectance  and/or  emissivity
spectrum, or by their thermal properties, one needs to apply some corrections to the
data, so that it will be expressed in reflectance values (for VIS-NIR passive Remote
Sensing),  or  in  emissivity  and  temperature  values  (for  thermal  passive  Remote
Sensing). These corrections are needed in order to:
•   comapre images from different sensors and/or dates.
•   preform environmental, ecological, or geophysical analyses.
•   improve classification results
In  order  to  do  these  corrections,  one  should  consider  the  complex  interactions  the
electromagnetic radiance arriving at the sensor goes through:
1.  The solar radiance reaching the upper atmosphere depends on the:
•   longitude and latitude
•   time (date and hour) of the year
2.  Going  through  the  atmosphere  (up  and  down),  the  Electromagnetic  radiance  is
affected by processes of:
•   (multiple) scattering
•   absorption
•   these  depend  on  the  spatial  and  temporal  distribution  of  the  different  gases,
molecules and aerosols in the atmosphere.
3.  The amount of incident radiance upon the surface, depends on the:
•   optical depth (height of the surface)
•   slope  and  aspect  of  the  surface  (defining  the  effective  angle  between  the
incident radiance and the surface)61
4.  Upon  hitting  the  ground,  the  radiation  is  partially  transmitted,  absorbed  and
reflected,  depending  on  the  physical  and  chemical  properties  of  the  surface,  and
on its roughness.
5.  The radiance reaching the sensor from the surface is also affected by the:
•   adjacency effect; atmopsheric scattering increases the autocorrelation between
adjacent pixels, the signal from a certain pixel partly mixed with signals from
its surrounding pixels.
•   optical length between the sensor and the surface, and the viewing geometry;
the more a pixel is located at the edge of an image and far from the nadir, the
more  the  signal  leaving  it  and  reaching  the  sensor  is  attenuated  by  the
The  higher  the  spectral  resolution  of  the  sensor,  the  more  sophisticated  will  be  the
correction  algorithms,  as  specific  absorption  of  gases  appear  (not  averaged  and
smoothed  away  by  wide  bands),  which  should  not  be  mistaken  with  absorption
features of the target materials one wants to identify.
As to date, there is no way to absolutely correct all these effects; different methods are
employed,  depending  on  the  purpose  of  the  work,  the  radiometric  and  atmospheric
data at hand, the hardware and the software. The more sophisticated ones recognize
the fact that the atmospheric attenuation is not uniform over the whole image (varying
topography, water vapor content, etc), and try to deal with it.
The different methods can be classified into the following, and they can be combined
or applied in steps (some of these methods are further explained below):
•   Disregarding the atmospheric and topographic effects, analysing the DN values.
•   Calibrating images from different dates to Like-Values; this technique states that it
is sufficient to convert the raw digital counts (DN’s) of one image to be consistent
with the counts for a chosen reference image, when comparing images to detect
•   Topographic correction;  correcting  for  the  effects  of  slope  and  aspect  (shading),
and  optical  path  length  (distance  between  sensor  and  surface)  using  a  Digital
Elevation Model.
•   Image based corrections:
Ø  Normalizing radiance values, in order to obtain values of relative reflectance.
These methods include the IARR (Internal Average Relative Reflectance) and
Ø  Extracting atmospheric correction from the image itself, e.g. water vapor and
the aerosol content (One of the greatest challenges in dealing with water vapor
is that its concentration changes with time, location, and altitude).
The estimation of the aerosol content is not yet operational. Statistical approaches use
the contrast reducing effect of strong scattering to estimate the aerosol content.
Histogram matching allows one to even obtain the spatial distribution of the
haze in the image - as long as the image data is statistically homogeneous. Simpler
approaches use dark tar-gets or a series of known spectrally homogeneous areas for an
estimate of the atmospherically scattered radiance within the image.
•   Multiple Frequency; Absorption of electromagnetic radiation by the atmosphere is
frequency  dependent.  By  measuring  the  radiance  at  different  wavelengths  a
correction can be calculated for the atmospheric absorption.62
•   Multiple  View;  The  path  length  of  atmosphere  traveled  by  radiation  affects  the
distance over which the radiation will be attenuated. If measurements are made of
the same location with different scan angles of known distance, a correction can
be made based upon this differential absorption.
•   Path  Length  Correction;  A  correction  can  be  included  which  attempts  to  correct
for varying scan angles. This generally applies a correction which is factored by
the path length of the line segment from the surface to the sensor.
•   Empirical line correction; Empirical corrections fully rely on the knowledge about
the spectrum of a group of pixels within the image (these are preferably gathered
using a field spectrometer at the time of the flight of the sensor). They are fast and
allow  a  pragmatic  processing  of  the  image  at  low  costs.  Their  disadvantage  is  a  low
reliability  in  mountaineous  terrain  and  for  changing  meteorologic  conditions  within  one
•   Adjacency  correction;  Each  pixel  has  to  be  corrected  with  respect  to  the  average
reflectance of the adjacent areas. This can be done by the definition of a spatial convolution
function  which  takes  a  distance-weighted  average  of  the  adjacent  area  in  the  image  to
calculate an adjacency weighting factor.
•   Atmospheric  modelling;  constructing  an  approximate  model  of  the  atmosphere,
based on world wide measurements of many atmopsheric parameters, gethered at
different  areas,  time  of  the  year,  and  atmospheric  layers.  Combining  the  above
data, the software looks up through the tables and models the atmosphere as it was
at the time of the flight. Atmospheric and meteorological data gathered at the time
of the flight, at the surface, or using a radiosonde, can improve the atmospheric
Analysing the results of an atmospheric correction, one should compare the absolute
values  of  reflectance  and  the  shape  of  the  reflectance  spectrum,  to  those  of  the
reflectance spectrum measured in the field or in a laboratory.
Below is given a more detailed description of some of the common methods used for
atmospheric correction of images.
4.2.1 Calibrating images from different dates to Like-Values
Invariant  targets  are  being  used  to  calibrate  the  digital  counts  of  coincident  or
overlapping  scenes  to  the  digital  counts  of  a  reference  image.  The  digital  counts  of
these targets are extracted from both images and robust regression techniques are used
to estimate the gains and offsets.
Guidelines for the Calibration of a Sequence of Images:
1.  Selection of reference image
The reference image is the scene to which the other scenes are related. It is
important that it is cloud free and contains no white outs (i.e. values >= 255).
2.  Selection of Invariant Targets
Invariant targets are features which have constant reflectance over time. The data
values are used to define linear functions to transform each overpass image to the
reference  image  by  assuming  these  targets  should  have  the  same  digital  count
values in each image. Targets must be selected for a range of bright, mid-range,
and dark values and you must have a balanced number of bright and dark targets.
Possible targets are listed below.
Dark: Ocean Lakes
Mid-range:  Rock outcrops Airfields63
Quarries, gravel pits and mines  Dams
Bright:        Catchments Sand
Targets  can  be  located  by  viewing  the  reference  image  on  the  screen  and  using
large-scale topographic maps. Vegetated targets should usually be avoided as they
show seasonal trends. When obtaining targets, it is important to locate targets over
a uniform section of the feature. . When obtaining targets, it is important to record
information about the target (Target type, size, line and pixel location, geographic
location, map sheet).
3.  Calculation of coefficiants
The next step is to calculate the regression coefficients which relate the overpass
images  to  the  reference  image.  Coefficients  for  several  methods,  such  as  least
squares,  s-estimation  and  weighted  least  squares  estimation  procedures  can  be
4.  Coefficient Examination
Once the  calibration functions have been calculated, the next step is to  examine
and select the best calibration line for each band. This is done by plotting the lines
and data. The y-axis is used for the reference image data and the x-axis is used for
the overpass image data.
Once the plots have been examined, the line of best fit is then chosen.
4.2.2 Internal Average Relative Reflectance (IARR)
The  IARR  calibration  technique  is  used  to  normalize  images  to  a  scene  average
spectrum.  This  is  particularly  effective  for  reducing  imaging  spectrometer  data  to
"relative  reflectance"  in  an  area  where  no  ground  measurements  exist  and  little  is
known  about  the  scene.  It  works  best  for  arid  areas  with  no  vegetation.  The  IARR
calibration is performed by calculating an average spectrum for the entire image and
using this as the reference spectrum. Apparent reflectance is calculated for each pixel
of the image by dividing the reference spectrum into the spectrum for each pixel.
4.2.3 Flat Field
The  "Flat  Field  Calibration"  technique  is  used  to  normalize  images  to  an  area  of
known "flat" reflectance. The method requires that you locate a large, spectrally flat,
spectrally  uniform  area  in  the  image  data,  usually  defined  as  a  Region  of  Interest
(ROI). The radiance spectrum from this area is assumed to be composed of primarily
atmospheric effects and the solar spectrum (that is, the spectrum of this ROI, which is
supposed  to  be  flat,  isn’t  so,  because  of  the  atmospheric  interference).  Relative
reflectance  is  calculated  for  each  pixel  of  the  image  by  dividing  the  reference
spectrum into the spectrum of each pixel.
4.2.4 Empirical line
The Empirical Line calibration technique is used to force image data to match selected
field reflectance. This method requires ground measurements and/or knowledge. Two
or more ground targets are identified and reflectance is measured in the field. Usually
the targets consist of at least one light and one dark area. The same two targets are
identified  in  the  image  and  average  spectra  are  extracted  for  Regions  of  Interest.  A
linear  regression  is  calculated  between  the  field  reflectance  spectra  and  the  image
radiance  spectra  (for  each  band  seperately)  to  determine  a  linear  transform  from
radiance to reflectance for each band of the data set. Gains and offsets calculated in
the regression are applied to the radiance spectra for each pixel to produce apparent
reflectance on a pixel-by-pixel basis.64
Empirical line scatter plot chart of a selected band, each point representing a Region of Interest
This  technique  assumes  that  the  atmospheric  effects  are  constant  across  the  scene,
assume that skylight scattered from the surface to the observer can be ignored, and it
is basically a single scattering solution for the atmosphere.
When using only two ROI’s, there is no redundancy, and therefore no computation of
the  RMS  of  the  empirical  line  model  is  possible.  When  more  than  two  targets  are
involved, by analysing the residuals on the scatter plot, it is possible to identify which
of the ROI’s serve well. Sometimes, different sets of ROI’s may be utilized in order to
preform  the  correction  over  different  ranges  of  the  spectrum.  The  method  is  also
applied as a further correction, after using an atmospheric model.
4.2.5 Atmospheric modelling
The state of a static atmosphere is described in terms of the parameters of the ideal
gas law:
P = nkT
where  P  is  the  atmospheric  pressure,  n  is  the  number  density,  k  is  Boltzmann's
constant,  and  T  is  the  temperature.  The  atmosphere  is  commonly  represented  as  a
stack of layers. Layers are defined in terms of temperature, composition, mixing, and
ionization distribution with altitude.
The mathematical description of the radiation propagation and emission processes that
lead to an observed radiance by a remote sensing platform is often referred to as the
“forward” model. The basic premise is that the observed radiance is a function of the
scene  in  view  and  the  composition  and  thermodynamic  state  of  the  intervening
atmosphere.  The  inverse  problem  (otherwise  known  as  retrieval  theory)  is  the  one  in
which the parameters affecting radiation are inferred from the observed radiance.
The radiometric formulation of radiative transfer is fairly standard across the literature
with the exception of the notation used. The radiance reaching a sensor is the sum of
the contributions from different propagation paths as shown in the following figure:65
Radiation propagation paths for reflective and thermal regions
The propagation paths depend on the region of the electro-magnetic spectrum of the
radiation  (see  the  figure  in  the  chapter  describing  the  electromagnetic  spectrum,
showing the respective emitted radiance of the Earth and the Sun).
In the reflective region, the dominant paths are:
A)  direct sunlight hits a target and reflects,
B)  sunlight scatters in the atmosphere and reaches a target and is then reflected,
C)  sunlight scatters in the atmosphere and reaches the sensor, and
G)  sunlight reflects off the background, reaches the target, and is then reflected.
In the thermal region, the dominant paths are:
D)  thermal photons emitted by a target reach the sensor,
E)  thermal radiation from the atmosphere reaches the target and is reflected,
F)  thermal photons from the atmosphere reach the sensor, and
H)  thermal photons from the background reach a target and are reflected.
Atmopsheric models need to account for:
•   the complex, time and space varying composition of the atmosphere, and
•   the  wavelength  dependent  interactions  experienced  by  the  radiance  reaching  the
A  generalized  form  of  the  radiative  transfer  equation  in  the  reflective  part  of  the
electromagnetic spectrum, is given below (this equation is wavelength dependent):
L sensor  = E 0 *cos(Q)*t*r/p + E d *t*r/p + L path
L sensor  = Radiance received at the sensor
E 0  = Solar radiance at the top of the atmosphere
Q = Incidence angle of solar radiance on the surface (0 for vertical, 90 for horizontal)
t = Transmittance factor
r = Reflectance factor
E d  = Scattered background radiation (of the sky)
L path  = Path scattered radiation reaching the sensor
The  radiative  transfer  equation  is  often  implemented  to  some  degree  in  computer
models  that  use  databases  containing  molecular  and  particulate  absorption  and
scattering  characteristics.  Various  models  exist  with  different  degrees  of  spectral
resolution, number of atmospheric constituents, cloud models, etc. Computer models
can  be  broken  down  into  two  major  classes:  band  transmittance  and  line-by-line
transmittance models. Band Transmittance Computer Models
These models are designed to lower the computational cost of computing the radiative
transfer  through  a  inhomogeneous  path  in  the  atmosphere.  This  is  done  by  fitting  a
band   model   (such   as   the   Lorentz   model)   to   line   spectra   obtained   through
measurements in the field or the laboratory. The model then simply uses parameters
such as the line width and strength to compute the overall absorption through a given
path. The most widely used model is the MODTRAN model developed by the United
States   Air   Force   Research   Laboratory   (AFRL),   Space   Vehicles   Directorate,
Battlespace   Environment   Division   (originally   a   group   within   the   Air   Force
Geophysics Laboratory). The current version of MODTRAN is version 4.
Band  models  are  useful  when  an  estimate  of  the  atmospheric  conditions  for  a
particular image collection is needed. In many instances, the approach involves using
a  standard  atmosphere  that  matches  the  particular  conditions  during  the  image
collection  (i.e.  choosing  the  “mid-latitude  summer  profile”  in  MODTRAN  for  an
image  collected  over  Oklahoma  on  a  summer  day).  Other  options  involve  using
radiosonde data as an input to the model. The spectral resolution of the MODTRAN
model  is  limited  to  contiguous  2  cm -1   bands  at  1  cm -1   intervals.  This  resolution  is
better  than  many  multispectral  sensors  and  reasonable  for  a  higher  resolution
hyperspectral sensor. However, as the channel widths of these new sensors decrease,
the band transmittance model starts to become inadequate due to errors arising from
spectral calibration. In other words, the sensor's band centers may not line up with the
band centers of the model. In that case, a line-by-line model may be needed. Line-by-line Models
Ground reflected Path scattered67
Line-by-line models are very high resolution radiative transfer models that use a large
database of molecular absorption and scattering measurements. The most widely used
database, also generated by the United States Air Force, is the HITRAN96 database.
The  model  FASCODE  taps  into  this  database  to  generate  high  spectral  resolution
transmittance calculations. The radiation propagation is then performed on a line-by-
line basis so that the Beer-Lambert Law holds. Because of this, the computation times
are larger and often less practical. The  University  of  South  Florida  has  released  the
latest version 2.51 of HITRAN-PC containing extensive updates, including the ability
to  read  the  1996  HITRAN  Database  available  on  CD-ROM.  The  1996  HITRAN
Database includes 999,363 absorption lines including data on new molecules.
Other  “fast"  forward  models  can  be  built,  such  as  PLOD  and  OPTRAN,  which  are
derivatives   of   the   GENLN2   model.   These   are   used   in   atmospheric   research
applications and is being currently used by the Atmospheric Infrared Sounder (AIRS)
team in support of the NASA EOS program.
As  Band  Transmittance  models  are  the  ones  more  used  for  Remote  Sensing
applications, a short description of some of these will be given. MODTRAN
MODTRAN  performs  accurate  and  speedy  calculation  from  the  UV  through  the
visible,  infrared  and  microwave  spectrum  (0  to  50,000  cm -1 ).  Calculations  in  the
visible and ultraviolet spectral regions are performed at lower spectral resolution (20
cm -1 ), while those in the IR and longer wavelengths are done at 2 cm -1  resolution.
PCModWin  (commercial  Windows  versions  of  the  Air  Force's  Phillips  Laboratory's
MODTRAN model) accommodates standard (WMO) atmospheric profiles, numerous
aerosol  models  (e.g.  fogs,  dust,  maritime,  etc.),  water  and  ice  cloud  models,  and
totally arbitrary geometric paths from sea level to 100 Kms. The molecular absorption
properties are based on the internationally recognized HITRAN atlas of spectroscopic
MODTRAN  itself  does  not  preform  an  atmospheric  correction  over  an  image
(ATREM and ATCOR, described later, receive as input an image, and their output is
the   same   image   atmospherically   corrected).   MODTRAN   receives   as   input   a
laboratory spectrum of some object (it can be also measured in the field, and be free
of atmospheric interferences), and its output are the spectrum of the same object as it
would be obtained on the sensor, and in addition many parameters of the atmospheric
profile, by wavelength and height above ground.
The workflow of using MODTRAN for preforming an atmospheric correction of an
image, goes on as follows:
1.  Preparing the reference background file containing the laboratory spectrum.
2.  Giving the software the input data (choosing from the available options) needed
for it to build the vertical atmospheric profile and calculate the parameters of the
radiative transfer model:
•   Average atmosphere model,
•   surface spectrum file and the object’s temperature,
•   aerosol model,
•   geometric  parameters  of  the  optical  path  (height  of  the  sensor  above  ground
and its oblique viewing angle),68
•   spectral resolution to preform the calculations, and
•   temporal  (day  of  year,  and  the  time  the  image  was  acquired)  and  locational
(longitude and latitude) data of the image
3.  Comparing the resulting output of the modelled “total radiance spectrum as seen
by the sensor”, with the actual radiance of the object as it is in the image.
If the modelled spectrum does not answer to out needs, we can change the input
parameters  given  to  MODTRAN,  in  order  to  approximate  more  closely  the
prevailing atmospheric conditions during the time the image was taken.
If the modelled spedctrum is good then we go and:
4.  Extracting the atmospheric parameters from the output file, and using them with
the radiative transfer model equation, on the image, in order to correct it.
When  using  the  output  parameters  for  the  radiative  transfer  equation,  one  should
remember the following points:
•   The  greater  the  reflectance  r  of  an  object,  the  value  of  L path   (path  scatered
radiation) will be higher. One can define a linear relation between the values of r
and  L sensor ,  by  running  MODTRAN  several  times,  each  time  with  a  theoretical
spectrally  flat  object  (e.g.,  0%,  10%,  20%,  30%,  60%),  and  thus  bypass  the
different values of L path  obtained when using different reference spectra.
•   This alone is not enough, as one should also take in account the sensor’s oblique
viewing  angle  (Initial  Zenith  Angle,  in  MODTRAN  terms).  For  Landsat,  this  effect  is
negligible and it can be assumed the entire scene is taken in the nadir. However,
for SPOT (which can be directed to “look” to the side), and for airborne scanners
(such as AVIRIS) that are closer to the ground and have a larger Field Of View, it
is important.
The  far  is  the  angle  from  the  nadir,  the  longer  is  the  optical  path,  there  will  be
more  scattering  on  the  one  hand,  and  more  absorption  on  the  other  hand.  These
two contrasting effects, create a difference between dark and bright objects.
In  the  far  angle,  over  dark  objects,  more  radiance  will  reach  the  sensor  (due  to
more scattering; in an angle of 30° the difference can reach more than 20%), while
over  bright  objects,  less  radiance  will  reach  the  sensor  (scattering  processes
having  a  smaller  ratio  of  the  total  radiance  at  the  sensor  relative  to  the  direct
reflectance  from  the  object,  therefore,  in  the  far  angle,  there  will  be  more
absorption over a bright object; in angle of 30° the difference can be between 5-
10%  in  specific  wavelengths  where  there  absorption  features  of  gases  in  the
To correct this, one can divide the image into several longitudinal stripes (5 to 10
degrees wide), computing in each one of them the regression between between the
values of r  and L sensor . This method, is implemented in ATCOR.
As MODTRAN is “building” an atmosphere over a reference spectrum, some of the
atmospheric interferences can not be taken in account:
•   the adjacency effect,
•   the topographic situation (shading, slope and aspect of the object’s location in the
real world), and
•   the atmospheric interferences above the scene are assumed to be uniform.
Ancillary data can help with the characterization of the atmosphere. This is often done69
through  radiosonde  measurements  in  which  a  sensor  is  mounted  on  a  balloon  and
launched  at  the  same  time  (or  as  near  as  possible  to  the  same  time)  as  the  time  of
image  acquisition.  These  radiosondes  then  measure  vertical  temperature,  pressure,
wind speed, and humidity profiles of the atmosphere. These profiles are then entered
into  a  radiative  transfer  model.  to  estimate  the  atmospheric  transmittance  and
The problem with radiosondes is that they are susceptible to drift during their ascent
and may not accurately represent the actual composition of the atmosphere for a given
column  of  air.  Furthermore,  the  logistics  of  successfully  launching  a  coincident
radiosonde for every remote sensing acquisition over the planet is impractical. Thus, it
is  preferable  to  develop  techniques  that  can  accomplish  atmospheric  compensation
using only the in-scene data inherent in an image.
This  task  has  not  been  possible  until  the  advent  of  hyperspectral,  or  imaging
spectrometer,  sensor  technology.  Here,  a  hyperspectral  sensor  is  any  remote  sensor
with  high  spectral  and  spatial  resolution.  The  spatial  resolution  aids  in  target
definition  and  measurement  of  small-scale  atmospheric  events.  By  increasing  the
number  of  spectral  measurements  of  the  Earth-leaving  radiance,  the  atmospheric
structure can be inferred directly from the measurement.
The ATmospheric REMoval Program (ATREM), described below, is an example to a
model that tries to do just that, for water vapor.
As it is very difficult to reconstruct with an atmospheric model the exact conditions
pevailing  at  the  time  of  the  flight,  an  emprical  line  correction  preformed  after  the
modelling of the atmosphere, might be helpful.
MODTRAN  can  be  downloaded  from  the  U.S.A.  Air  Force  Research  Laboratories,  at  the
following site:
The company marketing MODTRAN is ONTAR corporation (a demo version of the software can
be downloaded there): 2nd Simulation of Satellite Signal in the Solar Spectrum - 6S code
This code predicts the satellite signal from 0.25 to 4.0 micrometers assuming a cloud-
free  atmosphere.  The  main  atmospheric  effects  (gaseous  absorption  by  water  vapor,
carbon  dioxyde,  oxygen  and  ozone,  scattering  by  molecules  and  aerosols)  are  taken
into  account.  Non-uniform  surfaces  can  be  considered,  as  well  as  a  bidirectional
reflectance as boundary conditions.
The following input parameters are needed:
-    Geometrical conditions
-    Atmospheric model for gaseous components
-    Aerosol model (type and concentration)
-    Spectral condition
-    Ground reflectance (type and spectral variation)
At  each  step,  you  can  either  select  some  standard  conditions  (for  example,  spectral
bands of satellite for spectral conditions) or define your own conditions.
The authors of this package are:70
6S code: E. Vermote(3), D. Tanre(1), J.L. Deuze(1), M. Herman(1), J.J. Morcrette(2). Motif code: L.
(1)Laboratoire d'Optique Atmospherique
Universite des Sciences et Technologies de Lille
59655 Villeneuve d'Ascq Cedex - France
Reading - England
(3)Code 923 / GIMMS group:
Greenbelt, MD 20771 - USA ATmospheric REMoval Program (ATREM)
The  ATmospheric  REMoval  Program  (ATREM)  is  a  radiative  transfer  model-based
technique for deriving scaled surface reflectance from AVIRIS data without a priori
knowledge of surface characteristics. The atmospheric scattering is modeled using the
6S code.
The  spatial  and  temporal  variations  of  atmospheric  water  vapor  pose  difficulties  in
removing water vapor absorption features in hyperspectral data. In this algorithm, the
amount  of  water  vapor  is  derived  on  a  pixel-by-pixel  basis  from  the  data  using  the
0.94- and the 1.14-µm water vapor bands and a three-channel ratioing technique. The
derived water vapor values are then used for modeling water vapor absorption effects
in  the  entire  0.4-2.5  µm  region.  Together  with  the  solar  irradiance  curve  above  the
atmosphere,  and  transmittance  spectra  for  each  of  the  atmospheric  gases  CO2,  O3,
N2O,  CO,  CH4,  and  O2,  a  "scaled  surface  reflectance"  is  retrieved.  This  can  be
converted to real surface reflectance if surface topography is known.
For  six  gases  (CO2,  O3,  N2O,  CO,  CH4,  and  O2),  the  algorithm  assumes  that  the
amounts of the gases are uniform across the scene. Only one transmittance spectrum
is calculated for each of these gases.71
The algorithm treats the atmospheric water vapor differently. The atmospheric water
vapor concentrations vary significantly with time and altitude. The derivation of water
vapor values from imaging spectrometer data is mainly based on two facts. One is that
the surface reflectance curves vary nearly linearly with wavelength in the 0.94- and
the1.14-µm  water  vapor  band  absorption  regions  for  common  soils  and  rocks.  The
other is that under typical atmospheric conditions, the transmittances of the two water
vapor bands are sensitive to changes in the amount of atmospheric water vapor.
An apparent reflectance spectrum with relevant positions and widths of spectral regions used in the
three channel ratioing illustrated.72
A  three-channel  ratioing  technique  is  used  in  the  ATREM  derivation  of  the  water
vapor  value  from  a  radiance  spectrum.  The  mean  apparent  reflectance  at  the  water
vapor center is divided by one half of the sum of the mean apparent reflectances at the
two  window  regions.  The  ratio  effectively  removes  the  linear  surface  reflectance
effect and gives a mean observed transmittance for the 0.94-µm (and for the 1.14-µm)
water  vapor  band.  By  matching  the  mean  observed  transmittance  with  theoretically
calculated mean transmittances using atmospheric and spectral models, the amount of
water vapor in the Sun-surface-sensor path is obtained, and an atmospheric gaseous
transmittance spectrum corresponding to the pixel is then obtained.
The following systematic errors exist in ATREM:
•   The sensor is assumed to be looking only in nadir.
•   The atmospheric adjacency effect is not modeled.
•   Over mountainous terrain, one part of the terrain may be illuminated by scattering
from another part of the terrain. This introduces another kind of adjacency effect.
This  effect  can  be  refered  as  the  "topographic  adjacency  effect".  This  kind  of
adjacency effect is also not modeled in our program.
•   Some  of  the  surfaces,  such  as  vegetation,  snow,  ice,  and  iron-rich  soils  and
minerals, do not have linear reflectances in the 0.94- and the 1.14- µm water vapor
band  absorption  regions.  The  three  channel  ratios  calculated  from  the  data  over
these  surfaces  contain  the  surface  reflectance  effects.  Systematic  errors  are
therefore  introduced  in  the  derived  water  vapor  values.  In  order  to  decrease  the
errors in the derived water vapor amounts, the center positions and widths of the
window and water vapor absorption channels are all allowed to vary in the actual
implementation of the three-channel ratio technique.
ATREM can be obtained via anonymous ftp from or by contacting the Center for the
Study of Earth from Space at 303-492-5086. ATCOR
The  ATCOR  (ATmospheric  CORrection)  algorithm,  as  it  is  implemented  in  the
Remote Sensing software of ERDAS-Imagine, is designed for high spatial resolution
satellite  sensors  like  Landsat  Thematic  Mapper  (TM).  The  algorithm  works  with  a
catalogue of atmospheric correction functions stored in look-up tables. The catalogue
consists  of  a  broad  range  of  atmospheric  conditions  (different  altitude  profiles  of
pressure, air temperature, and humidity; several aerosol types; ground elevations from
0  to  1.5  km  above  sea  level;  solar  zenith  angles  ranging  from  0°  to  70°  ).  The
catalogue  covers  visibilities  (surface  meteorological  range)  from  5  km  to  80  km,
values can be extrapolated down to 4 km and up to 120 km. The 1996 edition of the
catalogue was compiled using the MODTRAN-2 and the SENSAT-5 codes.
The algorithm consists of an interactive and an automatic part. The interactive phase
serves for the definition of a reference target (dense dark vegetation or water) as well
as  haze  and  cloud.  The  reflectance  of  the  reference  target  in  a  single  spectral  band
(dark vegetation : TM band 3, water TM band 4) has to be specified. Additionally, the
image can be partitioned into subimages, called sectors. This phase also selects one of
the  atmospheres  available  in  the  catalogue,  i.e.  the  altitude  profile  of  pressure,
temperature and humidity as well as the aerosol type (e.g. rural) are fixed.73
The automatic phase first calculates the visibility in those regions containing reference
pixels.  The  visibility  is  obtained  by  matching  the  measured  signal  (i.e.  the  digital
number DN converted to a radiance using the sensor calibration) to the model-derived
signal  in  the  spectral  channel  of  known  target  reflectance.  The  sector-average
visibility over the reference regions is assigned to the non-reference areas.
The second step is the haze removal performed by histogram-matching the statistics
of the haze regions to the statistics of the clear part of the scene for each sector and
each  channel.  The  last  step  is  the  calculation  of  the  ground  reflectance  image
including  the  adjacency  correction,  and  the  computation  of  the  ground  brightness
temperature image (TM band 6).
Model  ATCOR2  is  restricted  to  satellite  sensors  with  a  small  swath  angle,  like
Landsat  TM,  MSS,  Resurs-01  MSU-E  and  SPOT.  The  model  assumes  a  terrain
consisting of horizontal surfaces of Lambertian reflectances.
It should be noted however, that the full version of ATCOR, outside of ERDAS, also
take in account the different viewing angles from the nadir and their influence on the
optical path, and can also handle hyperspectral image data.
Algorithm Developed by:
Dr. Rudolf Richter
DLR - IB 552-03/96
Institute for Optoelectronics
D-82234 Wessling
World-Wide Distribution and Support:
Riesstr. 10
82110 Germering
4.2.6 Temperature calibration of images
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
In  order  for  thermal  images  to  be  quantitative,  temperature  calibration  must  be
provided  for  the  scanners.  This  is  done  in  the  following  way:  electrically  heated
temperature calibration sources (Black Bodies) are mounted in the scanner, in either
side of the angular field of view. The scanner records the radiant temperature of the
first  calibration  source,  then  sweeps  the  terrain,  and  finally  records  the  radiant
temperature  of  the  second  source.  The  two  sources  are  set  at  two  different
temperatures, thus defining a scale for determining the temperature at any point along
the magnetic tape record of the scan line.
4.2.7 Thermal properties of materials
A thermal sensor picks up radiant emitted energy from a surface target heated through
radiation (solar insolation and sky radiance), convection (atmospheric circulation) and
conduction (through the ground). Thus, most sensed heat from surfaces has its origin
in  solar  illumination,  that  varies  with  both  diurnal  and  seasonal  changes  as  well  as74
cloud cover, but there is also a small, nearly constant contribution from internal heat
flux from the Earth's interior (much of this is due to thermal inputs from radioactive
decay).  Heat  is  transferred  into  and  out  of  near  surface  layers  owing  to  external
heating by the thermal processes of conduction, convection, and radiation.
A primary objective of temperature measurements and related thermal responses is to
infer something about the nature of the composition and other physical attributes of
materials  at  the  Earth's  surface  (and,  in  its  atmosphere).  For  any  given  material,
certain  characteristic  internal  properties  play  important  roles  in  governing  the
temperature of a body at equilibrium with its surroundings.
Heat  Capacity  (C):  The  measure  of  the  increase  in  thermal  energy  content  (Q)  per
degree of temperature rise. It is given in cgs units of calories per cubic cm. per degree
Centigrade, and its denotes the capacity of a material to store heat (recall from physics
that a calorie [cal] is the quantity of heat needed to raise one gram of water by one
degree  Centigrade).  Heat  capacity  is  calculated  as  the  ratio  of  the  amount  of  heat
energy,  in  calories,  required  to  raise  a  given  volume  of  a  material  by  one  degree
Centigrade  (at  a  standard  temperature  of  15°  Centigrade.)  to  the  amount  needed  to
raise the same volume of water by one degree Centigrade. A related quantity, specific
heat (c), is defined as C=c/r - units: calories per gram per degree Centigrade) where,
r = density ; this associates Heat Capacity to the thermal energy required to raise a
mass of 1 g(ram) of water by 1 degree Centigrade.
Thermal  Conductivity  (K):  The  rate  at  which  heat  will  pass  through  a  specific
thickness  of  a  substance,  measured  as  the  calories  delivered  in  1  second  across  a  1
centimeter  square  area  through  a  thickness  of  1  cm  at  a  temperature  gradient  of  1
degree Centigrade (units: calories per centimeter per second per degree Centigrade).
Thermal Inertia (P): The resistance of a material to temperature change, indicated by
the time dependent variations in temperature during a full heating/cooling cycle (a 24-
hour day for the Earth); defined as:
P = (Kcr) 1/2  = cr(k) 1/2
(k  is  a  term,  related  to  conductivity  K,  known  as  thermal  diffusivity,  in  units  of
calories per centimeter squared per square root of degree Centigrade seconds ). P is a
measure  of  the  heat  transfer  rate  across  a  boundary  between  two  materials.  e.g.,
air/soil.  Because  materials  with  high  P  possess  a  strong  inertial  resistance  to
temperature fluctuations at a surface boundary, they show less temperature variation
per heating/cooling cycle than those with lower thermal inertia.
Some characteristic values of these intrinsic thermal properties:
Water Sandy soil Basalt Stainless steel
K 0.0014 0.0014 0.0050 0.030
c 1.0 0.24 0.20 0.12
r 1.0 1.82 2.80 7.83
P 0.038 0.024 0.053 0.168
The interpretation of thermal data and images depicting temperature distribution over
an area is not a simple matter. In many instances, efforts must be confined to looking75
for patterns of relative temperature differences rather than the absolute values because
of the many complex factors that make quantitative determinations difficult, such as:
•   Number and distribution of different material classes in the instantaneous field
of view
•   Variations in the angle of thermal insolation relative to sensor position
•   Dependency  of  thermal  response  on  composition,  density  and  texture  of  the
•   Emissivities of the surface materials
•   Contributions from geothermal (internal) heat flux; usually small and local
•   Topographic irregularities including elevation, slope angle, and aspect (surface
direction relative to Sun's position)
•   Rainfall  history,  soil-moisture  content,  and  evaporative  cooling  effects  near
•   Vegetation canopy characteristics, including height, leaf geometry, plant shape
•   Leaf temperatures as a function of evapotranspiration and plant stress
•   Near surface (1 to 3 meters) air temperature; relative humidity; wind effects
•   Temperature history of atmosphere above surface zone
•   Cloud-cover history (during heating/cooling cycle)
•   Absorption and re-emission of thermal radiation by aerosols, water vapor, air
Some factors have fixed or constant effects; others vary with each sensor overpass. It
may be possible to correct for the influence of some of the variable factors but this is
difficult to do routinely. Measurements made at isolated individual points in a scene
and extrapolated to the general scene have limited validity.
Unlike  remote  sensing  of  reflected  light  from  surfaces  in  which  only  the  topmost
layers a few molecular layers thick are involved, thermal remote sensing is affected
by energy variations extending to varying shallow depths below the ground surface.
The most critical consideration in analyzing and interpreting thermal data and imagery
is that of knowing the physical and temporal conditions under which the near surface
layers are heated. Over the seasons, minor shifts in mean temperature in bedrock can
occur to depths of 10 meters (33 ft) or more. Materials at and immediately below the
surface are heated significantly during the day by incoming solar radiation and heat
transfer  from  the  air.  Temperatures  usually  drop  at  night  primarily  by  radiative
cooling  (with  maximum  radiative  cooling  under  cloudless  conditions),  accompanied
by some conduction and, for fluids, convection. During a single daily (diurnal) cycle,
the near surface layers (commonly, unconsolidated soils) experience alternate heating
and cooling to depths typically between 50 and 100 centimeters (20-40 inches). The
daily mean surface temperature is commonly near the mean air temperature. Observed
temperature changes are induced mainly by changes during the diurnal heating cycle
but seasonal differences (averages and range) in temperature and local meteorological
conditions also affect the cycle response from day to day.76
Changes in radiant temperatures of five surface-cover types during a 24-hour thermal cycle. From F.F.
Sabins, Jr., Remote Sensing: Principles and Interpretation. 2nd Ed., © 1987.
The  curves  shown  here  summarize  the  qualitative  changes  in  radiant  temperature
during a 24-hr cycle, beginning and ending at local midnight, for five general classes
of  materials  found  at  the  surface.  From  these  curves  one  may  estimate  the  relative
gray levels that would be recorded in a thermal image, as a function of the material
and  the  time  of  day.  Given  two  thermal  images  of  the  same  local,  taken  12  hours
apart,  about  noon  and  about  midnight,  one  might  determine  the  identities  of  co-
registered pixels based on their temperatures and their thermal inertia.
Thermal inertia cannot be measured remotely as each of the three parameters must be
measured  in  contact.  Instead,  Apparent  Thermal  Inertia  (ATI)  may  be  used  as  a
substitute, derived from the following equation:
ATI = (1 – albedo) / (T radmax - Trad min)
where  max  and  min  radiant  temps.  Can  be  obtained  from  day  and  night  thermal
Thermal crossover Radiant temperature curves of different materials may cross each
other  due  to  widely  differing  responses  of  each  by  day  and  by  night.  Occurs  at  or
about  local  sunrise  and  sunset.  Must  be  taken  into  account  in  mission  planning  in
order to maximise the probability of separating surface features by their temperature
Below  is  given  an  example  of  a  thermal  image  of  the  lakes  of  Erie  and  Ontario,
obtained from TM6. For experimental reasons, TM Band 6 on Landsat is occasionally
turned  on  at  night  to  obtain  thermal  images.  One  example,  a  full  scene  acquired  at
9:32  P.M.  on  August  22,  1982  shows  the  familiar  east  half  of  Lake  Erie  and  the
western part of Lake Ontario. The land appears moderately cool (darker tones), with
little detail, although the cities of Buffalo (east tip of Lake Erie), Toronto (top center)
and  Hamilton,  Ontario  (west  end  of  Lake  Ontario;  locally  hot  because  of  steel  mill
effluents) may be discernible on your monitor from street patterns and slightly lighter
(warmer)  tones.  A  mottled  pattern  of  variably  warmer  bands  characterizes  Lake
Ontario; these are related to thermal overturning effects (thermoclines) possible in this
deeper  (237  m  [782  ft])  lake.  Lake  Erie  is  uniformly  "hot"  because  its  shallowness77
(less  than  67  m  [220  ft])  inhibits  this  type  of  circulation.  Warm  rivers,  such  as  the
Niagara connecting the two lakes, stand in contrast to the land.
4.2.8 Retrieval of temperature and emissivity from radiance in thermal
The surface leaving radiance is the product of the temperature and emissivity effects.
Therefore, for a given radiance measurement, there is an infinite number of solutions
for  temperature  and  emissivity.  The  accuracy  of  the  temperature  measurement  is
therefore  directly  related  to  the  accuracy  of  the  estimate  of  the  target  emissivity.
Research has shown that emissivities must be known to accuracies of 0.01 or less in
order to obtain adequate estimates of temperature.
When measuring temperature over the water, the task is simplified by the fact that the
emissivity  is  well  known  and  spectrally  at  in  the  infrared.  Thus,  it  is  possible  to
measure  the  temperature  accurately  with  a  radiometer  that  has  a  limited  number  of
broad bands. Because of this, operational systems measuring ocean temperature have
been successful for years. When applying these algorithms to land, however, the result
is  not  the  same.  Unless  a  priori  knowledge  of  the  target  exists,  the  emissivity  of  a
particular pixel in a remotely sensed image is unknown.
In many applications, it is the emissivity itself that is of interest. The emissivity can be78
used, for instance, in a spectral classi_cation algorithm. The underlying assumption is
that a material can be uniquely described by a spectral emissivity curve. This inherent
property  of  the  material  is  then  used  to  classify  objects  in  a  scene.  To  do  this,
classification algorithms must have a library of spectral curves that correspond to each
material  of  interest.  The  measured  emissivity  is  then  compared  to  the  curves  in  the
library and the curve that matches it the most is selected.
The major challenge with this problem, however, is that the spectral surface radiance
consists of N measurements  while  the  unknown  spectral  emissivity  and  temperature
add  up  to  N  +  1  unknowns.  Therefore,  this  is  an  underdetermined  mathematical
problem  (different  materials  with  different  kinetic  temperatures,  can  emit  the  same
The  general  radiative  transfer  equation  in  the  thermal  region  of  the  electromagnetic
spectrum, is given below (the equation is wavelength dependent):
L sensor  =  e earth *E earth *t/p +  e sky *E sky *r*t/p + L path
L sensor  = Radiance received at the sensor
e earth  = Surface emissivity (usually between 0.7 to 1)
E earth  = Emittance of radiance of the surface according to Plank’s law (as a black body)
t = Transmittance factor (of the atmosphere)
r = Reflectance factor (of the surface; according to Kirchof’s law, equals to 1- e earth )
e sky  = Sky emissivity (usually between 0.5 to 0.75)
E sky  = Emittance of radiance from the sky according to Plank’s law (as a black body)
L path  = Upward Path emitted radiance
The  contribution  of  the  three  components  of  the  radiative  transfer  equation  int  the
thermal region (between 8.5 to 13mm), to the radiance at the sensor, as modelled by
Modtran for sand at 45 degrees, is as follows:
•   surface  reflected  –  less  than  3%  -  due  to  the  low  reflectance  spectrum  of  most
materials in the thermal region.
•   path thermal – between 15% to 50%, increasing with longer wavelengths, as an
effect  of  the  summing  together  of  emitted  radiance  from  different  atmospheric
layers; the higher the layer, the colder it is, the peak of emitted radiance moving
towards longer wavelengths (Wein’s law).
•   surface emission – between 50% to 85%, decreasing towards longer wavelength
(this depends off coarse upon the temperature of the surface being mointored, and
its emissivity).
There are several methods for atmospheric correction in the thermal region, trying to
extract  both  the  kinetic  temperature  of  the  surface  and  its  emissivity.  As  these
methods  are  more  sophisticated  (some  already  implemented  in  available  softwares),
they will not be dealt here, and one is invited to read in the apropriate journals.
Surface reflected      Path thermal Surface emission79
4.3 Geometric corrections
§     Domenico Visintini, Digital data stretching techniques,
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
All  remote  sensing  imagery  are  inherently  subject  to  geometric  distortions.  These
distortions  may  be  due  to  several  factors,  including:  the  perspective  of  the  sensor
optics;  the  motion  of  the  scanning  system;  the  motion  of  the  platform;  the  platform
altitude, attitude, and velocity; the terrain relief; and, the curvature and rotation of the
Earth. Geometric corrections are intended to compensate for these distortions so that
the  geometric  representation  of  the  imagery  will  be  as  close  as  possible  to  the  real
world. Many of these variations are systematic, or predictable in nature and can be
accounted  for  by  accurate  modeling  of  the  sensor  and  platform  motion  and  the
geometric  relationship  of  the  platform  with  the  Earth.  Other  unsystematic,  or
random, errors cannot be modeled and corrected in this way. Therefore, geometric
registration of the imagery to a known ground coordinate system must be performed.
Satellite  images  can  be  acquired  at  different  processing  levels  regarding  their
geometric quality; Examples will be given for the Landsat and SPOT satellites. The
methods for the geometric correction of satellite images are the ones that will be given
first.  Images  acquired  from  airborne  platform  suffer  from  more  severe  geometric
distortions.  The  methods  for  correcting  these  will  be  described  at  the  end  of  this
Systematic Errors:
Scan Skew: Caused by the forward motion if the platform during the time required
for  each  mirror  sweep.  The  ground  swath  is  not  normal  to  the  ground  track  but  is
slightly skewed, producing cross-scan geometric distortion
Mirror-Scan  Velocity  Variance:  The  mirror  scanning  rate  is  usually  not  constant
across a given scan, producing along-scan geometric distortion.
Panoramic Distortion: The ground area imaged is proportional to the tangent of the
scan  angle  rather  than  to  the  angle  itself.  Because  data  are  sampled  at  regular
intervals, this produces along-scan distortion
Platform Velocity: If the speed of the platform changes, the ground track covered by
successive mirror scans changes, producing along-track scale distortion
Earth Rotation: Earth rotates as the sensor scans the terrain. This results in a shift of
the ground swath being scanned, causing along-scan distortion.
Perspective:  For  some  applications  it  is  desirable  to  have  images  represent  the
projection  of  points  on  Earth  on  a  plane  tangent  to  Earth  with  all  projection  lines
normal to the plan. This introduces along-scan distortion.
Nonsystematic Errors
Altitude  Variance:  If  the  sensor  platform  departs  from  its  normal  altitude  or  the
terrain increases in elevation, this produces changes in scale
Platform Attitude: One sensor system axis is usually maintained normal to Earth's
surface  and  the  other  parallel  to  the  spacecraft's  direction  of  travel.  If  the  sensor
departs form this attitude, geometric distortion results.80
Geometric distortions of Landsat images
4.3.1 Geometric registration
The  geometric  registration  process  involves  identifying  the  image  coordinates  (i.e.
row, column) of several clearly discernible points, called ground control points (or
GCPs),  in  the  distorted  image  (A  -  A1  to  A4),  and  matching  them  to  their  true
positions in ground coordinates (e.g. latitude, longitude). The true ground coordinates
are typically measured from a map (B - B1 to B4), either in paper or digital format.
This  is  image-to-map  registration.  Once  several  well-distributed  GCP  pairs  have
been identified, the coordinate information is processed by the computer to determine
the proper transformation equations to apply to the original (row and column) image
coordinates  to  map  them  into  their  new  ground  coordinates.  Geometric  registration
may also be performed by registering one (or more) images to another image, instead
of to geographic coordinates. This is called image-to-image registration and is often
done  prior  to  performing  various  image  transformation  procedures,  which  involve
comparing images from different sensors or dates.81
There are several types of transformations to be applied on image coordiantes, so that
the image coordinates will transform into real world coordinates:
1.  Plane transformations are those which keep lines straight, being on the first order.
2.  Curvilinear (polynomial) transformations are higher order transformations that do
not necessarily keep lines straight and parallel.
3.  Triangulation.
4.  Piecewise   transformations   break   the   map   into   regions,   apply   different
transformations in each region.
The above transformations can be calssified as follows:
•   They are all point based transformations (the Ground Control Points).
•   The plane and polynomial transformations are global interpolators, determining a
single  function  that  applies  to  the  whole  region  (a  change  in  one  pair  of
coordinates affects the entire image).
Local interpolators (as the trinagulation, see below) apply an algorithm repeatedly
to a small portion of the total set of points.
•   Exact interpolators  honor  the  data  points  upon  which  the  interpolation  is  based,
i.e., the GCP’s are put exactly where supposed to (in triangulation).
With  approximate  interpolators  there  is  some  uncertainty  about  the  resulting
location of the GCP’s (in the plane and polynomial transformations). This utilizes
the belief that the whole image suffers from a uniform distortion (a global trend).
•   They are all deterministic methods, not using the theory of probability.
•   The plane and polynomial transformations are gradual  interpolators,  while  with
the triangulation method, abrupt changes can be intorduced.
•   They all asume the surface to be a plane, having relatively no relief variations. In
mountainous  areas,  georectification  of  the  image  needs  also  the  use  of  a  Digital
Elevation  Model  (DEM).  –  see  below  the  discussion  about  relief  displacement,
and about parametric geocoding. Plane transformations
Let  (x,y)  be  th  location  of  the  object  in  the  old  coordinate  system,  and  (u,v)  the
location  of  the  object  in  the  new  coordinate  system  (after  the  transformation).  The
four basic types (primitives) of plane transformations keep parallel lines parallel:
1.  Translation.
origin is removed, axes do not rotate.
u = x – a; v = y - b
origin is moved a units parallel to x and b units parallel to y.82
2.  Scaling.
both origin and axes are fixed, scale changes.
u = cx; v = dy
scaling of x and y may be different, if so, the shape of the object will change.
3.  Rotation.
origin fixed, axes move (rotate about origin).
u = x cos(a) + y sin(a); v = -x sin(a) + y cos(a)
a is measured counterclockwise.
4.  Reflection.
coordinate system is reversed, objects appear in mirror image.
to reverse y, but not x: u = x; v = c – y
this transformation is important for displaying images on video monitors as the
default coordinate system has the origin in the upper left corner and coordinates
which run across and down.
Usually a combination of these transformations (and some others) will be needed, as
the  distortion  of  the  image  are  complex,  and  cannot  be  attributed  to  one  of  the
In the following table are given the different kinds of complex plane transformations,
with their names, the unknown parameters, the distortion that will be experienced by
the  image,  and  the  minimum  number  of  ground  control  points.  Up  to  6  parameters,
parallel lines stay parallel.
Kind of
Deformation of the image        Unknown parameters        Minimum
number of
Congruency “roto-translation”, preservess
shape and scale
3 (E 0 , N 0 , Q) 2
Conformality         “roto-translation with (isotropy)
scale variation”, preserves shape
4 (E 0 , N 0 , Q, l) 2
“roto-translation with two
different scale variation”
5 (E 0 , N 0 , Q, l, m) 3
Affinity “roto translation with two
different scale variation and a
sliding angle”
6 (E 0 , N 0 , Q, l, m, g) 3
Longitudinal /
“roto-translation with two
different scale variation, a
sliding angle and a longitudinal /
transversal convergence angle”
7 (E 0 , N 0 , Q, l, m, g, g/h) 4
Homography “roto-translation with two
different scale variation, a
sliding angle and two
convergence angles” (projective
/ perspective transformation)
8 (E 0 , N 0 , Q, l, m, g, g, h) 4
E 0 East coordinate of the image frame g Transversal convergence angle
N 0 North coordinate of the image frame h Longitudinal convergence angle
Q Rotation angle from the image frame to the world coordinate frame
l Scale variation from the image frame to the world coordinate frame in x-direction
m Scale variation from the image frame to the world coordinate frame in y-direction
g Sliding angle in the world coordinate frame83 Polynomial Transformations
Simple linear transofrmation equations can be extended to higher powers:
u = a + bx + cy + gxy or u = a + bx + cy + gx 2  or u = a + bx + cy + gx 2  + hy 2  + ixy
equations of this form create curved surfaces, thus, straight lines may become curved.
The  higher  the  power  of  the  equation,  the  more  will  be  the  minimum  number  of
control points.
They usually give greater accuracy:
•   greater accuracy in the sense that when used to transform the control points, the
equations  faithfully  reproduce  the  known  coordinates  in  the  other  system  (the
RMS value will be lower)
•   however if error in measurement is present, and it always is to some degree, then
greater accuracy may not be desirable
•   a polynomial transformation may be more accurate for the control points, but less
accurate on average (the image may suffer great distortions). Triangulation
The  triangulation  method  can  be  seen  as  a  special  case  of  a  plane  transformation,
going on in the following way:
•   A  Triangulated  Irregular  Network  (TIN)  is  constructed,  one  for  the  GCPs  with
image coordinates, and one for the GCPs with real world coordinates (connecting
the same GCPs in both TIN’s).
•   Using  the  3  pairs  of  coordinates  in  each  triangle,  a  6  parametrs  affinity
transformation is performed, a different one for each triangle.
•   All  GCP’s  coordinates  are  honored,  the  method  assuming  there  is  no  error  in
chosing and locating them.
•   Abrupt distortions may be observed in the transformed image, if there are errors in
locating a GCP, or when adjacent triangles experience different distortions.
•   Resampling of pixels will occur only inside the TIN, the areas outside them being
“cut off” from the resampled image (unless another transformation is applied on
When an image was taken over a hilly area and we don’t have the DEM of that area,
or  when  an  airborne  image  (with  roll,  yaw  and  pitch  distortions)  was  taken  and  we
don’t   have   the   navigation   data   of   the   airborne   platform,   the   Triangulation
transformation might be our best option. However, a very large set of GCP needs to
be  collected  (~100),  and  the  result  will  be  not  as  accurate  as  the  one  that  would  be84
achieved using the navigation data and the DEM (see the discussion about parametric
geocoding below). Ground Control Points
In  order  to  compute  the  parameters  fot  the  transformation,  a  set  of  Ground  Control
Points (or tics) must be identified. The following rules can be given for selecting the
•   The  minimum  number  of  tics  should  be  equal  to  the  number  of  the  unknown
parameters.  However,  the  number  of  tics  collected  at  practice,  should  be  higher
than the minimum, in order to use the least squares adjustment method. The basic
idea, is to consider more information (observations) than those are strictly needed
to determine the unknowns (as in our tics there are always errors); than, to solve a
redundant linear equations system by performing an estimate (of the accuracy of
our transformation).
•   These control points must not be on a straight line (not collinear).
•   Best located to cover the edge, but also include some in the middle if possible.
•   Tics should be located much more carefully and accurately than the regular data.
•   The following kinds of points may be used as tics:
•   grid  intersections  and  corner  points  (relevant  for  scanned  maps,  not  for
•   prominent features, e.g. major intersections, depending on scale
•   marks made on the ground for the project, e.g. white crosses for aur photos
To each GCP is calculated the residual value (from the least square estimation), and
an overall RMS (Root Mean Square) error value is given. These indicate the accuracy
of the analysis. The following points should be noticed:
•   the spatial distribution of residuals may indicate weaknesses of the model.
•   may show the image has been distorted unevenly
•   magnitude of the residuals gives an estimate of the accuracy of the transformation
•   the desired RMS value equals the higher of the following:
Ø  the spatial resolution of the image
Ø  the scale/accuracy of the map being used as the reference for the correction
Ø  the  spatial  accuracy  of  the  real  world  coordinate  reference  map  for  the
correction, should be highrt than that of the image
•   for example, if we are working on the PANchromatic band of the SPOT image
(spatial resolution of 10 meters), the scale of the map being used for correcting
the image should be at least of 1:20,000 , the greater the better (or using DGPS
for locating the GCP’s). An RMS value of half the pixel size is very good.
Quantifying geocoding quality is difficult. Methods typically applied are:
•   calculate the location residuals of ground control points which were not used for
the prior calculation,
•   compare  the  image  results  with  the  DEM  along  terrain  lines  or  in  specific
mountainous areas,
•   overlay digital vector or raster maps on the geocoded results.85 Resmapling
In  order  to  actually  geometrically  correct  the  original  distorted  image,  a  procedure
called  resampling  is  used  to  determine  the  digital  values  to  place  in  the  new  pixel
locations  of  the  corrected  output  image.  The  resampling  process  calculates  the  new
pixel values from the original digital pixel values in the uncorrected image. There are
three  common  methods  for  resampling:  nearest  neighbour,  bilinear  interpolation,
and cubic convolution.
Nearest  neighbour  resampling  uses  the  digital  value  from  the  pixel  in  the  original
image which is nearest to the new pixel location in the corrected image. This is the
simplest method and does not alter the original values, but may result in some pixel
values  being  duplicated  while  others  are  lost.  This  method  also  tends  to  result  in  a
disjointed or blocky image appearance.
Nearest neighrbour Bilinear interpolation
Bilinear  interpolation  resampling  takes  a  weighted  average  of  four  pixels  in  the
original  image  nearest  to  the  new  pixel  location.  The  averaging  process  alters  the
original pixel values and creates entirely new digital values in the output image. This
may be undesirable if further processing and analysis, such as classification based on
spectral response, is to be done. If this is the case, resampling may best be done after
the classification process.
Cubic convolution86
 Cubic  convolution  resampling  goes  even  further  to  calculate  a  distance  weighted
average of a block of sixteen pixels from the original image which surround the new
output pixel location. As with bilinear interpolation, this method results in completely
new  pixel  values.  However,  these  two  methods  both  produce  images  which  have  a
much sharper appearance and avoid the blocky appearance of the nearest neighbour
method. Relief displacement
§     Antonio Arrighi, Elements of Photogrammetry and Air Photography with notes Photobathymetry,
1 st  Course on Port and Coastal Hydrography, IMO-IMA
§     Domenico Visintini, Errori nel Raddrizzamento,
The scale of an aerial photograph varies with the flying height of the aircraft. Thus,
variations  in  elevation  of  objects  on  the  ground  cause  scale  variations  in  aerial
photographs.  The  same  can  be  said  about  images  acquired  by  satellite  or  by
In  general,  the  higher  the  elevation  of  an  object,  the  farther  the  object  will  be
displaced  from  its  actual  position  away  from  the  principal  point  of  the  photograph
(the point on the ground surface that is directly below the camera lens; in the case of a
satellite  image,  there  is  no  prinicpal  point  –  rather,  for  each  scanned  line  there  the
pixel  located  directly  below  the  sensor,  at  the  nadir).  Conversely,  the  lower  the
elevation of an object, the more it will be displaced toward the principal point. This
effect, is called relief displacement.
Compare the map and photograph below. Both show the same gas pipeline clearing,
which passes through hilly terrain. Note the deformation of the pipeline route in the
photo relative to the shape of the route on the topographic map. The deformation in
the photo is caused by relief displacement.87
The equation for evaluating relief displacement is as follows:
d = r*h / H
where  d =       relief displacement
r =       distance on the image from the nadir to the displaced object (d and r
are expressed in the same units)
h =       height above datum of the object point whose image is displaced
H =      flying height above the datum selected for measurement of h
The characteristiscs of relief displacement are as follows:
•   d increases with increase in distance from the nadir (r)
•   d increases with increase in height of object (h) above datum (selected datum for
the image; e.g. the lowest point in the image, or the average height of the image)
•   d decreases with increase in flying height (H)
•   at the nadir, r = 0, d = 0
•   can be either outward (for points above datum) or inward (for points below datum)
•   causes objects to lean
•   causes straight lines on ground to appear crooked on photo
The following figure describes the basic geometry of relief displacement:88
Applying the above equation for two sensors, one on board of a satellite (SPOT), the
other  airborne  (DAIS),  gives  an  example  of  the  effect  of  relief  displacement  in
Remote  Sensing  images  (in  this  calculation  the  curvature  of  the  Earth  was  not
When SPOT is not directed to look obliquely, the far angle at the edge of the image is
about  2.06º.  It  can  be  seen  that  for  that  angle,  the  effect  of  relief  displacement  is
smaller  than  half  the  pixel  size  (10m  for  the  mono-chromatic  band  of  SPOT)  for
objects 100m above or below the image datum (thus, it is negligible). However, if the
imaged  area  is  mountainous,  or,  if  the  image  was  taken  at  some  oblique  angle,  the
effects of relief displacement cannot be neglected.89
For  airborne  scanners,  these  effects  of  relief  displacement  are  more  severe,  as  the
flying height is much lower, and the Field Of View is much larger (the FOV of the
DAIS is about 64º, compared to the 4.1º FOV of the SPOT). Thus comes the need for
parametric geocoding, discussed below.
When  performing  geometric  correction  of  an  image,  GCP  errors  due  to  relief
displacement in control points can be corrected, whenever elevations are available for
each point pair. This will be done prior to the calculation of the image transformation
from image coordinates to real world coordinates.
First, the image reference height datum needs to be determined. Then, for each GCP,
its  relief  displacement  on  the  image  (relative  to  the  nadir  pixel  for  each  line,  for
example)  should  be  calculated,  and  then  this  should  be  applied  to  the  image
coordinates  of  the  GCP  (relief  displacement  in  meters,  divided  by  the  pixel  size  in
meters, gives the relief displacement in pixels – thus changing the image coordinates
of the GCP in a positive or negative way –according to the relation between the object
and the image datum [above or below]).
By correcting the control points, relief is not introduced into the geometric model; it
should be remembered however, that relief in the imagery will not be corrected.90
4.3.2 LANDSAT - Geometric Characteristics
When   buying   satellite   images,   it   is   often   possible   to   acquire   them   already
geometrically  corrected,  in  different  levels  of  correction.  Here  will  be  given  an
example for this, regarding the LANDSAT satellite. TM Geometric Accuracy
The raw image contains geometric distortions induced by characteristics of the sensor.
These characteristics include:
•   non-linear mirror scanning velocity
•   varying average mirror speed between scans
•   sequential detector sampling (rather than "snapshot")
•   detector offsets in the focal plane.
The resulting distortions are modelled and corrected during product processing.
Geometric correction removes geometric distortions in an image based on knowledge
of the satellite and sensor, and re-maps the image to a regular grid in a standard map
projection. This is accomplished by constructing a mapping between pixel coordinates
in the image and geographic coordinates on the surface of the Earth.
The intrinsic geometric characteristics of the sensor, plus the knowledge of the precise
orbital  and  attitude  parameters  at  the  moment  of  acquisition,  make  it  possible  to
geolocate each pixel with an accuracy down to one half of the GSD (15m for Bands 1-
5 and 7). TM Data Processing Levels
Three  basic  levels  of  geometric  processing  are  available:  raw,  system  corrected  or
geocoded (levels A, B or C): Raw Data
LANDSAT  raw  data  is  available,  in  a  digital  format,  for  full,  quarter  and  mini  scenes  for  European
data, for full and sub-scenes for Worldwide data. A raw data product has no radiometric or geometric
correction,  but  the  scan  lines  are  re-aligned  in  the  same  across-track  direction.  In  addition,  Band  6
pixels are replicated (16x) to unify the geometry of all seven bands.
RAW  data  needs  to  be  processed  with  specialised  equipment.  This  format  is  not  suitable  for
inexperienced users System Corrected Products
In addition to the processing of raw data, System Corrected products are corrected for the geometric
distortion  caused  by  the  Earth's  curvature  and  rotation,  the  satellite's  attitude  and  the  "panoramic
distortion" inherent in the scanning geometry. Moreover, radiometric calibration is applied through a
look-up  table  that  models  the  detectors'  known  non-linearity.  System  correction  uses  one  of  two
resampling  algorithms  (`Cubic  Convolution'  or  `Nearest  Neighbour')  chosen  by  the  customer  when91
ordering the product; the defaults are Nearest Neighbour for European data and Cubic Convolution for
other data (more information about the algorithms is given below, see "Resampling Algorithm").
As the correction process is aiming to represent the curved Earth's surface in the flat plane of a two-
dimensional data array, System Corrected products have to use a map projection and Earth ellipsoid to
create  a  model  of  the  earth's  form  in  order  to  transform  the  pixels  The  built-in  map  projection  of
European data products is Space Oblique Mercator (SOM), which was developed specifically for use in
LANDSAT images because there is no distortion along the relatively narrow band of the satellite track
(path). The Earth ellipsoid is GRS 1980.
10-Year-Old U.S. data products are only available to this level of correction. Geocoded Products
Further  corrections  are  available  with  European  and  Worldwide  Geocoded  products.  These  are
precision processed products which, in addition to being system corrected, are geometrically rectified
according to the customer's specifications with respect to map projection, reference ellipsoid and pixel
alignment; furthermore any residual striping is suppressed by filtering and other statistical processing
operations.  Full  details  of  the  projections  and  ellipsoids  that  can  be  used  are  given  in  the  following
Three levels of geocoding are available, providing different degrees of accuracy of the final product, as
follows: Level A - `Without Ground Control Points'
The  information  about  the  orbit  and  attitude  of  the  satellite  which  is  downlinked  along  with  the
image data, is used to geolocate each pixel to a map projection and a reference ellipsoid which are
chosen by the customer. The absolute location accuracy is the same as for System Corrected (Path
Corrected) products - about 500 m; however, the internal geometry of the image is improved. The
orientation of the output image can be `true north', `along track' or any other specified angle. Level B `With Ground Control Points'
Geometric rectification is carried out using ground control points from maps, or from geodetic or
photogrammetric measurements. It is anticipated that maps will be provided by the customer, but
Eurimage is able to supply such maps (at cost) if required. The projection and reference ellipsoid
of the rectified image will correspond to that of the map unless the customer requests otherwise.
The quality of the ground control points directly affects the accuracy of the rectified image. The
rms  error  in  the  adjusted  satellite  model  is  typically  half  the  nominal  pixel  size,  provided  that
sufficiently accurate control points have been supplied. Terrain displacement errors, which are not
corrected  by  Level  B  geocoding,  are  proportional  to  the  instrument  viewing  angle  and  the  true
terrain variation within the image. It should be noted that this correction processing also corrects
for  non-systematic  distortions  such  as  those  caused  by  instantaneous  variation  of  the  satellite's
pitch and yaw. Level C `With Ground Control Points plus DTM'
In  addition  to  rectification  using  control  points,  images  are  corrected  pixel-by-pixel  for  local
terrain displacement errors by utilising a Digital Terrain Model (DTM); it is anticipated that this
will  be  provided  by  the  customer.  The  result  is  a  satellite  orthoimage.  The  location  accuracy  is
15m, even in areas of high terrain.92
4.3.3 SPOT – processing levels
The raw data received from the satellite undergo preprocessing with the aim of:
•   compensating for the instrument distortions,
•   generating ancillary data which brings information needed for the use and/or
the interpretation of the image data,
•   getting the product into its final packaging.
The preprocessing levels range from the least to the most sophisticated level (1A to
Ortho), covering a wide spectrum of utilization. Level 1A – no geometric correction
Only detector equalisation is performed: it consists in compensating for the differences of sensitivities
between the elementary detectors of the CCD (Charged Coupled Device) arrays, using a linear model.
Absolute  calibration  coefficients  posted  in  the  ancillary  data  can  be  used  to  convert  the  pixel  counts
into irradiance values.
No  geometric  corrections  are  performed,  and,  when  displayed,  the  SPOT  Scene  image  is  a  square.
Ancillary data (coordinates of the scene centre as well as the four corners) allow to locate the image
with an accuracy better than 500m (rms). Other information (ephemeris, attitude, look directions) can
be found and used to perform precise geometric processing. When displayed, the SPOT Scene is then a
These images are recommended for all people wishing to apply their own geometric processing such as
orthorectification  or  DEM  (Digital  Elevation  Model)  generation.  Level  1A  products  are  intended
specifically for users requiring scene data that have undergone minimum processing. This is essential
in  cartographic  applications,  for  further  precision  geometric  corrections  or  for  stereo  plotting,  and  in
detailed radiometric studies. Level 1B – compensating internal geometric distortions
The  same  detector  equalisation  as  for  level  1A  is  performed  (it  consists  in  compensating  for  the
differences of sensitivities between the elementary detectors of the CCD arrays). Absolute calibration
coefficients posted in the ancillary data can be used to convert the pixel counts into irradiances. The
geometric  corrections  are  performed  with  the  aim  of  compensating  for  the  internal  distortions  of  the
image caused by the imaging conditions (variations of the attitude of the spacecraft, panoramic effect,
earth curvature, rotation of the earth during the imaging of the scene, etc.). The transformation model
used for the level 1B is such that two consecutive scenes along a same data strip match together (i.e.,
there is a perfect registration between the overlapping portions of the two images). The orientation of
the processed image is the same as the one of the raw image (no rotation of the lines is performed).
When  displayed,  the  SPOT  Scene  image  is  a  parallelogram.  The  ancillary  data  (coordinates  of  the
scene centre and of the four corners, location model) allow to locate any point of the image with an
accuracy better than 500m (rms). The distortion of lengths is less than 2 x 10-3 (for a flat terrain).
Suitable  for  doing  geometric  measurements  (distances,  angles  and  superficies),  these  images  are
recommended for people who need a first level of geometric correction. Level 1AP - Photographic product for photogrametric use with analogue devices.
Same detector equalisation as level 1A is performed (compensation from the differences of sensitivities
between  the  elementary  detectors  of  the  CCD  arrays,  using  a  linear  model).  In  addition  a  filter
enhancing high spatial frequencies is applied in order to highlight linear features on the photographic
The  geometric  distortions  are  not  corrected,  with  the  exception  of  a  stretching  of  the  lines  with  the
purpose of coarsely compensating for the panoramic effect (due to non vertical viewing) for a better
comfort  of  the  stereoscopic  viewing.  When  displayed,  the  SPOT  Scene  image  is  a  rectangle.  Same
ancillary data as for level 1A can be delivered on a floppydisk; they allow to locate the image with an
accuracy better than 500m (rms), and can be ingested into stereo plotters for geometric modelisation.
This  product  is  designed  for  analogue  photogrammetry  (only  on  photographic  films).  The  set  of
radiometric  and  geometric  processing  was  designed  specifically  to  ease  the  task  of  stereoscopic
restitution of Spot imagery.93 Level 2A – entry level for cartographic products – projected, no control points
The  same  detector  equalisation  as  for  level  1A  is  performed  (it  consists  in  compensating  for  the
differences of sensitivities between the elementary detectors of the CCD arrays). Absolute calibration
coefficients posted in the ancillary data can be used to convert the pixel counts into irradiances. The
geometric  corrections  (geo-rectification)  use  a  re-sampling  model  which  takes  into  account  the
compensations  for  the  distortions  caused  by  the  imaging  conditions,  as  well  as  the  transformations
needed to put the image into the requested cartographic projection (conformal Lambert, UTM, oblique
equatorial, polar stereographic, polyconic, etc.). This model is computed on the basis of "system data"
known  a  priori  (ephemeris  of  the  spacecraft,  attitude  motion,  etc.),  and  does  not  use  external
information (ground control point, for example). Therefore, the accuracy of the location of any point
within the image is the same as for level 1B, i.e. better than 500m (rms). However, this error which is
mostly  a  translation  bias,  can  be  reduced  by  the  user  with  at  least  one  ground  control  point.  The
resulting location accuracy depends on the parallax error which comes from the differences of altitude
within  the  landscape,  combined  with  the  viewing  angle:  provided  that  the  terrain  is  flat  and  that  the
image  results  from  a  vertical  viewing,  and  after  having  removed  the  translation  bias,  the  order  of
magnitude of the residual location error which has been currently noticed is 50m (rms).
Apart from a translation bias due to the location uncertainty (less than 500m, rms) which can be easily
fixed by a user with at least one ground control point, the image is directly superimposable, with the
accuracy mentioned herabove, to any other geographic information in the same cartographic projection
such as vectors, scanned maps or other satellite images.
Typical users of this product are people wishing to combine geographic information of different types
and  sources  and  to  apply  their  own  image  processing  to  level  2A  images  in  order  to  extract  specific
information; this requires some remote sensing expertise. Level 2B – geocoded product, projected, with ground control points
The  same  detector  equalisation  as  for  level  1A  is  performed  (it  consists  in  compensating  for  the
differences  of  sensitivities  between  the  elementary  detectors  of  the  CCD  arrays).  For  SPOT  Scene
products, absolute calibration coefficients posted in the ancillary data can be used to convert the pixel
counts into irradiances.
The geometric corrections use a re-sampling model which takes into account the compensations for the
distortions  caused  by  the  imaging  conditions,  as  well  as  the  transformations  needed  to  put  the  image
into  the  requested  cartographic  projection  (conformal  Lambert,  UTM,  polar  stereographic,  polyconic,
etc.).  This  is  based  on  a  modelisation  of  the  flight  dynamic  of  the  spacecraft  using  "system  data"
(ephemeris of the satellite, attitude, etc.) plus additional geographic or cartographic data. These extra
data are ground control points whose cartographic or geographic coordinates are measured on a map or
on  the  ground  (GPS  points).  This  entails  a  major  improvement  of  the  location  accuracy  of  any  point
within the image which can range from 10 to 30m, depending on the quality of the maps (provided that
the landscape is flat).
Additional  cartographic  information  is  introduced  as  ancillary  data  (map  projection,  cartographic  co-
ordinates  of  the  first  pixel,  etc.).  When  displayed,  the  SPOT  Scene  image  is  a  parallelogram  having
undergone the rotation needed to match the cartographic reference frame (usually, the image is North
The availability of maps or GPS points is a pre-requisite for performing a level 2B preprocessing.
These products are designed to be used as digital maps. They provide a global, up-to-date, geographic
information to the geographic information community. The products rectified to a precision 2B level
can be used everywhere where the relief distortions are not too severe. Level ORTHO - The ultimate level of preprocessing for the best cartographic
accuracy: correction of the residual parallax errors brought by the relief.
The  same  detector  equalisation  as  for  level  1A  is  performed  (it  consists  in  compensating  for  the
differences  of  sensitivities  between  the  elementary  detectors  of  the  CCD  arrays).  For  SPOT  Scene
products, absolute calibration coefficients posted in the ancillary data can be used to convert the pixel
counts into irradiances.
The  geometric  corrections,  called  "Ortho-rectification",  use  a  re-sampling  model  which  takes  into
account  the  compensations  for  the  distortions  caused  by  the  imaging  conditions,  as  well  as  the
transformations  needed  to  put  the  image  into  the  requested  cartographic  projection  (conformal
Lambert,  UTM,  oblique  equatorial,  polar  stereographic,  polyconic,  etc.).  This  is  based  on  a
modelisation  of  the  flight  dynamic  of  the  spacecraft  using  "system  data"  (ephemeris  of  the  satellite,
attitude,  etc.)  plus  additional  geographic  or  cartographic  data.  These  extra  data  are  ground  control94
points whose cartographic or geographic co-ordinates are measured on a map or on the ground (GPS
points), plus a Digital Elevation Model which is used to fix the parallax error caused by the relief. This
entails a major improvement of the location accuracy of any point within the image which can range
from  10  to  30m,  depending  on  the  quality  of  the  maps,  irrespective  of  the  relief  and  of  the  viewing
angle. Additional cartographic information is introduced as ancillary data (map projection, cartographic
co-ordinates  of  the  first  pixel,  etc.).  When  displayed,  the  "SPOT  Scene"  image  is  a  parallelogram
having undergone the rotation needed to match the cartographic reference frame (usually, the image is
"North oriented").
The  availability  of  maps  (or  global  positionning  system  points)  and  DEM  is  a  pre-requisite  for
performing a level Ortho preprocessing.
They are thus ideal for mapping relief, offering the same accuracy as level 2B. A DEM of the region of
interest must obviously be available to generate level Ortho products.
4.3.4 Parametric Geocoding (based on the PARGE alogrithm)
§     Schläpfer  D.,  Meyer  P.,  and  Itten  K.I.,  1998:  Parametric  Geocoding  of  AVIRIS  Data  Using  a
Ground Control Point Derived Flightpath. Summaries of the Seventh JPL Airborne Earth Science
Workshop, published on the Web, JPL, Pasadena (CA), pp. 367-372
§     Schläpfer D., Meyer P., and Itten K.I.: PARGE: Parametric Geocoding based on GCP-Calibrated
Auxiliary Data,
§     these articles and others are available at -
§     Wiemaker  Rafael,  Registration  of  Airborne  Scanner  Imagery  using  Akima  Local  Quintic
Polynomial Interpolation
The position of scanning airborne systems (e.g. of the AVIRIS instrument) never is as
stable  as  the  behaviour  of  sensors  on  spacebrone  platoforms.  Thus,  geometric
distortions occur due to variations of the flight path as well as of the attitude (given by
roll, pitch and heading angles) of the plane. These distortions can not be corrected by
ground  control  point  based  traditional  georeferencing  procedures  easily,  since  the
movements  can  not  be  approximated  satisfyingly  by  polynomial  transformations  of
the  image,  due  to  the  non-instantaneous  image  formation  process,  unlike  the
photogrammetric practice:
Photogrammetry which deals with digitized photographic immagery can easily model
the  geometric  image  formation  process.  Photographs  are  taken  with  a  short  shutter
time  and  a  center  perspective  optics,  and  only  one  sensor  position  and  attitude  is
needed for the mapping equations. When terrain effects are neglected, photographic
aerial imagery is often geocoded using affine, bilinear and second degree polynomial
transformation functions. A small number of GCPs is sufficient in order to determine
the  necessary  number  of  coefficients;  a  higher  number  of  GCPs  can  increase  the
registration accuracy.
In contrast, multispectral remotely sensed imagery is often recoreded by line scanners
which are mounted on airborne platforms. Thus the process of image formation is not
instantaneous  but  depends  on  the  flight  path  and  the  attitude  of  the  instrument.
Moreover, the mapping itself cannot be described by a lens camera imaging model.
Therefore  the  ortho-rectification  of  scanner  recorded  images  –  especially  aerial
images from low flight altitudes with high spatial resolution – by conventional global
coordinate transforms is not satisfactory.
A linewise calculation has to be performed instead, to consider the behaviour of the
plane, and to allow for local rectification.95
For  an  exact  geometric  rectification  a  variety  of  input  data  is  required.  The  three
categories of input data are:
1.  Navigation data, consisting of location (longitude, latitude, height – from DGPS)
and engineering data (roll, pitch and true heading). This data should be resampled
exactly per line or per pixel of the scanner image.
2.  The  Digital  Elevation  Model  has  to  be  in  the  same  coordinate  system  as  the
airplane data. The resolution has to be based on the image nominal pixel size. The
Digital Elevation Model initiates the final geometry of the geocoded image.
3.  Image general information consists of exact information on FOV (field of view)
and  IFOV  (instantaneous  field  of  view),  scanning  frequency,  starting  time,
coordinates of first nadir point, missing lines, and dimensions of the image.
4.  Ground Control Points are used to determine the uncertain absolute calibration of
the airplane attitude angles, average height and position.
The full reconstruction of the geometry of the scanning process, can also provide the
required  data  for  the  atmospheric  correction  –  the  viewing  angle  per  pixel,  the
absolute  distance  from  the  sensor  to  each  pixel  location,  or  the  relative  air  mass
between sensor and pixel.
The following steps are performed during the main processing algorithm:
•   Calculate the current observation geometry (following figure); the theoretic look
angle vector (L) is calculated between the airplane position and a supposed ‘flat’
DEM, using the instruments FOV and the pixel position information. This vector
is then transformed to the effective look angle (L t ), applying the data from the roll,
pitch and true heading.
•   Find the intersection point on the surface; the intersection procedure used in the
PARGE algorithm calculates a height profile along the footprint of L t  and searches
for a point of equal height L t .96
•   Map  the  image  coordinates;  the  pixel  coordinates  of  the  image  (pixel  and  line
number)  are  written  to  an  array  in  DEM  geometry  at  the  intersection  point
position.  The  result  of  this  procedure  is  a  ‘remapping  array’  which  contains  the
indices  of  the  raw  image  coordinates,  mapped  to  the  correct  poistions  on  the
•   Gap  filling;  triangulation  and  nearest  neighbour  techniques  are  used  to  create  a
spatially continous image (see below).
In  order  not  to  lose  much  information  of  the  raw  image,  the  resolution  of  the  final
Digital  Elevation  Model  (and  image)  has  to  be  taken  slightly  higher  than  of  the
original image data. This results with the creation of gaps, that is, pixels that contain
no information from the scanner. In order to fill these gaps, the method prefered by
the authors, is the triangulation method, as in the following figure:97
The flightpath  normally is provided with a dataset (of the navigation data). If no path
is  available,  a  flightpath  reconstruction  procedure  is  applied  based  on  a  number  of
GCP’s: the xy-position of the plane is determined for each GCP, and an average flight
height  is  derived  from  the  statistics  of  additional  GCPs.  The  following  assumptions
allow such a flightline reconstruction based on a number of GCPs:
•   the flight altitude is constant within the required accuracy
•   the flight velocity is constant within GCP distances, and
•   the flight is more or less straight without any hard turns.
The aircraft position is then calculated using a cubic spline interoplation between the
position points.98
5. Image Processing:
§     Idrisi on-line help
Imaging  Remote  Sensing  data  is  organized  in  a  matrix  (raster).  The  columns  are
usually  termed  as  samples,  and  the  rows  as  lines.  As  an  image  scene  contains
information from several bands, there can be different ways to organize the data. The
data  is  stored  as  a  binary  stream  of  bytes  in  either  Band  Sequential  (BSQ),  Band
Interleaved by Pixel (BIP), or Band Interleaved by Line (BIL) format.
•   BSQ  is  the  simplest  format,  with  each  line  of  data  followed  immediately  by
the  next  line  of  the  same  spectral  band.  BSQ  format  is  optimal  for  spatial
(X,Y) access to any part of a single spectral band.
•   BIP  format  provides  optimal  spectral  processing  performance.  Images  stored
in BIP format have the first pixel for all bands in sequential order, followed by
the second pixel for all bands, followed by the third pixel for all bands, etc.,
interleaved  up  to  the  number  of  pixels.  This  format  provides  optimum
performance for spectral (Z) access of the image data.
•   BIL  format  provides  a  compromise  in  performance  between  spatial  and
spectral  processing  and  is  the  recommended  file  format  for  most  Remote
Sensing processing tasks. Images stored in format have the first line of the first
band followed by the first line of the second band, followed by the first line of
the  third  band,  interleaved  up  to  the  number  of  bands.  Subsequent  lines  for
each band are interleaved in similar fashion.
As an image file contains only pixel values, additional information is needed in order
to display it, and to keep track of it. Depending on the software used, this metadata is
located  either  at  a  separate  file  (usually  an  ASCII  file  with  the  same  name,  using  a
different  extension),  or  in  the  same  file  containing  the  data,  before  the  data.  This
information is known as header, or documentation.
Given  below  is  the  structure  of  an  image  documentation  file,  as  is  in  the  software
Idrisi. They may be broken down into four major groups:
Information about the image as a whole:
•   title: A descriptive name of the file. It is this text that appears at the bottom of
the LIST dialog box when an Image Documentation  file is highlighted.
•   data type: The type of numbers stored in the file. Allowable entries are byte,
integer and real.
•   file type: The format in which the Image file is stored. Allowable entries are
ASCII, Binary and Packed Binary.
•   columns: The number of columns in the image. This is extremely important as
it tells IDRISI for Windows modules how to construct the rectangular image
from the stored values.
•   rows: The number of rows in the image.99
Information about the georeferencing system of the file:
•   ref. system: The name of the geographic referencing system used with the file.
This  may  be  Plane,  Lat/Long,  or  a  specific  referencing  system  defined  by  a
Reference System Parameter file.
•   ref.  units:  The  unit  of  measure  used  in  the  specified  reference  system.
Allowable entries are m, ft, mi, km, deg and radians.
•   unit  dist.:  The  scaling  factor  between  the  given  coordinates  and  actual
measurements on the ground. This will almost always be 1. The unit distance
answers  the  question,  "If  I  move  one  unit  in  the  reference  system  described
here, how far have I moved on the ground, measuring in reference units?"
•   min X: The minimum X coordinate (left edge) of the image.
•   max X: The maximum X coordinate (right edge) of the image.
•   min Y: The minimum Y coordinate (bottom edge) of the image.
•   max Y: The maximum Y coordinate (top edge) of the image.
•   pos'n error: A measure of the accuracy of the positions in the image. This field
can  be  used  to  record  the  RMS  (Root  Mean  Square)  error  of  locational
positions  in  the  reference  system.  At  present,  this  field  is  not  analytical,  but
rather is for informational purposes only.
•   resolution:  The  inherent  resolution  of  the  image.  In  most  cases,  this  should
correspond to the result of dividing the range of reference coordinates in X by
the number of columns in the image. However, there are some rare instances
where it might differ from this result. An example is the case of LANDSAT
Band 6 (Thermal) imagery. The resolution of those data is actually 120 meters,
and would be recorded as such in this field. However, the data is distributed in
an apparent 30 meter format to make them physically match the dimensions of
the  other  bands  of  the  scene.  The  resolution  field  is  a  way  of  correctly
indicating the underlying resolution of these data.
Information about the values stored in the file:
•   min value: The minimum value in the image.
•   max value: The maximum value in the image.
•   value units: The unit of measure of the values in the image. It is suggested that
the  term  classes  be  used  for  all  qualitative  data  sets,  and  that  whenever
standard linear units are appropriate, that the same abbreviations that are used
for reference units should also be used (m, ft, mi, km, deg, rad).
•   value  error:  This  field  is  very  important  and  should  be  filled  out  whenever
possible. It records the error in the data values that appear in image cells. For
qualitative   data,   this   should   be   recorded   as   a   proportional   error.   For
quantitative data, the value here should be an RMS error figure. For example,
for a DEM, an RMS error of 3 would indicate 68% of all values will be within
± 3 meters of the stated elevation, that approximately 95% will be within ± 6
meters, and so on. This field is analytical for some modules (e.g., PCLASS)
and it is intended that it will become more so in the future.100
•   flag  value:  Any  value  in  the  image  that  is  not  a  data  value,  but  rather  has  a
special meaning. If there is no flag value, this entry should remain blank.
•   flag def'n: Definition of the above flag value. The most common data flags are
those  used  to  indicate  background  cells  and  missing  data  cells.  This  field  is
analytical for some modules (e.g., SURFACE) and will become more so in the
future.   The   key   words   background   and   missing   data   are   specifically
recognized  by  some  modules.  Other  terms  are  only  informational  in  this
version. If there is no flag value, this entry should remain blank.
•   legend  cats:  The  number  of  legend  categories  present.  Legend  entries  are
optional.  If  there  is  no  legend,  0  should  be  entered  in  this  field.  If  legend
categories do exist, there will be as many lines following as there are legend
categories. If no legend categories exist, these lines are not necessary. (See the
example below.)
Other information about the file:
The following four entries are optional, and any number of each may be entered at the
end  of  the  file,  so  long  as  each  has  the  correct  term  in  the  14  character  descriptive
field  to  the  left.  These  are  all  text  fields  and  are  included  to  facilitate  complete
documentation of the Image file. At present, these last fields are for information only
and are not read by IDRISI for Windows modules (although some modules will write
•   comment: Any additional information about the data may be recorded here.
•   lineage: Description of the history by which the values were recorded/derived.
•   completeness:  The  degree  to  which  the  values  describe  the  subject  matter
•   consistency: The logical consistency of the file.
Note  that  the  lineage,  consistency  and  completeness  fields  are  intended  to  meet  the
recommendations  of  the  U.S.  National  Committee  for  Digital  Cartographic  Data
Standards (NCDCDS). Along with the pos'n error and value error fields, they provide
a means of adhering to the proposed standards for reporting digital cartographic data
quality.  For  further  information  about  the  NCDCDS  standard,  refer  to  the  January
1988 issue of The American Cartographer, 15(1).
Other  software  header  files  can  have  more  (e.g.  name  of  the  sensor,  storage  format
(BSQ,BIL,BIP), etc) or less metadata information.
Further discussion of this subject is given at the last chapter of  this text, under the
title of metadata.
5.2 Image enhancment:
§     Ilwis documentation:
§     Idrisi on line help
Image  enhancement  deals  with  the  procedures  of  making  a  raw  image  more
interpretable for a particular application. In this section, commonly used enhancement
techniques are described, which improve the visual impact of the raw remotely sensed
data on the human eye.101
Image   enhancement   techniques   can   be   classified   in   many   ways.   Contrast
enhancement,  also  called  global  enhancement,  transforms  the  raw  data  using  the
statistics  computed  over  the  whole  data  set.  Examples  are:  linear  contrast  stretch,
histogram equalized stretch and piece-wise contrast stretch. Contrary to this, spatial or
local enhancement takes into consideration local conditions only and these can vary
considerably  over  an  image.  Examples  are  image  smoothing  and  sharpening  (These
will be dealt here under the section on filters).
5.2.1 Histogram, stretching, colour palettes Contrast enhancement
The objective of this section is to understand the concept of contrast enhancement and
to be able to apply commonly used contrast enhancement techniques to improve the
The sensitivity of the on-board sensors of satellites, have been designed in such a way
that  they  record  a  wide  range  of  brightness  characteristics,  under  a  wide  range  of
utilizes the brightness range of the detectors. The goal of contrast enhancement is to
improve the visual interpretability of an image, by increasing the apparent distinction
distinguishing and interpreting spatial features in an image, the eye is rather poor at
discriminating  the  subtle  differences  in  reflectance  that  characterize  such  features.
them readily observable.
Contrast  stretch  is  also  used  to  minimize  the  effect  of  haze.  Scattered  light  that
objects  at  the  earth  surface,  is  called  haze  or  path  radiance.  Haze  results  in  overall
higher DN values and this additive effect results in a reduction of the contrast in an
blue, and lowest in the infra red range of the electromagnetic spectrum.
Techniques used for a contrast enhancement are: the linear stretching technique and
types the piece-wise linear contrast stretch can be applied.
A  computer  monitor  on  which  the  satellite  imagery  is  displayed  is  capable  of
satellite images, as their digital numbers also vary within the range of 0 to 255. To
produce  an  image  of  optimal  contrast,  it  is  important  to  utilize  the  full  brightness
In  order  to  decide  which  type  of  stretch  to  use,  and  to  see  the  effect  of  a  contrast
stretch,  in  addition  to  just  watching  the  image,  it  is  very  usefull  to  observe  the
distribution  where  the  widths  of  contiguous  vertical  bars  are  proportional  to  class
widths  of  the  variable,  and  the  heights  of  the  bars  are  proportional  to  the  class Linear Stretch
linear stretch is the simplest contrast enhancement. A DN value in the low end of
the  original  histogram  is  assigned  to  extreme  black,  and  a  value  at  the  high  end  is
assigned to extreme white. The remaining pixel values are distributed linearly between these
extremes. One drawback of the linear stretch, is that it assigns as many display levels to the102
rarely  occurring  DN  values  as  it  does  to  the  frequently  occurring  values.  However,  linear
contrast stretch, putting (min, max) at (0,255) in most cases still produces a rather dull image.
Even though all gray shades of the display are utilized, the bulk of the pixels are displayed in
mid  gray.  This  is  caused  by  the  more  or  less  normal  distribution,  within  the  minimum  and
maximum values in the tail of the distribution. For this reason it is common to cut off the tails
of the distribution at the lower and upper range (usually be defining the size of the tails by their
percentage from the total). Histogram equalization
The histogram equalization technique is a non-linear stretch. In this method, the DN values are
redistributed  on  the  basis  of  their  frequency.  More  different  gray  tones  are  assigned  to  the
frequently occurring DN values of the histogram.
A  histogram  of  the  resulting  image  will  thus  appear  flat  --  hence  the  name.  In  theory  (i.e.,
Information Theory), this leads to an image that carries the maximum amount of information
for any given number of classes. However, this does not imply that the resulting image is more
meaningful. In fact, since the nature of the histogram has been altered, you will have lost one of
the  more  informative  characteristics  of  the  image.  However,  in  cases  where  it  is  difficult  to
develop a good visual display, this will usually provide an excellent visual result.  Histogram
equalization is not available for use with integer or real data. Piece-wise linear stretch
The piece-wise linear contrast stretch is very similar to the linear contrast stretch, but the linear
interpolation of the output values is applied between user defined DN values. This method is
useful to enhance only a certain cover type, for example water. The data values for this feature
are in the range of 5 to 18, and in order to be able to discriminate as much as possible, it is wise
to use all available gray levels for this feature only. In this way detailed differences within the
feature of interest appear, where as the remaining features are assigned to a single gray tone.103
Following is an example of 3 types of stretching applied to band 1 of a SPOT image that was
taken in 1992 (from the exercise files of the Idrisi software). Given are the images and their
corresponding histograms.

 104 RGB false colour composit
§     Ilwis documentation:
In this section, the objective is to understand the concept of color composites and to
be able to create different color composites.
The spectral information stored in the separate bands can be integrated by combining
them into a color composite. Many combinations of bands are possible. The spectral
information  is  combined  by  displaying  each  individual  band  in  one  of  the  three
primary colors: Red, Green and Blue.
Every combination of bands used to create a color composite image is called a  False
Color Composite (FCC). In one of the more common FCC, the red color is assigned
to the near-infrared band, the green color to the red visible band and the blue color to
the green visible band. The green vegetation will appear reddish, the water bluish and
the  (bare)  soil  in  shades  of  brown  and  gray.  For  SPOT  multi-spectral  imagery,  the
bands 1, 2 and 3 are displayed respectively in blue, green and red (for TM imagery
this FCC will be composed of the bands 2, 3 and 4). Another combination used very
often  for  TM  imagery  is  the  one  that  displays  in  red,  green  and  blue  the  respective
bands 5, 4 and 3. Other band combinations are also possible. Some combinations give
a color output that resembles natural colors: water is displayed as blue, (bare) soil as
red  and  vegetation  as  green.  Hence  this  combination  leads  to  a  so-called  Pseudo
Natural Color composite.
Bands  of  different  images  (from  different  imaging  systems  or  different  dates),  or
layers  created  by  band  rationing  or  principal  component  analysis,  can  also  be
combined  using  the  color  composite  technique.  An  example  could  be  the  multi-
temporal combination of vegetation indices for different dates, or the combination of
2 SPOT XS bands with a SPOT PAN band, (giving in one color image the spectral
information  of  the  XS  bands  combined  with  the  higher  spatial  resolution  of  the
panchromatic band). Colour Definition Models, fusion of TM and SPOT
§     Paint Shop Pro 5 on line help Colour Definition Models
There are several methods for defining the projected colors that appear on a computer
monitor. The following table outlines the Color dialog box red, green, and blue (RGB)
and  hue,  saturation,  and  lightness  (HSL)  settings  for  the  standard  white  light  color
Color RGB Settings HSL Settings
Red     Green  Blue    Hue     Sat       Light
Red 255      0          0          0          240      120
Orange 255      128      0          20        240      120
Yellow 255      255      0          40        240      120
Green 0          255      0          80        240      120
Azure 0          255      255      120      240      120
Indigo 0          0          255      160      240      120
Violet 255      0          255      200      240      120 Red, Green, and Blue (RGB)
The  most  popular  method  for  defining  a  projected  color  is  as  a  combination  of  red,
green, and blue, with values ranging from 0 to 255. For example, pure red is defined105
by red = 255 (100%), green =0 ( 0%), and blue =0 ( 0%). Pure black has red, green,
and blue values of 0%, and pure white has red, green, and blue values of 100%.  The
RGB model, one of the additive color models, is used on computer monitors. Hue, Saturation, and Lightness (HSL)
A  projected  color  can  be  defined  by  the  three  components  of  hue,  saturation,  and
Hue describes the color's shade or tint. It is measured on a circular spectrum running
from red to green to blue and returning to red.
Saturation describes the hue's purity. A color with 100% saturation is bright and vivid,
and a color with 0% saturation is a shade of grey.
Lightness  (sometimes  termed  value)  describes  the  color's  brightness.  A  color  with
100%  lightness  is  always  pure  white,  and  a  color  with  0%  lightness  is  always  pure
black. CMYK Model
The CMYK model, a subtractive color model, is based on light being absorbed and
reflected  by  paint  and  ink.  This  model  is  often  used  when  printing.  The  primary
colors, cyan, magenta, and yellow, are mixed to produce the other colors. When all
three are combined, they produce black. Because impurities in the ink make it difficult
to produce a true black, a fourth color, black (the K), is added when printing. Fusion by HSV Transformation
Bands from different sensors with different spatial resolutions can be fused together,
in  order  to  take  advantage  of  the  higher  spatial  resolution  of  one  of  them,  and  the
higher spectral resolution of the other.
The  most  widely  applied  fusion  procedure  is  the  merging  of  panchromatic  SPOT
imagery (10 m) with three-color SPOT imagery (20 m) or multispectral LANDSAT
TM  imagery  (30  m).  The  simplest,  most  wide-spread  and  probably  most  intuitive
technique works as follows:
1.  Take three spectral bands from the multispectral imagery;
2.  Register the low resolution color image to the high resolution panchromatic image
(i.e.  essentially  to  magnify  the  color  image  to  the  same  pixel  size  as  the
panchromatic image);
3.  Transform  the  magnified  color  image  from  an  RGB  color  system  into  the  HSV
color system (Hue, Saturation, Value);
4.  Replace the ``Value'' image by the high resolution panchromatic image;
5.  Transform back into the RGB color system.
This technique is known to work well for moderate resolution ratios (such as 1:3 for
SPOT + LANDSAT TM). The results are still helpful but less reliable for resolution
ratios  such  as  1:20,  e.g.  for  fusion  of  SPOT  color  images  with  panchromatic  aerial
It has to be noted, however, that fusion by HSV transformation can be applied only to
multispectral imagery consisting of three bands, since the image has to be coded as an
RGB image before fusion can take place.
Another  method  for  fusing  images,  the  fusion  by  relative  contribution,  will  not  be
discussed here. Spatial filters – noise reduction (low-pass), edge enhancment
§     Cracknell A.P. and Hayes L.W.B. (1993), Introduction to Remote Sensing, pp. 170-173
§     Jain Anil K. (1989), Fundamentals of Digital Image Processing, pp. 53-54, 244-252, 347-357
§     PCI (1996), Using PCI software vol. I, pp. 217-226
§     Schott John R. (1997), Remote Sensing – The Image Chain Approach, pp. 241-248, 347-353
The  operation  called  filtering  is  an  operation  that  changes  the  value  of  a  pixel
according  to  the  “gray”  values  of  its  surrounding  pixels  (it  is  a  so  called  “focal
operation”).  This  might  be  called  a  spatial  stretching  operation,  unlike  a  contrast
stretch that is operating on the single pixel.
When  filtering,  we  are  actually  analysing  the  variance  of  the  pixel  values  in  space,
that is, we are dealing with the spatial frequency of an image. Any image can be seen
as made of two basic components of information: a high frequency component (high
pass)  and  a  low  frequency  component  (low  pass),  their  sum  makes  up  the  original
image.  High  frequencies  describe  sudden\big  changes  in  pixel  values  along  a  short
distance (linear features: geologic faults, rivers, roads, pipelines, etc or the borders of
area features: buildings, etc). Low frequencies describe gradual\small changes in pixel
values in space (such as: water bodies, agricultural fields, natural vegetation, etc).
The  spatial  operation  of  filtering  receives  its  name  due  to  the  fact  that  it  lets  only
those  frequencies  in  the  image  that  we  want  to  remain.  High-pass  filters  will
emphasize spatial change (borders), while low-pass filters will generate an image that
will  appear  smooth  or  blurred  relative  to  the  original  (such  a  filter  can  be  used  to
eliminate noises). The basic principles of a filter
Although there are many kinds of filters that can be applied on an image, the basic
principles of their operation are similar:
•   The window size of the filter is to be defined, usually having an odd number of
columns and rows (the number of the rows equals that of the columns).
•   A weight is assigned to each pixel in the window. The pixel values in the window,
with the window, are termed “kernel”.
•   The  filtered  image  is  obtained  by  moving  the  window  we  have  defined,  and
multiplying each weight in that window, with the value of the corresponding pixel
(concerning its location in the window) in the original image. The sumproduct (the
sum of the multiplications) is the new value of the pixel in the new image.
Moving the window over the whole image, this operation is preformed on a pixel by
pixel  basis.  Mathematically,  this  operation  is  known  as  “convolution”.  In  order  that
pixels along the perimeter of an image can be a center for the window, the perimeter
pixels  are  duplicated  (temporarily)  during  the  filter  operation,  as  described  in  the
following figure:107 Low pass filters
Low pass filters remove high frequency features from an image, by diminishing the
deviations  from  the  central  parameter.  The  bigger  the  moving  window  is,  the  more
blurred  will  be  the  output  image;  sometimes  it  would  be  necessary  to  perform  a
contrast stretch after that filter, in order to use the full dynamic range of gray values.
These filters will be used mainly to remove noises from an image. Noises in an image
can  be  classified  in  two  according  their  structure:  random  noise,  and  systematic  or
repetitive noise. Systematic noise is usually due to problems in the sensor. Random
noise might occur (for example) on a camera film, when some points on the film itself
were damaged.
Such filters can be used also to lower the variability between pixels that belong to the
same category, and thus might help in performing a better classification.
The low-pass filters of MEAN and GAUSSIAN are used to smooth an image. In the
MEAN filter all pixels have an equal weight, while in the GAUSSIAN, the far a pixel
is  located  from  the  central  pixel,  its  weight  is  smaller  (according  to  gaussian  [bell]
distribution);  a  filter  where  the  weight  of  a  pixel  depends  on  its  distance  from  the
central  pixel  is  termed  “distance  weighted”  –  these  filters  smooth  more  gently  than
“equal weighted” filters.
These filters are also applicable in GIS, for example:
•   Smoothing a DEM created from contours.
•   Computing density from point data.
In the MEDIAN and MODE filters there are no kernel values, and the value of the
new pixel in the filtered output image is just the value of the median or mode of the
pixels  in  the  moving  window.  The  MEDIAN  filter  is  good  for  removing  random
noise. The MODE filter is good for filling in empty areas between polygons after a
vector-to-raster  conversion,  or  after  performing  classification  in  order  to  remove
1/9   1/9   1/9
1/9          1/9          1/9
1/9          1/9          1/9
11/121 11/121
7/121 7/121
7/121 7/121
3/121 3/121
3/121 2/121
2/121 2/121     1/121
1/121 1/121
isolated pixels that belong to a different category than their neighbors. A MODE filter
does not create a new value, in the sense that, for example, the data type remains the
same.  If  the  values  of  the  input  image  are  integers,  the  same  will  be  in  the  output
image  (notice  the  first  figure  presented  in  this  section,  where  applying  the  MEAN
filter changed the values from integer to real).
An example of applying filters will be given using a SONAR image in which we can
see a pipeline. The SONAR image is quite noisy (similar to speckles seen on a Side
Looking RADAR).
The image was taken from the following site:
In  the  following  figure  the  different  effect  of  four  different  low-pass  filters  on  the
image can be seen, comparing them on a vertical profile line.109
Notice the following:
•   The  pipeline’s  location  on  the  profile  is  above  the  “a”  in  the  X  title  “Vertical
•   The  high-pass  features  reduced  to  the  maximum  with  the  MEAN  filter,  leaving
only low-pass features.
•   The  GAUSSIAN  filter  is  following  the  original  profile  better  than  the  MEAN,
which might create a negative spike where it is actually positive.
•   The MEDIAN filter leaving many small spikes.
•   The MODE filter being the most unsuitable here, not removing much of the noise,
and even creating grouped spikes.
For  visual  comparison  are  given  the  images  produced  by  the  MEAN  and  MODE
MEAN filtered image MODE filtered image
In  low-pass  filters  the  sum  of  the  kernel  values  equals  1  (in  some  software,  it  is
possible  to  define  the  kernel  values  in  integer  numbers  –  thus  avoiding  the  need  to
calculate the exact numbers - and then define that the values should be normalized).
Another characteristic of low-pass filters is that they maintain the units of the variable
being filtered (unlike high-pass filters, as will be seen).110 High pass filters
High  pass  filters  emphasize  sudden  changes  in  the  values  of  pixels  over  distance.
Those  abrupt  differences  stand  for  high  spatial  frequencies,  termed  “edges”.  In
images,  edges  will  be  found  in  point  features  (random  noise),  linear  features  (road,
river) or borders between objects (the border between a field and a forest).
Generally, a high-pass filter will emphasize objects smaller than half the size of the
window being used, and blur objects larger than half the window size.
Before  examining  some  high-pass  filters,  it  would  be  fruitful  to  know  the  effect
known as “Mach bands”.
The  spatial  interaction  of  the  reflected  light  from  an  object  with  its  environment,
creates  a  phenomena  called  the  effect  of  Mach  bands.  This  effect  demonstrates  that
the brightness of an object, is not a simple function of its luminance.
The gray bands in the image have a uniform reflectance, but the apparent brightness is
not  uniform.  On  the  edge  between  two  gray  bands,  we  can  see  differences  in  the
brightness  value.  The  thin  line  in  the  chart  demonstrates  that  effect,  with  the
overshoots and undershoots. Some of the high-pass filters are actually creating such a
phenomena,  creating  overshoots  and  undershoots  of  pixel  values  on  the  edges  of
different objects.
A graphical example of creating a high-pass filter is given in the following figure:111
The procedure described in the figure is the most simple high-pass filter. It is based on
the above mentioned idea that each image is built of two components, one of a low
frequency, the other of a high frequency. By subtracting the low frequency component
from  the  image,  we  receive  the  high  frequency.  This  filter  is  called  HIGH  PASS
High-pass filters examine the difference between adjacent pixels. As these differences
are usually small, a contrast stretch should be performed on the output filtered image,
in  order  to  see  better  the  details.  These  filters  can  be  used  in  algorithms  for  pattern
recognition,  when  the  frequency  of  an  “edge”  in  unit  of  area  can  indicate  certain
objects.  The  output  filtered  image  can  be  also  used  as  another  band  helping  in  the
classification procedure.
One    of    the    most    known    high-pass    filters    is    the    LAPLACIAN    EDGE
ENHANCMENT. Its meaning can be thus understood: We subtract the image from a
blurred version of itself created from the averaging of the four nearest neighbors. This
enhances  edges  and  isolated  pixels  with  extreme  values.  However,  in  the  classic
version of this filter, very bright values are obtained in dark edges, and dark values in
bright  edges.  This  is  visually  confusing,  therefor  we  can  use  an  alternative,  the
Modified LaPlacian. There are many other versions of this filter. One of them is given
below too, on the right.
There are two basic kinds for edge detection operators:
•   Gradient operators, working in two orthogonal directions (horizontal and vertical).
•   Compass operators, working in a user defined direction.
The basic idea, is to calculate the slope (that is, the difference in pixel values over a
distance) of an image, and then to define a certain cutting value above which a pixel is
considered as representing an edge point or an edge line.
There   are   many   edge   detection   filters,   one   of   them   is   the   SOBEL   EDGE
DETECTOR, that is working in the following way:
New value = sqrt ( X 2  + Y 2  ) where
X = the resulting image from applying the kernel Kx (below) to the input image
Y = the resulting image from applying the kernel Ky (below) to the input image
Kx = Ky =
0      1      0
1        -4       1
0      1      0
0      -1      0
-1       4        -1
0        -1       0
-1/9  -1/9  -1/9
-1/9  17/9  -1/9
-1/9  -1/9  -1/9
1      0      -1
-2       0      2
-1       0      1
1      2      1
0      0      0
-1        -2        -1112
The  Sobel  edge  detector  is  a  gradient  operator.  Two  high-pass  filters  are  being
calculated,  one  in  the  horizontal  direction,  the  other  in  the  vertical  (their  respective
kernel  values  are  rotated  by  90  degrees  the  one  to  the  other).  The  output  images
resulting are raised by the power of 2, and then we calculate the square root of their
sum. Other gradient operators work basically in a similar way, only having different
kernels (the weights given to the pixels, and the window size used).
Compass operators measure the slope in the requested direction. The resulting slope
images are again used to define a “cutting” value, above which pixels are considered
as  edges.  Below  are  given  the  3*3  kernels  of  the  8  principal  directions;  in  order  to
calculate  directional  edge  enhancement  in  other  angles,  the  moving  window  size
should be enlarged.
An  example  of  using  a  directional  edge  detector  filter  will  be  given,  with  the  same
SONAR  image  of  the  pipeline,  with  the  chosen  direction  of  North  and  East  (it  was
preformed  on  the  MEAN  filtered  image,  to  achieve  better  results).  Notice  the  easy
detection of the pipeline in the North filter.
North edge enhancement East edge enhancement
1      1      1
1        -2        -1
1        -1        -1
1      1      1
1        -2       1
-1        -1        -1
1      1      1
-1        -2       1
-1        -1       1
-1      1      1
1        -2        -1
1      1        -1
1      1      -1
-1        -2       1
-1       1      1
-1     -1      1
1        -2        -1
1      1      1
-1     -1     -1
1        -2       1
1      1      1
1      -1     -1
-1        -2       1
1      1      1
In high-pass filters, the sum of the kernel values is usually one, but can be any other
number  too.  An  important  thing  to  remember  is  that  the  measurement  units  of  the
variable (image) being filtered are not maintained (as in contrast stretching), thus, the
output image is being used to highlight the phenomena we are interested in, not more.
5.3 Multi-band operations
§     Ilwis documentation:
To enhance or extract features from satellite images which cannot be clearly detected
in  a  single  band,  you  can  use  the  spectral  information  of  the  object  recorded  in
multiple  bands.  These  images  may  be  separate  spectral  bands  from  a  single  multi
spectral  data  set,  or  they  may  be  individual  bands  from  data  sets  that  have  been
recorded  at  different  dates  or  using  different  sensors.  The  operations  of  addition,
subtraction, multiplication and division, are performed on two or more co-registered
images   (see   previous   exercise   on   image   to   image   registration)   of   the   same
geographical  area.  This  section  deals  with  multi-band  operations.  The  following
operations will be treated:
- The use of ratio images to reduce topographic effects.
- Vegetation indexes, some of which are more complex than ratio's only.
- Multi-band statistics.
- Principal components analysis.
- Image algebra, and;
- Image fusion.
5.3.1 Image ratios: Brightness variations
When  a  satellite  passes  over  an  area  with  relief,  it  records  both  shaded  and  sunlit
areas. These variations in scene illumination conditions are illustrated in the figure. A
red silt stone bed shows outcrops on both the sunlit and the shadowed side of a ridge.
The observed DNs are substantially lower on the shaded side compared to the sunlit
areas. This makes it difficult to follow the silt stone bed around the ridge.
In the individual Landsat-TM bands 3 and 4, the DNs of the silt stone are lower in the
shaded  than  in  the  sunlit  areas.  However,  the  ratio  values  are  nearly  identical,
irrespective of illumination conditions. Hence, a ratioed image of the scene effectively
compensates for the brightness variation, caused by the differences in the topography
and emphasizes by the color content of the data.
Table: Differences in DN values of selected bands and the ratio values
TM Band 3       TM Band 4       Ratio: Band3/Band4
Sunlit slope          94  42  2.24
Shaded slope       76  34  2.23114
5.3.2 Normalized Difference Vegetation Index
Ratio images are often useful for discriminating subtle differences in spectral variations, in a
scene  that  is  masked  by  brightness  variations.  Different  band  ratios  are  possible  given  the
number of spectral bands of the satellite image. The utility of any given spectral ratio, depends
upon  the  particular  reflectance  characteristics  of  the  features  involved  and  the  application  at
hand. For example a near-infrared / red ratio image might be useful for differentiating between
areas of stressed and non stressed vegetation.
Various mathematical combinations of satellite bands, have been found to be sensitive
indicators   of   the   presence   and   condition   of   green   vegetation.   These   band
combinations  are  thus  referred  to  as  vegetation  indices.  Two  such  indices  are  the
simple vegetation index (VI) and the normalized difference vegetation index (NDVI).
NDVI = (NIR – Red) / (NIR + Red)
Both are based on the reflectance properties of vegetated areas as compared to clouds,
water and snow on the one hand, and rocks and bare soil on the other. Vegetated areas
have a relatively high reflection in the near-infrared and a low reflection in the visible
range of the spectrum. Clouds, water and snow have larger visual than near-infrared
reflectance. Rock and bare soil have similar reflectance in both spectral regions. The
effect of calculating VI or the NDVI is clearly demonstrated in next table.
Table: Reflectance versus ratio values
TM Band 3     TM Band 4     VI        NDVI
Green vegetation        21  142  121      0.74
Water  21  12  -9        -0.27
Bare soil  123  125  2          0.01
It is clearly shown that the discrimination between the 3 land cover types is greatly
enhanced by the creation of a vegetation index. Green vegetation yields high values
for the index. In contrast, water yield negative values and bare soil gives indices near
zero. The NDVI, as a normalized index, is preferred over the VI because the NDVI is
also compensating changes in illumination conditions, surface slopes and aspect.
5.3.3 Principal components analysis
Another  method  (which  is  not  spatial  and  applied  in  many  fields),  called  principal
components analysis (PCA), can be applied to compact the redundant data into fewer
layers. Principal component analysis can be used to transform a set of image bands, as
that  the  new  layers  (also  called  components)  are  not  correlated  with  one  another.
These  new  components  are  a  linear  combination  of  the  original  bands.  Because  of
this, each component carries new information. The components are ordered in terms
of the amount of variance explained, the first two or three components will carry most
of  the  real  information  of  the  original  data  set,  while  the  later  components  describe
only the minor variations (sometimes only noise). Therefore, only by keeping the first
few  components  most  of  the  information  is  kept.  These  components  can  be  used  to
generate  an  RGB  color  composite,  in  which  component  1  is  displayed  in  red,
component  2  and  3  in  green  and  blue  respectively.  Such  an  image  contains  more
information than any combination of the three original spectral bands.115
To perform PCA, the axis of the spectral space are rotated, the new axis are parallel to
the axis of the ellipse (see figure). The length and the direction of the widest transect
of  the  ellipse  are  calculated.  The  transect  which  corresponds  to  the  major  (longest)
axis of the ellipse, is called the first principal component of the data. The direction of
the first principal component is the first eigenvector, and the variance is given by the
first  eigenvalue.  A  new  axis  of  the  spectral  space  is  defined  by  the  first  principal
omponent.  The  points  in  the  scatter  plot  are  now  given  new  coordinates,  which
correspond to this new axis. Since in spectral space, the coordinates of the points are
the  pixel  values,  new  pixel  values  are  derived  and  stored  in  the  newly  created  first
principal component.
The second principal component is the widest transect of the ellipse that is orthogonal
(perpendicular)  to  the  first  principal  component.  As  such,  PC2  describes  the  largest
amount  of  variance  that  has  not  yet  been  described  by  PC1.  In  a  two  dimensional
space, PC2 corresponds to the minor axis of the ellipse. In n-dimensions there are n
principal  components  and  each  new  component  is  consisting  of  the  widest  transect
which is orthogonal to the previous components.
The next figure shows graphically the result of a PC transform in which the data are
presented without correlation. Through this type of image transform the relationship
with  raw  image  data  is  lost.  The  basis  is  the  covariance  matrix  from  which  the
eigenvectors and eigenvalues are mathematically derived. It should be noted that the
covariance  values  computed  are  strongly  depending  on  the  actual  data  set  or  subset
The  following  figure  exemplifies  the  (possible)  information  content  of  the  new
components, and the ability to filter out some of the noise by performing an inverse
PCA transformation, not including the new component containing only noise.
The output of a PCA contains the following tables:
•   Variance/covariance matrix between the original bands (variables).
•   Correlation matrix between the original bands.
•   Component’s eigenvalues – the amount of variance explained by each of the new
bands (%variance = eigenvalue / S[eigenvalue]).
•   Eigenvectors – the parameters for the linear combination of the new bands for an
inverse transformation, back to the original bands.
•   Component’s factor loadings (factor pattern matrix) – factors with a high loading
parameter with an original band, have a high correlation with it.117
5.3.4 Image classification
Measured reflection values in an image depend on the local characteristics of the earth
surface;  in  other  words  there  is  a  relationship  between  land  cover  and  measured
eflection values. In order to extract information from the image data, this relationship
must   be   found.   The   process   to   find   the   relationship   is   called   classification.
Classification can be done using a single band, in a process called density slicing, or
using many bands (multi-spectral classification). Density slicing
In theory, it is possible to base a classification on a single spectral band of a remote
sensing  image,  by  means  of  single  band  classification  or  density  slicing.  Density
slicing  is  a  technique,  whereby  the  DNs  distributed  along  the  horizontal  axis  of  an
image  histogram,  are  divided  into  a  series  of  user-specified  intervals  or  slices.  The
number of slices and the boundaries between the slices depend on the different land
covers in the area. All the DNs falling within a given interval in the input image are
then displayed using a single class name in the output map.
Firstly, the ranges of the DN values representing the different land cover types have to
be determined. Secondly, a domain Group with the slice boundaries, names and codes
has to be created.
By  a  visual  interpretation  of  the  spectral  reflectance  of  the  cover  classes  in  the  image,
differences  are  observed  with  regard  to  their  gray  tones.  Density  slicing  will  only  give
reasonable  results,  if  the  DN  values  of  the  cover  classes  are  not  overlapping  each  other,  as
indicated in the following figure:
In the figure, a clear distinction is possible between cover type A and cover types B/C
(slice with minimum number at position DN 1), as the distribution of the pixels over
the  digital  range  is  different  from  cover  type  B  and  C.  If  the  distributions  of  cover
types B and C are studied, an overlap can be observed. To discriminate between B and
C several options are possible, but all will result in an improper slice assignment. To
classify cover class B, the slice DN 1 to DN 2 will include the lower DN values of C
and the higher DN values of B are excluded. To include all the pixels belonging to B,
the slice assignment is from DN 1 to DN 4. This slice assignment is including even
more pixels that belong to cover class C. When trying to classify C, the same problem
occurs. Multi-spectral image classification118
Multi  spectral  image  classification  is  used  to  extract  thematic  information  from
satellite images in a semi-automatic way. Different methods for image classification
exist; some of them are based on the theory about probabilities. Looking at a certain
image  pixel  in  M  bands  simultaneously,  M  values  are  observed  at  the  same  time.
Using multi-spectral SPOT images, where M=3, three reflection values per pixel are
given. For instance, (34, 25, 117) in one pixel, in another (34,24,119) and in a third
(11, 77, 51). These values found for 1 pixel in several bands are called feature vectors.
It can be recognized that the first two sets of values are quite similar and that the third
is different from the other two. The first two probably belong to the same (land cover)
class and the third belongs to another one.
In  classification  jargon  it  is  common  to  call  the  three  bands  “features”.  The  term
features instead of bands is used because it is very usual to apply transformations to
the  image,  prior  to  classification.  They  are  called  “feature  transformations”,  their
results “derived features”. Examples are: Principal components, HIS transformations
In one pixel, the values in the (three) features can be regarded as components of a 3-
dimensional  vector,  the  feature  vector.  Such  a  vector  can  be  plotted  in  a  3-
dimensional  space,  called  feature  space.  Pixels  belonging  to  the  same  (land  cover)
lass and having similar characteristics, end up near to each other in the feature space,
regardless  of  how  far  they  are  from  each  other  in  the  terrain  and  in  the  image.  All
pixels belonging to a certain class will (hopefully) form a cluster in the feature space.
Moreover,  it  is  hoped  that  other  pixels,  belonging  to  other  classes,  fall  outside  this
cluster (but in other clusters, belonging to those other classes).
A  large  number  of  classification  methods  exist.  To  make  some  order,  the  first
distinction is between unsupervised and supervised classification. For satellite image
applications, the second is generally considered more important. In order to make the
classifier  work  with  thematic  (instead  of  spectral)  classes,  some  “knowledge”  about
the relationship between classes and feature vectors must be given.
Theoretically, this could be done from a database in which the relationships between
(thematic)  classes  and  feature  vectors  is  stored.  It  is  tempting  to  assume  that  in  the
past,  enough  images  of  each  kind  of  sensor  have  been  analyzed,  as  to  know  the
spectral characteristics of all relevant classes. This would mean, for example, that a
pixel with feature vector (44, 32, 81) in a multi spectral SPOT image always means:
Grass,  whereas  (12,  56,  49)  is  always  a  forest  pixel.  Unfortunately,  the  observed
feature vectors in a particular image are influenced by a large amount of other factors
than  land  cover,  such  as:  Atmospheric  conditions,  sun  angle  (as  function  of
latitude/time of day/date and as a function of terrain relief), soil types, soil humidity,
growing stage (vegetation), wind (affecting orientation of leafs), etc. The problems we
meet  when  trying  to  take  all  these  influences  into  account  vary  from  quite  easy  to
practically  impossible;  at  least,  vast  amounts  of  additional  data  (DEM’s,  soil  maps,
etc.) would be required.119 Supervised Classification Sampling
Therefore, classifications methods are much more widely used, where the process is
divided  into  two  phases:  a  training  phase,  where  the  user  “trains”  the  computer,  by
assigning for a limited number of pixels to what classes they belong in this particular
image,  followed  by  the  decision  making  phase,  where  the  computer  assigns  a  class
label  to  all  (other)  image  pixels,  by  looking  for  each  pixel  to  which  of  the  trained
classes this pixel is most similar.
During  the  training  phase,  the  classes  to  be  use  are  previously  defined.  About  each
class some “ground truth” is needed: A number of places in the image area that are
known to belong to that class. This knowledge must have been acquired beforehand,
for instance as a result of fieldwork, or from an existing map (assuming that in some
areas  the  class  membership  has  not  changed  since  the  map  was  produced).  If  the
ground  truth  is  available,  training  samples  (small  areas  or  individual  pixels)  are
indicated in the image and the corresponding class names are entered. These training
samples are also termed as “Regions Of Interest” (ROI).
Guidelines for selecting training areas:
•   Training areas should be homogenous. This can be tested by graphic histograms,
numeric  summaries,  2-band  scatter  plot  for  investigating  separability  of  feature
classes  by  pairs  of  bands,  3-D  plot  of  3-band  feature  space  (if  the  software
•   One large ‘uniform’ training area per feature class is preferable to several smaller
training areas, though this must depend upon the degree of variability within each
class from site to site, and degree of variability within individual site.
•   Easy  to  extract  more  than  is  needed,  and  then  examine  site  statistics  before
making decision.
•   Each training area should be easily located in the image: use a topographic map,
nautical chart, or aerial photos to assist, though differential GPS observations may
•   If  a  smaller  training  area  is  necessary,  then  the  minimum  size  is  critical.  What
should be the size of the training site?
•   Note CCRS statement for MSS: individual training area should be minimum
of 3 - 4 pixels East-West by 6 pixels North-South.
•   Others [e.g. Swain and Davis, IDRISI] state (10 x # bands used), e.g. area of
40 pixels if all four MSS bands used (or approx 6 pixels x 7 pixels). Classification
It  is  the  task  of  the  decision  making  algorithm  to  make  a  partitioning  of  the  feature  space,
according to our training samples. For every possible feature vector in the feature space, the
program must decide to which of the sets of training pixels this feature vector is most similar.
After that, the program makes an output map where each image pixel is assigned a class
label, according to the feature space partitioning. Some algorithms are able to decide
that feature vectors in certain parts of the feature space are not similar to any of the
trained classes. They assign to those image pixels the class label “unknown”. In case
the area indeed contains classes that were not included in the training phase, the result
“unknown” is probably more realistic than to make a “wild” guess.120
In the following scatter plots, the different symbols are used:
C – corn, D – deciduous forest, S – sand, U – urban, W - water
•  - ROI mean, 1 – pixel to be classified
To find the relationship between classes and feature vectors is not as trivial as it may
seem.  Therefore,  various  decision  making  algorithms  are  being  used;  they  are
different in the way they partition the feature space. Three of them are:
The Box classifier is the simplest classification method: In 2-D space, rectangles are
created  around  the  training  feature  vector  for  each  class;  in  3-Dimension  they  are
actually boxes (blocks). The position and sizes of the boxes can be exactly around the
feature vectors (Min-Max method), or according to the mean vector (this will be at the
center  of  a  box)  and  the  standard  deviations  of  the  feature  vector,  calculated
separately per feature (this determines the size of the box in that dimension). In both
cases, the user is allowed to change the sizes by entering a multiplication factor. In
parts  of  the  features  space  where  boxes  overlap,  it  is  usual  to  give  priority  to  the
smallest  box.  Feature  vectors  in  the  image  that  fall  outside  all  boxes  will  be
The  Minimum  Distance-to-mean  classifier,  first  calculates  for  each  class  the  mean
vector of the training feature vectors. Then, the feature space is partitioned by giving
to  each  feature  vector  the  class  label  of  the  nearest  mean  vector,  according  to
Euclidean metric. Usually it is possible to specify a maximum distance threshold: If
the nearest mean is still further away than that threshold, it is assumed that none of the
classes is similar enough and the result will be “unknown”.
Gaussian  Maximum  Likelihood  classifiers  assume  that  the  feature  vectors  of  each
class  are  (statistically)  distributed  according  to  a  multivariate  normal  probability
density  function.  The  training  samples  are  used  to  estimate  the  parameters  of  the
distributions. The boundaries between the different partitions in the feature space are
placed where the decision changes from one class to another. They are called decision
boundaries. Accuracy Assessment
Accuracy may be defined, in a working sense, as the degree (often as a percentage) of
correspondence between observation and reality. Accuracy is usually judged against
existing maps, large scale aerial photos, or field checks. Two fundamental questions
about accuracy can be posed: Is each category in a classification really present at the
points specified on a map? Are the boundaries separating categories valid as located?
Various  types  of  errors  diminish  accuracy  of  feature  identification  and  category
distribution. Most are made either in measurement or in sampling. Three error types
1.  Data  Acquisition  Errors:  These  include  sensor  performance,  stability  of  the
platform, and conditions of viewing. They can be reduced or compensated by
making systematic corrections (e.g., by calibrating detector response with on-
board light sources generating known radiances). Corrections, often modified
by ancillary data such as known atmospheric conditions during overpass, can
be made during initial processing of raw data.
2.  Data Processing Errors: An example is possible misregistration of equivalent
pixels  in  the  different  bands  of  the  Landsat  Thematic  Mapper.  The  goal  in
geometric  correction  is  to  hold  the  mismatch  to  a  displacement  of  no  more
than  1  pixel.  Under  ideal  conditions,  and  with  as  many  as  25  ground  control
points (GCP) spread around a scene, this goal can be realized. Misregistrations
of several pixels significantly compromise accuracy.
3.  Scene-dependent Errors: One such error relates to how the class is defined and
established which, in turn, is sensitive to the resolution of both the observing
system and the reference map or photo. Three variants of these are exemplified
by  a  common  geologic  situation  in  which  the  sensor  data  are  processed
primarily to recognize rock types at the surface. In this there are pitfalls. First,
geologists in the field seek to map bedrock but over large parts of a surface,
soil  and  vegetation  cover  mask  the  bedrock  at  many  places.  The  geologist
makes  logical  deductions  in  the  field  as  to  the  rock  type  most  likely  buried
under the surface and shows this on a map (in which these masking materials
are treated as invisible [ignored]). This, unfortunately, does not correspond to
what  the  sensor  "sees".  Second,  most  geologic  maps  are  stratigraphic  rather
than  lithologic,  i.e.,  consist  of  units  identified  by  age  rather  than  rock  type.
Thus,  the  same  or  similar  rock  types  will  be  shown  by  different  symbols  or122
colors  on  the  map,  so  that  checking  for  ground  truth  requires  conversion  to
lithologies (often difficult because a unit may be diverse lithologically but was
chosen for some other mode of uniformity). Third, a rock type may need to be
considered  in  context  with  its  surroundings  in  order  to  be  properly  named.
Granite,  and  the  sedimentary  rock  called  arkose  derived  from  it,  both  have
similar spectral properties. The latter, however, typically is arranged in strata,
being   distributed   as   a   depositional   formation,   whose   spatial   patterns
(especially  when  exposed  as  folded  or  inclined  layers)  are  usually  quite
distinct from those of massive granites and are often revealed by topographic
These above points raise the question: Accuracy with respect to what? The maps we
use as the standards are largely extrapolations, or more rightly, abstractions. They are
often  thematic,  recording  one  or  more  surface  types  or  themes  -  the  signals  -  but
ignoring others -the noise. But, the sensor - whether or not it can resolve them - sees
all. When quantifying accuracy, the lack of equivalence and totality must be adjusted
for,  if  possible.  Another,  often  overlooked  point  about  maps  as  reference  standards
concerns their intrinsic or absolute accuracy. The map itself requires an independent
frame  of  reference  to  establish  its  own  validity.  For  centuries,  most  maps  were
constructed without regard to assessment of their inherent accuracy; in recent years,
some  maps  have  been  published  with  a  statement  of  level  of  confidence.  The  U.S.
Geological Survey has reported results of accuracy assessments of the 1:250,000 and
1:1,000,000 land use maps of Level 1 classifications based on aerial photos that meets
the 85% accuracy criterion at the 95% confidence level.
As a general rule, the level of accuracy obtainable in a remote sensing classification
depends  on  such  diverse  factors  as  the  suitability  of  training  sites,  the  size,  shape,
distribution,  and  frequency  of  occurrence  of  individual  areas  assigned  to  each  class,
the  sensor  performance  and  resolution,  and  the  methods  involved  in  classifying
(visual photointerpretation versus computer-aided statistical classifier), and others. A
quantitative  measure  of  the  mutual  role  of  improved  spatial  resolution  and  size  of
target on decreasing errors appears in this plot:123
The dramatic improvement in error reduction around 30 m (100 ft) is related in part to
the nature of the target classes - coarse resolution is ineffective in distinguishing crop
types but high resolution (< 20 m) adds little in recognizing these other than perhaps
identifying  species.  As  the  size  of  crop  fields  increases,  the  error  decreases  further.
The  anomalous  trend  for  forests  (maximum  error  at  high  resolution)  may  (?)  be  the
consequence  of  the  dictum:  "Can't  see  the  forest  for  the  trees."  -  here  meaning  that
high resolution begins to display individual species and breaks in the canopy that can
confuse the integrity of the class "forest". Two opposing trends influence the behavior
of these error curves: 1) statistical variance of the spectral response values decreases
whereas 2) the proportion of mixed pixels increases with lower resolution.
A  study  of  classification  accuracy  as  a  function  of  number  of  spectral  bands  shows
The increase from 1 to 2 bands produces the largest improvement in accuracy. After
about  4  bands  the  accuracy  increase  flattens  or  increases  very  slowly.  Thus,  extra
bands   may   be   redundant,   as   band-to-band   changes   are   cross-correlated   (this
correlation   may   be   minimized   and   even   put   to   advantage   through   Principal
Components  Analysis).  However,  additional  bands,  such  as  TM  5  and  7,  can  be
helpful  in  rock  type  (geology)  identification  because  compositional  absorption
features  diagnostic  of  different  types  reside  in  these  spectral  intervals.  Note  that
highest  accuracy  associates  with  crop  types  because  fields,  consisting  of  regularly-
space rows of plants against a background of soil, tend to be more uniform .
In practice, accuracy of classification may be tested in four ways: 1) field checks at
selected points (usually non-rigorous and subjective) chosen either at random or along
a  grid;  2)  estimate  (non-rigorous)  of  agreement  on  theme  or  class  identify  between
class map and reference maps, determinedusually by overlaying one on the other(s);
3) statistical analysis (rigorous) of numerical data developed in sampling, measuring,
and processing data, using such tests as root mean square, standard error, analysis of
variance,  correlation  coefficients,  linear  or  multiple  regression  analysis,  and  Chi-
square testing (see any standard text on statistics for explanation of these tests); and 4)
confusion  matrix  calculations  (rigorous).  This  last  approach  is  best  explained  by  the
writer's  study  made  of  a  subscene  from  a  July,  1977  Landsat  image  that  includes
Elkton, Maryland.
A  1:24,000  aerial  photo  that  falls  within  this  subscene  was  acquired  from  the  EPA.
Starting with a field visit in August, 1977 during the same growing season as the July
overpass, the crops in many individual farms located in the photo were identified, of
which  about  12  were  selected  as  training  sites.  Most  were  either  corn  or  soybeans;
others   were   mainly   barley   and   wheat.   A   Maximum   Likelihood   supervised
classification was then run.125
With the class identities in the photo as the standard, the number of pixels correctly
assigned  to  each  class  and  those  misassigned  to  other  classes  were  arranged  in  the
confusion  matrix  used  to  produce  the  summary  information  shown  in  the  following
table, listing errors of commission, omission, and overall accuracies:
Errors  of  commission  result  when  pixels  associated  with  a  class  are  incorrectly
identified  as  other  classes,  or  from  improperly  separating  a  single  class  into  two  or
more  classes.  Errors  of  omission  occur  whenever  pixels  that  should  have  been
identified  as  belonging  to  a  particular  class  were  simply  not  recognized  as  present.
Mapping accuracy for each class is stated as the number of correctly identified pixels
within  the  total  in  the  displayed  area  divided  by  that  number  plus  error  pixels  of
commission and omission. To illustrate, in the table, of the 43 pixels classed as corn
by  photointerpretation  and  ground  checks,  25  of  these  were  assigned  to  corn  in  the
Landsat classification, leaving 18/43 = 42% as the error of omission; likewise, of the
43, 7 were improperly identified as other than corn, producing a commission error of
16%.  Once  these  errors  are  determined  by  reference  to  "ground  truth",  they  can  be
reduced  by  selecting  new  training  sites  and  reclassifying,  by  renaming  classes  or
creating new ones, by combining them, or by using different classifiers. With each set
of  changes,  the  classification  procedure  is  iterated  until  a  final  level  of  acceptable
accuracy is reached.
It is not recommended to use the same sample map for both the classification and the
accuracy assessment, because this will produce figures that are too optimistic.126 Unsupervised classification (clustering)
One  way  to  perform  a  classification  is  to  plot  all  pixels  (all  feature  vectors)  of  the
image  in  a  feature  space,  and  then  to  analyze  the  feature  space  and  to  group  the
feature vectors into clusters. The name of this process is unsupervised classification.
In this process there is no knowledge about “thematic” land cover class names, such
as town, road, potatoes etc. All it can do, is to find out that there appears to be (for
example) 16 different “things” in the image and give them numbers(1 to 16). Each of
these “things” are called “spectral classes”. The result can be a raster map, in which
each  pixel  has  a  class  (from  1  to  16),  according  to  the  cluster  to  which  the  image
feature vector of the corresponding pixel belongs. After the process is finished, it is up
to  the  user  to  find  the  relationship  between  spectral  and  thematic  classes.  It  is  very
well possible, that it is discovered that one thematic class is split into several spectral
ones, or, worse, that several thematic classes ended up in the same cluster.
Various unsupervised classification (clustering) algorithms exist. Usually, they are not
completely automatic; the user must specify some parameters such as the number of
clusters (approximately) you want to obtain, the maximum cluster size (in the feature
space),  the  minimum  distance  (also  in  the  feature  space),  that  is  allowed  between
different  clusters  etc.  The  process  “builds”  clusters  as  it  is  scanning  through  the
image.  Typically,  when  a  cluster  becomes  larger  than  the  maximum  size,  it  is  split
into two clusters; on the other hand, when two clusters get nearer to each other than
the minimum distance, they are merged into one.
5.3.5 Unmixing
Natural  surfaces  are  rarely  composed  of  a  single  uniform  material.  Spectral  mixing
occurs  when  materials  with  different  spectral  properties  are  represented  by  a  single
image  pixel.  A  variety  of  factors  interact  to  produce  the  signal  received  by  the
imaging spectrometer:
•   A  very  thin  volume  of  material  interacts  with  incident  sunlight.  All  the
materials present in this volume contribute to the total reflected signal.
•   Spatial mixing of materials in the area represented by a single pixel result in
spectrally mixed reflected signals.
•   Variable illumination due to topography (shade) and actual shadow in the area
represented by the pixel further modify the reflected signal, basically mixing
with a "black" endmember.
•   The imaging spectrometer integrates the reflected light from each pixel. Modeling Mixed Spectra
The simplest model of a mixed spectrum is a linear one, in which the spectrum is a
linear  combination  of  the  "pure"  spectra  of  the  materials  located  in  the  pixel  area,
weighted by their fractional abundance.
This simple model can be formalized in three ways: a physical model a mathematical
model, and a geometric model. The physical model as discussed above includes the
Ground Instantaneous Field of View (GIFOV) of the pixels, the incoming irradiance,
the  photon-material  interactions,  and  the  resulting  mixed  spectra.  A  more  abstract
mathematical  model  is  required  to  simplify  the  problem  and  to  allow  inversion,  or
unmixing (that is, seperating the “mixed” materials that make the pixel).127
A  spectral  library  forms  the  initial  data  matrix  for  the  analysis.  The  ideal  spectral
library contains endmembers that when linearly combined can form all other spectra.
The  mathematical  model  is  a  simple  one.  The  observed  spectrum  (a  vector)  is
considered  to  be  the  product  of  multiplying  the  mixing  library  of  pure  endmember
spectra (a matrix) by the endmember abundances (a vector). An inverse of the original
spectral  library  matrix  is  formed  by  multiplying  together  the  transposes  of  the
orthogonal  matrices  and  the  reciprocal  values  of  the  diagonal  matrix  (Boardman,
1989). A simple vector-matrix multiplication between the inverse library matrix and
an  observed  mixed  spectrum  gives  an  estimate  of  the  abundance  of  the  library
endmembers for the unknown spectrum.
The  geometric  mixing  model  provides  an  alternate,  intuitive  means  to  understand
spectral mixing. Mixed pixels are visualized as points in n -dimensional scatter-plot
space (spectral space), where n is the number of bands. In two dimensions, if only two
endmembers  mix,  then  the  mixed  pixels  will  fall  in  a  line  (Figure  a).  The  pure
endmembers  will  fall  at  the  two  ends  of  the  mixing  line.  If  three  endmembers  mix,
then the mixed pixels will fall inside a triangle (Figure b). Mixtures of endmembers
"fill in" between the endmembers.
All mixed spectra are "interior" to the pure endmembers, inside the simplex formed
by the endmember vertices, because all the abundances are positive and sum to unity.
This "convex set" of mixed pixels can be used to determine how many endmembers
are present and to estimate their spectra. The geometric model is extensible to higher
dimensions where the number of mixing endmembers is one more than the inherent
dimensionality of the mixed data. Practical Unmixing Methods
Two very different types of unmixing are typically used: Using "known" endmembers
and using "derived" endmembers.
Using known endmembers, one seeks to derive the apparent fractional abundance of
each endmember material in each pixel, given a set of "known" or assumed spectral
endmembers.  These  known  endmembers  can  be  drawn  from  the  data  (averages  of
regions picked using previous knowledge), drawn from a library of pure materials by
interactively browsing through the imaging spectrometer data to determine what pure
materials  exist  in  the  image,  or  determined  using  expert  systems  (image  processing
methods  not  covered  here  –  MNF,  PPI,  and  others)  or  other  routines  to  identify
The mixing endmember matrix is made up of spectra from the image or a reference
library.  The  problem  can  be  cast  in  terms  of  an  overdetermined  linear  least  squares
problem. The mixing matrix is inverted and multiplied by the observed spectra to get
least-squares estimates of the unknown endmember abundance fractions. Constraints
can be placed on the solutions to give positive fractions that sum to unity. Shade and
shadow are included either implicitly (fractions sum to 1 or less) or explicitly as an
endmember (fractions sum to 1).
The  second  unmixing  method  uses  the  imaging  spectrometer  data  themselves  to
"derive"  the  mixing  endmembers  (Boardman  and  Kruse,   1994).   The   inherent
dimensionality of the data is determined using a special orthogonalization procedure
related to principal components:
•   A linear sub-space, or "flat" that spans the entire signal in the data is derived.
•   The data are projected onto this subspace, lowering the dimensionality of the
unmixing and removing most of the noise.
•   The convex hull of these projected data is found.
•   The  data  are  "shrink-wrapped"  by  a  simplex  of  n  -dimensions,  giving
estimates of the pure endmembers.
•   These derived endmembers  must  give  feasible  abundance  estimates  (positive
fractions that sum to unity).
Spectral unmixing is one of the most promising hyperspectral analysis research areas.
Analysis  procedures  using  the  convex  geometry  approach  already  developed  for
AVIRIS data have produced quantitative mapping results for a a variety of materials
(geology, vegetation, oceanography) without a priori knowledge. Combination of the
unmixing approach with model-based data calibration and expert system identification
capabilities  could  potentially  result  in  an  end-to-end  quantitative  yet  automated
analysis methodology.129
6. Active Remote Sensing
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
An active Remote Sensing system supplies its own source of energy, which is directed
at the object in order to measure the returned energy. Active microwave sensors are
generally divided into two distinct categories: imaging and non-imaging. The most
common  form  of  imaging  active  microwave  sensors  is  RADAR.  RADAR  is  an
acronym  for  RAdio  Detection  And  Ranging,  which  essentially  characterizes  the
function  and  operation  of  a  radar  sensor.  The  sensor  transmits  a  microwave  (radio)
signal  towards  the  target  and  detects  the  backscattered  portion  of  the  signal.  The
strength  of  the  backscattered  signal  is  measured  to  discriminate  between  different
targets  and  the  time  delay  between  the  transmitted  and  reflected  signals  determines
the distance (or range) to the target.
Non-imaging  microwave  sensors  include  altimeters  and  scatterometers.  In  most
cases these are profiling devices which take measurements in one linear dimension, as
opposed  to  the  two-dimensional  representation  of  imaging  sensors.  Radar  altimeters
transmit short microwave pulses and measure the round trip time delay to targets to
determine their distance from the sensor. Generally altimeters look straight down at
nadir below the platform and thus measure height or elevation (if the altitude of the
platform  is  accurately  known).  Radar  altimetry  is  used  on  aircraft  for  altitude
determination and on aircraft and satellites for topographic mapping and sea surface
height estimation. Scatterometers are also generally non-imaging sensors and are used
to  make  precise  quantitative  measurements  of  the  amount  of  energy  backscattered
from  targets.  The  amount  of  energy  backscattered  is  dependent  on  the  surface
properties (roughness) and the angle at which the microwave energy strikes the target.
Scatterometry measurements over ocean surfaces can be used to estimate wind speeds
based on the sea surface roughness. Ground-based scatterometers are used extensively
to  accurately  measure  the  backscatter  from  various  targets  in  order  to  characterize
different  materials  and  surface  types.  This  is  analogous  to  the  concept  of  spectral
reflectance curves in the optical spectrum.
Here  will  be  dealt  four  active  systems  which  have  interesting  applications  for
Hydrography: the Side  Looking  Airborne  Radar  (SLAR),  radar  altimeters,  Airborne
Laser  Scanning  (ALS),  and  also  the  Global  Positioning  System  (GPS)  -  being  used
not for positioning!.
6.1 Side Looking Airborne Radar (SLAR):
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
Radar is the acronym for Radio Detection and Ranging, indicating that it operates in
the radio and microwave bands of the Electromagnetic spectrum ranging from a meter
to a few milimeters in wavelength.
RADAR  systems  can  be  operated  independently  of  lighting  conditions  (i.e.  also  at
night) and largely independent of weather. In addition, the terrain can be illuminated
in the optimum direction to enhance features of interest.130
A typical radar system consists of (1) a pulse generator that discharges timed pulses of
microwave/radio  energy,  to  (2)  a  transmitter,  then  through  (3)  a  duplexer,  to  (4)  a
directional antenna that shapes and focuses each pulse into a stream transmitted to the
target;  (5)  returned  pulses  are  then  picked  up  by  the  same  antenna  and  sent  to  a
receiver  that  converts  (and  amplifies)  these  into  video  signals  carried  to  (6)  a
recording device which can store these digitally for later processing and/or produce a
realtime analog display on a cathode ray tube (CRT) or drive a moving light spot to
record  on  film.  Each  pulse  lasts  only  microseconds  (typically  there  are  about  1500
pulses per second); pulse length - an important factor along with band width in setting
the  system  resolution  -  is  the  distance  traveled  during  the  pulse  generation.  The
duplexer  serves  to  separate  the  outgoing  and  returned  pulses  (i.e.,  eliminate  their
mutual  interferences)  by  functioning  as  an  on-off  switch  that  blocks  out  reception
during  transmission  and  vice  versa.  The  antenna  on  a  ground  system  is  generally  a
parabolic "dish".
RADAR was developed during World War II for navigation and target location, using
the familiar rotating antenna and circular cathode-ray display. The conaissance strip-
mapping capability of Side-Looking Airborne Radar, or SLAR, was developed in the
1950s to aquire recnnaissance imagery without the necessity of overflying unfriendly
A  typical  SLAR  is  composed  of  an  antenna  that  both  transmits  the  radar  pulse  and
receives the return from  the terrain. The bursts of Electromagnetic energy from the
transmitter are of specific wavelength and duration, or pulse length. The timing of the
returned energy pulse determines the position of terrain features on the image.
A real aperture SLAR system operates with a long (~5-6 m) antenna usually shaped as
a  section  of  a  cylinder  wall.  This  type  produces  a  beam  of  noncoherent  pulses  and
utilizes its length to obtain the desired resolution (related to angular beamwidth) in the
azimuthal  (flight  line)  direction.  At  any  instant  the  transmitted  beam  is  propagated
outward within a fan-shaped plane perpendicular to the flight line.
A  second  type  of  system,  Synthetic  Aperture  Radar  (SAR),  is  exclusive  to  moving
platforms. It uses an antenna of much smaller physical dimensions, which sends forth
its signals from different positions as the platform advances, simulating a real aperture
by  integrating  the  pulse  "echoes"  into  a  composite  signal.  It  is  possible  through
appropriate  processing  to  simulate  effective  antennae  lengths  up  to  100  meters  or
more. This system depends on the Doppler effect to determine azimuth resolution. As
coherent  pulses  transmitted  from  the  radar  source  reflect  from  the  ground  to  the
advancing platform (air- or spacecraft), the target acts as though in apparent (relative)
motion. This motion results in changing frequencies, which give rise to variations in
phase  and  amplitude  in  the  returned  pulses.  These  data  are  recorded  for  later
processing  (employing  optical  [using  coherent  laser  light]  or  digital  correlation
methods), in which the moderated pulses are analyzed and recombined to synthesize
signals equivalent to those obtained by a narrow beam, real aperture system.131
The pulses of energy transmitted from the antenna iluminate narrow strips of terrain
oriented normal to the aircraft flight direction. In the following figure such a strip of
terrain and the shape of the energy pulse that it returns to the antenna, are illustrated.
The mountain front facing the antenna has a strong return because of its orientation
with respect to the radar antenna. The mountain blocks the transmitted pulse from the
terrain immediately downrange, and there is no return from that terrain. The resulting
dark signature on the image is called a radar shadow. Because of the diverse shapes
and  orientation,  vegetation  produces  a  speckled  signature  of  intermediate  intensity.
Metallic objects, such as the bridge, produce very strong returns and bright signatures
because of their geometry and electrical properties. Radar energy striking calm water
is almost totally reflected with the angle of reflection equal and opposite to the angle
of incidence; very little energy is returned to the antenna and a dark signature results.
Smooth surfaces such as calm water, dry lake beds, highways and airport runways are
called specular targets because of their mirror reflection of radar energy. When these
surfaces are located in the extreme near range position, a bright return can result; at
other locations they produce dark returns.
The return pulse is converted to scan line by assigning the the darkest tones of a gray
scale to the lower intensity returns and the lightest tones of a gray scale to the highest
intensity  returns.  In  addition  to  being  an  active  system,  radar  differs  from  other
Remote  Sensing  because  data  are  recorded  on  the  basis  of  time  rather  than  angular
distance.  Time  can  be  much  more  precisely  measured  and  recorded  than  angular
distance;  hence,  radar  imagery  can  be  acquired  at  long  ranges  with  high  resolution.132
Also,   atmospheric   absorption   and   scattering   are   minimal   at   most   microwave
6.1.1 Frequencies
Radar  operates  in  part  of  the  microwave  region  of  the  electromagnetic  spectrum,
specifically  over  the  frequency  interval  from  40000  to  300  megahertz  (MHz),  the
latter  extending  just  into  the  higher  frequency  end  of  the  radio  (broadcast)  region.
Commonly used frequencies and their corresponding wavelengths are specified by a
band nomenclature (these code letters were given during World War II, and remain to
this day), as follows:
•   K-alpha Band: 40000-26000 MHz (0.8 - 1.1 cm); K Band: 26500-18500 MHz (1.1
-  1.7  cm)  -  very  short  wavelengths  used  in  early  airborne  radar  systems  but
uncommon today.
•   X Band: 12500-8000 MHz (2.4 - 3.8 cm) - used extensively on airborne systems
for military reconnaissance and terrain mapping.
•   C  Band:  8000-4000  MHz  (3.8  -  7.5  cm)  -  common  on  many  airborne  research
systems  (CCRS  Convair-580  and  NASA  AirSAR)  and  spaceborne  systems
(including ERS-1 and 2 and RADARSAT).
•   S Band: 4000-2000 MHz (7.5 – 15.0 cm) - used on board the Russian ALMAZ
•   L Band: 2000-1000 MHz (15.0 - 30.0 cm) - used onboard American SEASAT and
Japanese JERS-1 satellites and NASA airborne system.
•   P Band: 1000- 300 Mhz (30.0 - 100.0 cm) - longest radar wavelengths, used on
NASA experimental airborne research system.133
6.1.2 Polarization
When  discussing  microwave  energy,  the  polarization  of  the  radiation  is  also
important.  Polarization  refers  to  the  orientation  of  the  electric  field  (recall  the
definition  of  electromagnetic  radiation  from).  Most  radars  are  designed  to  transmit
microwave  radiation  either  horizontally  polarized  (H)  or  vertically  polarized  (V).
Similarly,   the   antenna   receives   either   the   horizontally   or   vertically   polarized
backscattered energy, and some radars can receive both. These two polarization states
are designated by the letters H for horizontal, and V, for vertical. Thus, there can be
four combinations of both transmit and receive polarizations as follows:
•   HH - for horizontal transmit and horizontal receive,
•   VV - for vertical transmit and vertical receive,
•   HV - for horizontal transmit and vertical receive, and
•   VH - for vertical transmit and horizontal receive.
The first two polarization combinations are referred to as like-polarized because the
transmit  and  receive  polarizations  are  the  same.  The  last  two  combinations  are
referred  to  as  cross-polarized  because  the  transmit  and  receive  polarizations  are
opposite  of  one  another.  Similar  to  variations  in  wavelength,  depending  on  the
transmit   and   receive   polarizations,   the   radiation   will   interact   with   and   be
backscattered  differently  from  the  surface.  Both  wavelength  and  polarization  affect
how  a  radar  "sees"  the  surface.  Therefore,  radar  imagery  collected  using  different
polarization and wavelength combinations may provide different and complementary
information about the targets on the surface.
6.1.3 Viewing Geometry and Spatial Resolution
The imaging geometry of a radar system is different from the framing and scanning
systems  commonly  employed  for  optical  remote  sensing  described  in  Chapter  2.
Similar  to  optical  systems,  the  platform  travels  forward  in  the  flight  direction  (A)
with the nadir (B) directly beneath the platform. The microwave beam is transmitted
obliquely at right angles to the direction of flight illuminating a swath (C) which is
offset from nadir. Range (D) refers to the across-track dimension perpendicular to the
flight direction, while azimuth (E) refers to the along-track dimension parallel to the
flight  direction.  This  side-looking  viewing  geometry  is  typical  of  imaging  radar
systems (airborne or spaceborne).
Imaging geometry of a radar system
The  portion  of  the  image  swath  closest  to  the  nadir  track  of  the  radar  platform  is
called  the  near  range  (a)  while  the  portion  of  the  swath  farthest  from  the  nadir  is
called the far range (b). The incidence angle is the angle between the radar beam
and ground surface (A) which increases, moving across the swath from near to far134
range. The look angle (B) is the angle at which the radar "looks" at the surface. In the
near range, the viewing geometry may be referred to as being steep, relative to the far
range,  where  the  viewing  geometry  is  shallow.  At  all  ranges  the  radar  antenna
measures  the  radial  line  of  sight  distance  between  the  radar  and  each  target  on  the
surface. This is the slant range distance (C). The ground range distance (D) is the
true  horizontal  distance  along  the  ground  corresponding  to  each  point  measured  in
slant range.
Near\far range and incidence angles
Unlike  optical  systems,  a  radar's  spatial  resolution  is  a  function  of  the  specific
properties  of  the  microwave  radiation  and  geometrical  effects.  If  a  Real  Aperture
Radar  (RAR)  is  used  for  image  formation  (as  in  Side-Looking  Airborne  Radar)  a
single transmit pulse and the backscattered signal are used to form the image. In this
case, the resolution is dependent on the effective length of the pulse in the slant range
direction and on the width of the illumination in the azimuth direction. The range or
across-track  resolution  is  dependent  on  the  length  of  the  pulse  (P).  Two  distinct
targets  on  the  surface  will  be  resolved  in  the  range  dimension  if  their  separation  is
greater than half the pulse length. For example, targets 1 and 2 will not be separable
while  targets  3  and  4  will.  Slant  range  resolution  remains  constant,  independent  of
range.  However,  when  projected  into  ground  range  coordinates,  the  resolution  in
ground  range  will  be  dependent  of  the  incidence  angle.  Thus,  for  fixed  slant  range
resolution, the ground range resolution will decrease with increasing range.
Range or across-track resolution
The  azimuth  or  along-track  resolution  is  determined  by  the  angular  width  of  the
radiated  microwave  beam  and  the  slant  range  distance.  This  beamwidth  (A)  is  a
measure of the width of the illumination pattern. As the radar illumination propagates
to  increasing  distance  from  the  sensor,  the  azimuth  resolution  increases  (becomes
coarser). In this illustration, targets 1 and 2 in the near range would be separable, but
targets  3  and  4  at  further  range  would  not.  The  radar  beamwidth  is  inversely135
proportional to the antenna length (also referred to as the aperture) which means that a
longer antenna (or aperture) will produce a narrower beam and finer resolution.
Azimuth or along-track resolution
The equation for range resolution is:
R r  = tc / 2cosg
And for azimuth resolution:
R a  = 0.7Sg / D
t    Pulse length
c          speed of light
g    the angle between the horizontal plane and the line connecting the target and
the antenna (1 – incidence angle)
D         antenna length
Finer range resolution can be achieved by using a shorter pulse length, which can be
done  within  certain  engineering  design  restrictions.  Finer  azimuth  resolution  can  be
achieved by increasing the antenna length. However, the actual length of the antenna
is limited by what can be carried on an airborne or spaceborne platform. For airborne
radars, antennas are usually limited to one to two metres; for satellites they can be 10
to  15  metres  in  length.  To  overcome  this  size  limitation,  the  forward  motion  of  the
platform and special recording and processing of the backscattered echoes are used to
simulate a very long antenna and thus increase azimuth resolution.
Increase azimuth resolution
This figure illustrates how this is achieved. As a target (A) first enters the radar beam
(1), the backscattered echoes from each transmitted pulse begin to be recorded. As the
platform  continues  to  move  forward,  all  echoes  from  the  target  for  each  pulse  are
recorded during the entire time that the target is within the beam. The point at which
the target leaves the view of the radar beam (2) some time later, determines the length
of the simulated or synthesized antenna (B). Targets at far range, where the beam is
widest will be illuminated for a longer period of time than objects at near range. The
expanding beamwidth, combined with the increased time a target is within the beam
as  ground  range  increases,  balance  each  other,  such  that  the  resolution  remains
constant  across  the  entire  swath.  This  method  of  achieving  uniform,  fine  azimuth136
resolution  across  the  entire  imaging  swath  is  called  synthetic  aperture  radar,  or
SAR. Most airborne and spaceborne radars employ this type of radar.
6.1.4 Radar Image Distortions
As with all remote sensing systems, the viewing geometry of a radar results in certain
geometric distortions on the resultant imagery. However, there are key differences for
radar imagery which are due to the side-looking viewing geometry, and the fact that
the radar is fundamentally a distance measuring device (i.e. measuring range). Slant-
range scale distortion occurs because the radar is measuring the distance to features in
slant-range rather than the true horizontal distance along the ground. This results in a
varying image scale, moving from near to far range. Although targets A1 and B1 are
the same size on the ground, their apparent dimensions in slant range (A2 and B2) are
different. This causes targets in the near range to appear compressed relative to the far
range.  Using  trigonometry,  ground-range  distance  can  be  calculated  from  the  slant-
range  distance  and  platform  altitude  to  convert  to  the  proper  ground-range  format.
This conversion comparison shows a radar image in slant-range display (top) where
the fields and the road in the near range on the left side of the image are compressed,
and the same image converted to ground-range display (bottom) with the features in
their proper geometric shape.
Similar to the distortions encountered when using cameras and scanners, radar images
are also subject to geometric distortions due to relief displacement. As with scanner
imagery, this displacement is one-dimensional and occurs perpendicular to the flight
path.  However,  the  displacement  is  reversed  with  targets  being  displaced  towards,
instead  of  away  from  the  sensor.  Radar  foreshortening  and  layover  are  two
consequences which result from relief displacement.
When the radar beam reaches the base of a tall feature tilted towards the radar (e.g. a
mountain)  before  it  reaches  the  top  foreshortening  will  occur.  Again,  because  the
radar measures distance in slant-range, the slope (A to B) will appear compressed and
the  length  of  the  slope  will  be  represented  incorrectly  (A'  to  B').  Depending  on  the
angle of the hillside or mountain slope in relation to the incidence angle of the radar
beam, the severity of foreshortening will vary. Maximum foreshortening occurs when
the radar beam is perpendicular to the slope such that the slope, the base, and the top
are  imaged  simultaneously  (C  to  D).  The  length  of  the  slope  will  be  reduced  to  an
effective length of zero in slant range (C'D'). This figure shows a radar image of steep
mountainous  terrain  with  severe  foreshortening  effects.  The  foreshortened  slopes
appear as bright features on the image.137
Layover  occurs  when  the  radar  beam  reaches  the  top  of  a  tall  feature  (B)  before  it
reaches  the  base  (A).  The  return  signal  from  the  top  of  the  feature  will  be  received
before  the  signal  from  the  bottom.  As  a  result,  the  top  of  the  feature  is  displaced
towards the radar from its true position on the ground, and "lays over" the base of the
feature (B' to A'). Layover effects on a radar image look very similar to effects due to
foreshortening.  As  with  foreshortening,  layover  is  most  severe  for  small  incidence
angles, at the near range of a swath, and in mountainous terrain.
Both foreshortening and layover result in radar shadow. Radar shadow occurs when
the  radar  beam  is  not  able  to  illuminate  the  ground  surface.  Shadows  occur  in  the
down range dimension (i.e. towards the far range), behind vertical features or slopes
with  steep  sides.  Since  the  radar  beam  does  not  illuminate  the  surface,  shadowed
regions will appear dark on an image as no energy is available to be backscattered. As
incidence angle increases from near to far range, so will shadow effects as the radar
beam  looks  more  and  more  obliquely  at  the  surface.  This  image  illustrates  radar
shadow effects on the right side of the hillsides which are being illuminated from the
6.1.5 Target Interaction and Image Appearance
The  brightness  of  features  in  a  radar  image  is  dependent  on  the  portion  of  the
transmitted energy that is returned back to the radar from targets on the surface. The
magnitude  or  intensity  of  this  backscattered  energy  is  dependent  on  how  the  radar
energy  interacts  with  the  surface,  which  is  a  function  of  several  variables  or
parameters. These parameters include the particular characteristics of the radar system
(frequency, polarization, viewing geometry, etc.) as well as the characteristics of the
surface   (landcover   type,   topography,   relief,   etc.).   Because   many   of   these
characteristics are interrelated, it is impossible to separate out each of their individual
contributions to the appearance of features in a radar image. Changes in the various
parameters may have an impact on and affect the response of other parameters, which
together will affect the amount of backscatter. Thus, the brightness of features in an
image  is  usually  a  combination  of  several  of  these  variables.  However,  for  the
purposes of our discussion, we can group these characteristics into three areas which
fundamentally control radar energy/target interactions. They are:
•   Surface roughness of the target
•   Radar viewing and surface geometry relationship
•   Moisture content and electrical properties of the target
The surface roughness of a feature controls how the microwave energy interacts with
that  surface  or  target  and  is  generally  the  dominant  factor  in  determining  the  tones
seen on a radar image. Surface roughness refers to the average height variations in the
surface  cover  from  a  plane  surface,  and  is  measured  on  the  order  of  centimetres.
Whether a surface appears rough or smooth to a radar depends on the wavelength and
incidence angle.139
Surface roughness Spectral diffusity of the surface
Simply put, a surface is considered "smooth" if the height variations are much smaller
than the radar wavelength. When the surface height variations begin to approach the
size  of  the  wavelength,  then  the  surface  will  appear  "rough".  Thus,  a  given  surface
will  appear  rougher  as  the  wavelength  becomes  shorter  and  smoother  as  the
wavelength becomes longer. A smooth surface (A) causes specular reflection of the
incident  energy  (generally  away  from  the  sensor)  and  thus  only  a  small  amount  of
energy  is  returned  to  the  radar.  This  results  in  smooth  surfaces  appearing  as  darker
toned areas on an image. A rough surface (B) will scatter the energy approximately
equally in all directions (i.e. diffusely) and a significant portion of the energy will be
backscattered  to  the  radar.  Thus,  rough  surfaces  will  appear  lighter  in  tone  on  an
image.  Incidence  angle,  in  combination  with  wavelength,  also  plays  a  role  in  the
apparent roughness of a surface. For a given surface and wavelength, the surface will
appear smoother as the incidence angle increases. Thus, as we move farther across the
swath,  from  near  to  far  range,  less  energy  would  be  returned  to  the  sensor  and  the
image would become increasingly darker in tone.
The relationship of radar wavelength and depression angle to surface roughness may
be described by the Rayleigh criterion that considers a surface to be smooth if
h < l / 8sing
h = the height of surface irregularities, or surface roughness
l = the radar wavelength
g = the grazing angle between the terrain and the incident radar wave
both h and l are given in the same units, usually centimeters. For horizontal terrain
surfaces, the grazing angle equals the antenna depression angle. For an X-band radar
(l = 3 cm) at a depression angle of 45º, the roughness value at which the surface will
appear smooth is:        h < 0.53 cm
This  means  that  a  vertical  relief  of  0.53  cm  is  the  theoretical  boundary  between
smooth  and  rough  surfaces  for  the  given  wavelength  and  depression  angle.  The
Rayleigh  criterion  does  not  consider  the  important  category  of  surface  relief  that  is
intermediate between definitely smooth and definitely rough surfaces. The Rayleigh
criterion was modified by Peake and Oliver to define the upper and lower values of h140
for surfaces of intermediate roughness. Their smooth criterion considers a surface to
be smooth if
h < l / 25sing
For  the  above  parameters,  a  vertical  relief  of    0.17  cm  is  the  boundary  between
smooth surfaces and surfaces of intermediate roughness.
Smooth surface with specular reflection; no return
They have also derived a rough criterion that considers a surface to be rough if
h > l / 4.4sing
Which for the above parameters gives the result of 0.96 cm.
Smooth surfaces produce strong returns at depression angles near vertical, but little or
no return at lower angles. For a rough surface, the relatively uniform return decreases
somewhat at low depression angles because of the longer two way travel distance.
Return Intensity as a function of depression angle Local incidence angle
We  have  already  discussed  incidence  or  look  angle  in  relation  to  viewing  geometry
and  how  changes  in  this  angle  affect  the  signal  returned  to  the  radar.  However,  in
relation to surface geometry, and its effect on target interaction and image appearance,
the  local  incidence  angle  is  a  more  appropriate  and  relevant  concept.  The  local141
incidence angle is the angle between the radar beam and a line perpendicular to the
slope at the point of incidence (A). Thus, local incidence angle takes into account the
local  slope  of  the  terrain  in  relation  to  the  radar  beam.  With  flat  terrain,  the  local
incidence  angle  is  the  same  as  the  look  angle  (B)  of  the  radar.  For  terrain  with  any
type of relief, this is not the case. Generally, slopes facing towards the radar will have
small  local  incidence  angles,  causing  relatively  strong  backscattering  to  the  sensor,
which results in a bright-toned appearance in an image.
As  the  concept  of  local  incidence  angle  demonstrates,  the  relationship  between
viewing geometry and the geometry of the surface features plays an important role in
how the radar energy interacts with targets and their corresponding brightness on an
image. Variations in viewing geometry will accentuate and enhance topography and
relief  in  different  ways,  such  that  varying  degrees  of  foreshortening,  layover,  and
shadow  (previous  section)  may  occur  depending  on  surface  slope,  orientation,  and
The  look  direction  or  aspect  angle  of  the  radar  describes  the  orientation  of  the
transmitted radar beam relative to the direction or alignment of linear features on the
surface. The look direction can significantly influence the appearance of features on a
radar  image,  particularly  when  ground  features  are  organized  in  a  linear  structure
(such  as  agricultural  crops  or  mountain  ranges).  If  the  look  direction  is  close  to
perpendicular to the orientation of the feature (A), then a large portion of the incident
energy will be reflected back to the sensor and the feature will appear as a brighter
tone.  If  the  look  direction  is  more  oblique  in  relation  to  the  feature  orientation  (B),
then  less  energy  will  be  returned  to  the  radar  and  the  feature  will  appear  darker  in
tone.  Look  direction  is  important  for  enhancing  the  contrast  between  features  in  an
image.  It  is  particularly  important  to  have  the  proper  look  direction  in  mountainous
regions  in  order  to  minimize  effects  such  as  layover  and  shadowing.  By  acquiring
imagery from different look directions, it may be possible to enhance identification of
features with different orientations relative to the radar.
Features which have two (or more) surfaces (usually smooth) at right angles to one
another, may cause corner reflection to occur if the 'corner' faces the general direction
of the radar antenna. The orientation of the surfaces at right angles causes most of the
radar energy to be reflected directly back to the antenna due to the double bounce (or
more) reflection. Corner reflectors with complex angular shapes are common in urban
environments   (e.g.   buildings   and   streets,   bridges,   other   man-made   structures).
Naturally occurring corner reflectors may include severely folded rock and cliff faces
or upright vegetation standing in water. In all cases, corner reflectors show up as very
bright targets in an image, such as the buildings and other man-made structures in this
radar image of a city.142
The presence (or absence) of moisture affects the electrical properties of an object or
medium.  Changes  in  the  electrical  properties  influence  the  absorption,  transmission,
and  reflection  of  microwave  energy.  Thus,  the  moisture  content  will  influence  how
targets and surfaces reflect energy from a radar and how they will appear on an image.
Generally,  reflectivity  (and  image  brightness)  increases  with  increased  moisture
content. For example, surfaces such as soil and vegetation cover will appear brighter
when they are wet than when they are dry.
When  a  target  is  moist  or  wet,  scattering  from  the  topmost  portion  (surface
scattering) is the dominant process taking place. The type of reflection (ranging from
specular to diffuse) and the magnitude will depend on how rough the material appears
to the radar. If the target is very dry and the surface appears smooth to the radar, the
radar  energy  may  be  able  to  penetrate  below  the  surface,  whether  that  surface  is
discontinuous  (e.g.  forest  canopy  with  leaves  and  branches),  or  a  homogeneous
surface (e.g. soil, sand, or ice). For a given surface , longer wavelengths are able to
penetrate further than shorter wavelengths.
If  the  radar  energy  does  manage  to  penetrate  through  the  topmost  surface,  then
volume  scattering  may  occur.  Volume  scattering  is  the  scattering  of  radar  energy
within a volume or medium, and usually consists of multiple bounces and reflections
from different components within the volume. For example, in a forest, scattering may
come  from  the  leaf  canopy  at  the  tops  of  the  trees,  the  leaves  and  branches  further
below, and the tree trunks and soil at the ground level. Volume scattering may serve
to  decrease  or  increase  image  brightness,  depending  on  how  much  of  the  energy  is
scattered out of the volume and back to the radar.143
6.1.6 Radar Image Properties
All  radar  images  appear  with  some  degree  of  what  we  call  radar  speckle.  Speckle
appears as a grainy "salt and pepper" texture in an image. This is caused by random
constructive and destructive interference from the multiple scattering returns that will
occur within each resolution cell. As an example, an homogeneous target, such as a
large  grass-covered  field,  without  the  effects  of  speckle  would  generally  result  in
light-toned  pixel  values  on  an  image  (A).  However,  reflections  from  the  individual
blades of grass within each resolution cell results in some image pixels being brighter
and some being darker than the average tone (B), such that the field appears speckled.
Speckle Grass covered field
Speckle is essentially a form of noise which degrades the quality of an image and may
make interpretation (visual or digital) more difficult. Thus, it is generally desirable to
reduce  speckle  prior  to  interpretation  and  analysis.  Speckle  reduction  can  be
achieved in two ways:
•   multi-look processing, or
•   spatial filtering.
Multi-look processing refers to the division of the radar beam (A) into several (in this
example, five) narrower sub-beams (1 to 5). Each sub-beam provides an independent
"look" at the illuminated scene, as the name suggests. Each of these "looks" will also
be subject to speckle, but by summing and averaging them together to form the final
output image, the amount of speckle will be reduced.
While  multi-looking  is  usually  done  during  data  acquisition,  speckle  reduction  by
spatial  filtering  is  performed  on  the  output  image  in  a  digital  (i.e.  computer)  image
analysis environment. Speckle reduction filtering consists of moving a small window
of a few pixels in dimension (e.g. 3x3 or 5x5) over each pixel in the image, applying a
mathematical  calculation  using  the  pixel  values  under  that  window  (e.g.  calculating144
the  average),  and  replacing  the  central  pixel  with  the  new  value.  The  window  is
moved  along  in  both  the  row  and  column  dimensions  one  pixel  at  a  time,  until  the
entire image has been covered. By calculating the average of a small window around
each pixel, a smoothing effect is achieved and the visual appearance of the speckle is
reduced.  This  graphic  shows  a  radar  image  before  (top)  and  after  (bottom)  speckle
reduction  using  an  averaging  filter.  The  median  (or  middle)  value  of  all  the  pixels
underneath  the  moving  window  is  also  often  used  to  reduce  speckle.  Other  more
complex filtering calculations can be performed to reduce speckle while minimizing
the amount of smoothing taking place.
3*3 average filter window Speckle reduction using an average filtering
Both  multi-look  processing  and  spatial  filtering  reduce  speckle  at  the  expense  of
resolution,  since  they  both  essentially  smooth  the  image.  Therefore,  the  amount  of
speckle reduction desired must be balanced with the particular application the image
is being used for, and the amount of detail required. If fine detail and high resolution
is  required  then  little  or  no  multi-looking/spatial  filtering  should  be  done.  If  broad-
scale interpretation and mapping is the application, then speckle reduction techniques
may be more appropriate and acceptable.
Another  property  peculiar  to  radar  images  is  slant-range  distortion,  which  was
discussed  in  some  detail  in  section  3.4.  Features  in  the  near-range  are  compressed
relative to features in the far range due to the slant-range scale variability. For most
applications,  it  is  desirable  to  have  the  radar  image  presented  in  a  format  which
corrects  for  this  distortion,  to  enable  true  distance  measurements  between  features.
This requires the slant-range image to be converted to 'ground range' display. This can
be done by the radar processor prior to creating an image or after data acquisition by
applying a transformation to the slant range image. In most cases, this conversion will
only be an estimate of the geometry of the ground features due to the complications
introduced by variations in terrain relief and topography.
A  radar  antenna  transmits  more  power  in  the  mid-range  portion  of  the  illuminated
swath than at the near and far ranges. This effect is known as antenna pattern and
results  in  stronger  returns  from  the  center  portion  of  the  swath  than  at  the  edges.
Combined with this antenna pattern effect is the fact that the energy returned to the
radar decreases dramatically as the range distance increases. Thus, for a given surface,
the strength of the returned signal becomes smaller and smaller moving farther across
the swath. These effects combine to produce an image which varies in intensity (tone)
in  the  range  direction  across  the  image.  A  process  known  as  antenna  pattern
correction may be applied to produce a uniform average brightness across the imaged
swath, to better facilitate visual interpretation.145
The range of brightness levels a remote sensing system can differentiate is related to
radiometric resolution (section 2.5) and is referred to as the dynamic  range.  While
optical  sensors,  such  as  those  carried  by  satellites  such  as  Landsat  and  SPOT,
typically produce 256 intensity levels, radar systems can differentiate intensity levels
up  to  around  100,000  levels!  Since  the  human  eye  can  only  discriminate  about  40
intensity  levels  at  one  time,  this  is  too  much  information  for  visual  interpretation.
Even a typical computer would have difficulty dealing with this range of information.
Therefore, most radars record and process the original data as 16 bits (65,536 levels of
intensity),  which  are  then  further  scaled  down  to  8  bits  (256  levels)  for  visual
interpretation and/or digital computer analysis.
Calibration  is  a  process  which  ensures  that  the  radar  system  and  the  signals  that  it
measures are as consistent and as accurate as possible. Prior to analysis, most radar
images  will  require  relative  calibration.  Relative  calibration  corrects  for  known
variations in radar antenna and systems response and ensures that uniform, repeatable
measurements can be made over time. This allows relative comparisons between the
response of features within a single image, and between separate images to be made
with confidence. However, if we wish to make accurate quantitative measurements
representing the actual energy or power returned from various features or targets for
comparative purposes, then absolute calibration is necessary.
Absolute calibration, a much more involved process than relative calibration, attempts
to relate the magnitude of the recorded signal strength to the actual amount of energy
backscattered from each resolution cell. To achieve this, detailed measurements of the
radar  system  properties  are  required  as  well  as  quantitative  measurements  of  the
scattering  properties  of  specific  targets.  The  latter  are  often  obtained  using  ground-
based scatterometers, as described in section 3.1. Also, devices called transponders
may  be  placed  on  the  ground  prior  to  data  acquisition  to  calibrate  an  image.  These
devices receive the incoming radar signal, amplify it, and transmit a return signal of
known strength back to the radar. By knowing the actual strength of this return signal
in the image, the responses from other features can be referenced to it.
6.1.7 Advanced Radar Applications
In  addition  to  standard  acquisition  and  use  of  radar  data,  there  are  three  specific
applications worth mentioning.
The first is stereo radar which is similar in concept to stereo mapping using aerial
photography (described in section 2.7). Stereo radar image pairs are acquired covering
the  same  area,  but  with  different  look/incidence  angles  (A),  or  opposite  look
directions (B). Unlike aerial photos where the displacement is radially outward from
the nadir point directly below the camera, radar images show displacement only in the
range  direction.  Stereo  pairs  taken  from  opposite  look  directions  (i.e.  one  looking146
north  and  the  other  south)  may  show  significant  contrast  and  may  be  difficult  to
interpret  visually  or  digitally.  In  mountainous  terrain,  this  will  be  even  more
pronounced  as  shadowing  on  opposite  sides  of  features  will  eliminate  the  stereo
effect. Same side stereo imaging (A) has been used operationally for years to assist in
interpretation  for  forestry  and  geology  and  also  to  generate  topographic  maps.  The
estimation of distance measurements and terrain height for topographic mapping from
stereo  radar  data  is  called  radargrammetry,  and  is  analogous  to  photogrammetry
carried out for similar purposes with aerial photographs.
Radargrammetry  is  one  method  of  estimating  terrain  height  using  radar.  Another,
more advanced method is called interferometry. Interferometry relies on being able
to measure a property of electromagnetic waves called phase. Suppose we have two
waves with the exact same wavelength and frequency traveling along in space, but the
starting  point  of  one  is  offset  slightly  from  the  other.  The  offset  between  matching
points on these two waves (A) is called the phase difference. Interferometric systems
use  two  antennas,  separated  in  the  range  dimension  by  a  small  distance,  both
recording the returns from each resolution cell. The two antennas can be on the same
platform (as with some airborne SARs), or the data can be acquired from two different
passes  with  the  same  sensor,  such  has  been  done  with  both  airborne  and  satellite
radars. By measuring the exact phase difference between the two returns (A), the path
length  difference  can  be  calculated  to  an  accuracy  that  is  on  the  order  of  the
wavelength (i.e centimetres). Knowing the position of the antennas with respect to the
Earth's  surface,  the  position  of  the  resolution  cell,  including  its  elevation,  can  be
determined.  The  phase  difference  between  adjacent  resolution  cells,  is  illustrated  in
this interferogram, where colours represents the variations in height. The information
contained  in  an  interferogram  can  be  used  to  derive  topographic  information  and
produce three-dimensional imagery of terrain height.
Phase difference Interferometric systems Interferogram
The concept of radar polarimetry was already alluded to in our discussion of radar
fundamentals in section 3.2. As its name implies, polarimetry involves discriminating
between the polarizations  that  a  radar  system  is  able  to  transmit  and  receive.  Most147
radars   transmit   microwave   radiation   in   either   horizontal   (H)   or   vertical   (V)
polarization,  and  similarly,  receive  the  backscattered  signal  at  only  one  of  these
polarizations.   Multi-polarization   radars   are   able   to   transmit   either   H   or   V
polarization and receive both the like- and cross-polarized returns (e.g. HH and HV or
VV  and  VH,  where  the  first  letter  stands  for  the  polarization  transmitted  and  the
second letter the polarization received). Polarimetric radars are able to transmit and
receive both horizontal and vertical polarizations. Thus, they are able to receive and
process all four combinations of these polarizations: HH, HV, VH, and VV. Each of
these   "polarization   channels"   have   varying   sensitivities   to   different   surface
characteristics and properties. Thus, the availability of multi-polarization data helps to
improve the identification of, and the discrimination between features. In addition to
recording the magnitude (i.e. the strength) of the returned signal for each polarization,
most polarimetric radars are also able to record the phase information of the returned
signals.  This  can  be  used  to  further  characterize  the  polarimetric  "signature"  of
different surface features.148
7. Remote Sensing Applications for the sea (passive and SLAR)
7.1 Sea Surface Temperature
Estimation  of  Sea  Surface  Temperature  (SST)  is  important  for  weather  models,
estimation of heat content and heat flux, and also for climate monitoring and climate
change detection.
SST  can  be  measured  either  by  Thermal  Infrared  Remote  Sensing  or  by  Passive
Microwave  Remote  Sensing,  the  former  (Thermal  IR)  being  more  applied.  Thermal
infrared remote sensing is the oldest technique used to measure SST. This is because
the thermal infrared is near the peak of the Planck function at terrestrial temperatures,
and the signal is thus strongest at the wavelengths coupled with the fact that thermal
infrared detectors are relatively easy to build and reliable. A middle infrared channel
is also often used at night (when not contaminated by reflected solar), because this is
in   a   spectral   region   of   strong   variance   in   the   Planck   function   at   terrestrial
•   Strengths
This strong signal of temperature in the thermal wavelengths provides the thermal
infrared with the ability to measure with good accuracy and high resolution. This
combination is very important for analysis of long term climate variability, as well
as for shorter term weater forecasting. The time series of available infrared SST
measurements (20 years) also provides a long term record which is very valuable.
•   Weaknesses
The  biggest  problem  with  thermal  infrared  remote  sensing  is  the  presence  of  an
atmosphere. This atmospheric interference takes on two forms, total disruption of
the  signal  in  the  presence  of  clouds  and  atmospheric  attenuation  of  infrared
radiation  by  gases,  water  vapor,  and  aerosols.  If  a  cloud  is  present  in  a  satellite
field  of  view,  the  signal  recieved  by  a  satellite  sensor  is  a  measurement  of  the
temperature of the cloud top, and thus cannot be used to calculate an SST. Even
when no clouds are present, atmospheric  corrections  are  required  to  compensate
for radiation scattered into or out of the field of view or absorbed by atmospheric
constituents (this subject was dealt in the chapter about corrections, previously).
The following steps are followed in taking a satellite bit stream and then converting it
to an SST map:
•   Radiance calibration (from DN to radiance), followed by a temperature calibration
using  the  information  provided  by  on  board  blackbody,  we  get  the  brightness
temperature (radiant, not kinetic).
•   Geometric correction.
•   Cloud detection and filtering
•   Atmospheric  correction  (for  water  vapor,  gaseous  constituents  and  aerosol
attenuation) and SST calcualtion.
•   Quality check using a climatology/algorithm intercomparison.
The most used sensor for SST is the NOAA AVHRR, which has the following
•   Global Coverage Daily
•   Long Heritage (20 years)
•   Source for US Operational SST
•   3.7, 11, and 12 Micron Thermal Bands
•   1 and 4 km Resolution Data
•   Accuracy ~ 0.6K rms
Other   sensors   used   are   the   Along   Track   Scanning   Radiometer   (ATSR),   the
Geostationary   Operational   Environmental   Satellite   (GOES),   and   the   TRMM
Microwave Imager (TMI). Each one of them has its own advantages and weaknesses.
Regarding the GOES, two points are interesting:
On  the  one  hand,  as  it  stays  over  a  fixed  point,  it  enables  a  study  of  the  diurnal
variation in the cycle of SST.
On the other hand, its position in a fixed orbit also causes problems as it goes into and
out  of  the  shadow  of  the  earth  each  day  and  one  side  "freezes"  as  the  other  side
"cooks", creating a (known) bias of up to 0.5K .
The following problems exist in the Remote Sensing of SST:
•   Bulk/Skin Temperature Difference
A  temperature  gradient  usually  exists  across  the  millimeter  immediately  below  the
ocean's surface or skin. This temperature gradient is related to the net air-sea heat flux
and influences air-sea interactions and temperature measurements taken at or near the
Typically the 1mm skin is about 0.3-0.5K cooler than the bulk of the water beneath.
This is mainly a problem at thermal wavelengths, emission from top micron of ocean.
As microwave emission penetrates deeper, there the effect is less important.
•   Emissivity Variations of Seawater
The emissivity of seawater in the thermal region of the electromagnetic spectrum is
very nearly one. However, when seeking an accuracy of .2K or better from a signal of
approximately 300K, an error of .1% is enough to. The emissivity of water is affected
by the salinity of the water, and as this may vary throughout the ocean, it becomes a
factor that is not readily corrected for, even though its magnitude is very small. The150
emissivity of water is also a function of viewing angle, and this must be corrected for.
Normally it is implicitly absorbed in the viewing angle dependent terms of the SST
retrieval equations.
The  variation  of  emissivity  of  seawater  at  microwave  frequencies  is  much  more
substantial than in thermal wavelengths. It is dependent not only on salinity here, but
on surface roughness as well. The fact that surface roughness is dependent upon wind
speed  has  led  to  the  development  of  wind  speed  retrieval  algorithms  based  on
microwave  emission  and  its  variation  due  to  wind  induced  roughness  emissivity
7.2 Oil Spill Detection
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
Oil  spills  can  destroy  marine  life  as  well  as  damage  habitat  for  land  animals  and
humans. To limit the areas affected by the spill and facilitate containment and cleanup
efforts, a number of factors have to be identified:
•   Spill location
•   Size and extent of the spill
•   Direction and magnitude of oil movement
•   Wind, current and wave information for predicting future oil movement
Remote  sensing  offers  the  advantage  of  being  able  to  observe  events  in  remote  and
often  inaccessible  areas.  For  example,  oil  spills  from  ruptured  pipelines,  may  go
unchecked  for  a  period  of  time  because  of  uncertainty  of  the  exact  location  of  the
spill, and limited knowledge of the extent of the spill. Remote sensing can be used to
both  detect  and  monitor  spills.  Detecting  oil  seeps  can  be  also  used  in  exploration
For  ocean  spills,  remote  sensing  data  can  provide  information  on  the  rate  and
direction  of  oil  movement  through  multi-temporal  imaging,  and  input  to  drift
prediction  modelling  and  may  facilitate  in  targeting  clean-up  and  control  efforts.
Remote sensing devices used include the use of infrared video and photography from
airborne platforms, thermal infrared imaging, airborne laser fluourosensors, airborne
and  space-borne  optical  sensors,  as  well  as  airborne  and  spaceborne  SAR.  SAR
sensors  have  an  advantage  over  optical  sensors  in  that  they  can  provide  data  under
poor  weather  conditions  and  during  darkness.  Users  of  remotely  sensed  data  for  oil
spill applications include the Coast Guard, national environmental protection agencies
and  departments,  oil  companies,  shipping  industry,  insurance  industry,  fishing
industry, national departments of fisheries and oceans, and departments of defence.
The key operational data requirements are fast turnaround time and frequent imaging
of the site to monitor the dynamics of the spill. For spill identification, high resolution
sensors  are  generally  required,  although  wide  area  coverage  is  very  important  for
initial monitoring and detection. Airborne sensors have the advantage of frequent site
specific  coverage  on  demand,  however,  they  can  be  costly.  Spills  often  occur  in
inclement weather, which can hinder airborne surveillance.
Laser fluorosensors are the best sensors for oil spill detection (where it is needed to
separate between the quite similar spectra of oil and water; with UV oil films as thin151
as 15 mm  can  be  detected),  and  have  the  capability  of  identifying  oil  on  shores,  ice
and snow, and determining what type of oil has been spilled. However, they require
relatively cloud free conditions to detect the oilspill.
On  daytime  and  nighttime  IR  images  both  refined  and  crude  oils  have  cooler
signatures  than  the  adjacent  clean  water.  The  oil  and  water  have  the  same  kinetic
temperature beacause the two liquids are in direct contact. However, the emissivity of
pure water is 0.993, and a thin film of petroleum reduces the emissivity to 0.972. For
a  kinetic  emperature  of  18ºC,  the  radiant  temperature  of  pure  water  will  be  17.5ºC,
while that of the oil film will be just 15.9ºC. IR images have the advantage of being
available   both   for   day   and   night   operations.   However,   rain   and   fog   preven
temperature image acquisition, and the interpreter must be careful to avoid confusing
cold water currents with oil slicks.
SAR  sensors  can  image  oilspills  through  the  localized  suppression  of  Bragg  scale
waves. Oilspills are visible on a radar image as circular or curvilinear features with a
darker  tone  than  the  surrounding  ocean  (the  small  waves  that  cause  backscatter  on
radar  images  of  the  sea  are  dampened  by  a  thin  film  of  oil).  The  detection  of  an
oilspill  is  strongly  dependent  upon  the  wind  speed.  At  wind  speeds  greater  than  10
m/s, the slick will be broken up and dispersed, making it difficult to detect. Another
factor that can play a role in the successful detection of an oilspill is the difficulty in
distinguishing  between  a  natural  surfactant  and  an  oilspill.  Multi-temporal  data  and
ancillary information can help to discriminate between these two phenomena.
7.2.1 Case study 1 – oil slick
A  supertanker,  the  Sea  Empress,  was  grounded  near  the  town  of  Milford  Haven,
Wales  on  February  15,  1996.  After  hitting  rocks,  the  outer  hull  was  breached  and
approximately 70,000 tonnes of light grade crude oil was dispersed southward under
storm conditions.
In this RADARSAT image taken a week after the spill, the extent of the oil is visible.
The  dark  areas  off  the  coast  represent  the  areas  where  oil  is  present  and  areas  of
lighter  tone  directly  south  are  areas  where  dispersant  was  sprayed  on  the  oil  to
encourage emulsification. Oil, which floats on the top of water, suppresses the ocean's
capillary  waves,  creating  a  surface  smoother  than  the  surrounding  water.  This
smoother  surface  appears  dark  in  the  radar  image.  As  the  oil  starts  to  emulsify  and
clean-up efforts begin to take effect, the capillary waves are not as effectively damped
and  the  oil  appears  lighter.  Size,  location  and  dispersal  of  the  oil  spill  can  be
determined using this type of imagery.
7.2.2 Case study 2 – oil seep
EarthSat  developed  the  Seep  Enhancement  Algorithm  (SEA),  to  select  and  process
satellite  images  for  offshore  hydrocarbon  seep  detection.  SEA  operates  with  image152
data from a variety of satellites (primarily Landsat, ERS-1 and SPOT), detecting seeps
under a broad range of environmental and imaging conditions.
Offshore Oil Seeps
Seeps  are  direct  evidence  of  hydrocarbon  occurrences.  Seeps  are  associated  with
many major onshore discoveries. Seeps provide significant targets for more intensive
exploration  efforts.  Many  prospective  offshore  areas  have  yet  to  be  searched  for
Finding Seeps by Satellite
Thin oil slicks form where bubbles of oil and gas reach the surface. Oil slicks affect
water in two important ways that are readily detected by imaging satellites:
•   Spectral: oil slicks increase reflectance in the visible through near-infrared portion
of the electromagnetic spectrum.
•   Textural: oil slicks smooth the sea surface, reducing the amount of reflected sun
glint ("glitter") and radar backscatter.
The type of detection (spectral vs. textural) depends on seepage rate, oil composition,
sea state and illumination. Only satellite imagery provides detailed data on the shape
and size of natural oil slicks to pinpoint seep location and estimate seepage rates.
7.3 Ice motion and monitoring
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
Ice  moves  quickly  and  sometimes  unpredictably  in  response  to  ocean  currents  and
wind. Vessels can be trapped or damaged by the pressure resulting from these moving
ice floes. Even offshore structures can be damaged by the strength and momentum of
moving ice. For these reasons it is important to understand the ice dynamics in areas
of construction or in the vicinity of a shipping/fishing route.
Remote  sensing  gives  a  tangible  measure  of  direction  and  rate  of  ice  movement
through mapping and change detection techniques. Ice floes actually have individual
morphological characteristics (shape, structures) that allow them to be distinguished
from  one  another.  The  floes  can  be  mapped  and  their  movement  monitored  to
facilitate in planning optimum shipping routes, to predict the effect of ice movement
on  standing  structures  (bridges,  platforms).Users  of  this  type  of  information  include
the shipping, fishing, and tourism industries, as well as engineers involved in offshore
platform and bridge design and maintenance.
Monitoring  of  ice  movement  requires  frequent  and  reliable  imaging.  The  revisit
interval  must  be  frequent  enough  to  follow  identifiable  features  before  tracking
becomes  difficult  due  to  excessive  movement  or  change  in  appearance.  Active
microwave  sensing  (radar)  provides  a  reliable  source  of  imaging  under  all  weather
and  illumination  conditions.  RADARSAT  provides  this  type  of  sensor  and  is  a
spaceborne  platform,  which  is  advantageous  for  routine  imaging  operations.  The
orbital path ensures that Arctic areas are covered daily which meets the requirement
for frequent imaging.
Surface  roughness  explains  most  of  the  signature  characteristics  of  sea  ice,  but  the
brine content of the ice also influences the radar return. Young ice has a high content153
of brine that is progressively reduced with time to a minimum in multi-year ice. Brine
has a higher dielectric constant than pure ice, which reduces the radar return.
The resolution and imaging frequency requirements for ice motion tracking vary with
the size of floes and the ice dynamics in a region. In areas of large slow moving floes
(e.g. Beaufort Sea), 1km resolution data over 10 day intervals is adequate. In dynamic
marginal ice zones (e.g. Gulf of St. Lawrence), 100m resolution data over 12-24 hr
intervals is required.
It is also possible to use passive Remote Sensing for identifying sea ice, thoush it is
prone  to  atmospheric  attenuation,  and  the  orbital  and  temporal  of  most  of  these
satellites (TM, SPOT) are not adequate. In order to distinguish ice from clouds, the
following criteria can be used: the more uniform brightness of ice, the sharp contacts
with water of the ice floes, and clous shadows.
ERS-1 image, Northern Lake Erie, 1 Feb 1994,
7.3.1 Case study (example): E.-A. Herland, Operational ice monitoring with ERS-1
SAR, Space Technology, Technical Research Centre of Finland,
The  ice  season  in  the  northern  Baltic  Sea  lasts  for  more  than  six  months  in  the
northernmost part during normal winters, with the maximum extent occurring in the
January-March period. This severely affects marine traffic in the area, and the Finnish
Board of Navigation operates nine icebreakers in the area in order to support marine
traffic  and  keep  important  harbours  open  during  the  ice  season.  The  Ice  Service
compiles ice charts daily during the ice season, and information on the ice condition is
distributed  to  shipping  and  harbour  authorities  and  to  ships  in  the  area.  Since  1989
icebreakers have used an image workstation, developed by VTT (Technical Research
Centre  of  Finland),  for  viewing  NOAA  images.  This  workstation  has  later  been
expanded to incorporate ERS-l SAR images, too.154
Each ERS-1 SAR image covers an area of 100 x 100 km with 30 m resolution and is
produced  in  near  real  time  at  the  receiving  stations  in  Kiruna,  Sweden,  and  Fucino,
Italy.  During  the  winter  1994  the  images  used  were  low-resolution  (100  m)  images
generated by lowpass filtering FD-images. During the winters 1992 and 1994 ERS-1
was in the 3-day cycle orbit and in 1993 in the 35-day cycle orbit.
Images  were  transmitted  in  digital  form  to  the  icebreakers,  where  the  images  are
displayed on an image workstation. In orderto be able to combine different types of
images,  map  information,  and  other  information  available  to  the  icebreakers,  all
images  are  transformed  to  the  Mercator  projection  before  transmission  to  the
icebreakers.  For  this  purpose  a  special  image  format  has  been  developed,  which
includes all necessary information on image acquisition time, sensor and geographical
The  experiences  from  the  three  winters  of  1992-4  have  shown  that  it  is  technically
feasible to use satellite SAR images for real-time ice monitoring. The main bottleneck
has turned out to be time delay from satellite overpass until the user has the image.
With  the  ERS-1  coverage,  it  is  also  not  possible  to  cover  all  interesting  areas  in  a
timely  fashion. With  systems  like  Radarsat,  the  coverage  problem  is  alleviated,  but
large areal coverage will most likely require a satellite distribution channel even for
100  m  resolution  images.  In  order  to  allow  near  real-time  use  of  the  images,  this
distribution must be done immediately after the processing at the ground station, and
the  processing  capacity  must  be  sufficient  for  the  required  areal  coverage.  The
usefulness of the images has been clearly demonstrated, both for icebreaker operation
and ice charting. If a recent satellite image is available, the icebreaker workstation lets
the captain optimise the route for the icebreaker and advise ships by radio on how to
find useful leads in the ice, thus minimising the need for icebreaker assistance. For ice
charting,  SAR  images  offer  a  unique  capability,  especially  during  cloudy  weather
Concept of optimal ice routing between ice floes in a 10 by 10 km area using SAR images (bold red
line) compared with theoretical straight line between two waypoints (dashed green line).
Taken from:
7.4 Mapping the sea floor (bathymetry)
§     Sabins Floyd F. (1976), Remote Sensing – Principles and Interpretation, Freeman
Many parts of the world’s sea floor are charted only in a general fashion or the charts
are  inaccurate  (being  out  of  date,  etc).  Accurate  charts  are  especially  needed  for
shallow  shelf  areas  where  submarine  deposition,  erosion,  and  growth  of  coral  reefs
can  change  the  bottom  topography  within  a  few  years  after  a  bathymetric  survey  is
Remote Sensing of the sea floor from aircraft or satellites is restricted by the fact that
water  absorbs  or  reflects  most  wavelengths  or  electromagnetic  energy.  IR  energy  is
absorbed  by  waterm,  and  only  visible  wavelengths  penetrate  water.  The  depth  of
penetration is influenced by the turbidity of the water and the wavelength of the light.
A  10-m  layer  of  clear  ocean  water  transmits  almost  50  percent  of  the  incident
radiation  from  0.4  to  0.6  mm  in  wavelength,  but  transmits  less  than  10  percent  of
radiation from 0.6 to 0.7 mm wavelength. The increased turbidity results in a decrease
of light transmittance and a shift in the wavelength of maximum transmittance to the
0.5 to 0.6 mm region.
Spectral transmittance for 10m of various types of water
The signatures on satellite images are influenced not only by water depth, but also by
water  clarity,  reflectance  of  the  bottom  sediment,  and  by  atmospheric  conditions.
Therefore  the  depth  associated  with  image  signatures  in  one  island  will  not
necessarily be the same in other areas.
7.4.1 Case study – SHOM’s nautical space chart (La spatiocarte marine)
The SHOM (Service Hydrographique et Océanographique de la Marine) is in charge
of   the   hydrography   and   of   the   nautical   cartography   in   areas   under   French
responsability.  Thus,  Overseas  French  Regions  (DOM)  and  Territories  (TOM),
Southern lands and Antarctica are as concerned as the metropolitan territory is.
But,  if  the  hydrography  and  cartography  of  the  metropolitan  territory  and  of  the
Overseas Regions of France (DOM) have reached a high level, numerous regions of
the  Southern  Pacific  and  of  the  Indian  Ocean  are  not  completely  hydrographied.
Topographic or bathymetric data are too often fragmented or isolated, even obsolete,
in order to allow the publication of nautical charts of high quality.156
With  the  present  classic  tools,  several  years  would  be  necessary  to  carry  out
hydrographic  works  on  these  areas  and  enable  to  cartography  them  matching
acceptable standards.
Since several years, satellite imagery is one of the source of data used by the SHOM
for  nautical  chart  elaboration.  Mariners  do  not  always  notice  it.  Indeed,  once
processed,  the  informations  provided  by  satellites  are  reproduced  according  to  the
graphics of classic cartography.
With  nautical  space  charts,  a  further  step  was  reached  because  spatial  data  are
becoming  the  chart  basis  (terrestrial  topography,  sea  beds  geomorphology).  These
space charts are distributed as the other SHOM products.
The  satellite  images  analysis  will  enable  the  SHOM  to  reactualize,  within  the  ten
years to come, the nautical charts covering the whole of the South Pacific islands and
French atolls.
The studies carried out since the Preliminary Evaluation Program of SPOT (PEPS -
1984) on bathymetry have led to the implementation of a water height measurement
The results of the studies on bathymetry are directly linked to the characteristics of the
SPOT satellite.
The  sensors  enable  to  measure  the  radiometry.  Radiometry  is  the  intensity  of  the
radiation  reflected  by  the  floor  and  by  the  sea  floor  through  the  water  layer.  This
intensity is measured within the visible spectrum for well determined frequency bands
called "channels". On the images, it can be noticed that deeper the sea bed is, greater
the radiance is absorbed and lower is the measured radiometry level.
Spectral resolution enables the observation of objects, in clear waters, up to 22 to 25
meters depths on the XS1 channel and up to 5 to 7 meters depths on the XS2 channel.
As for the XS3 channel, it doesn't provide any bathymetric information because the
infrared  is  absorbed  by  the  water.  This  channel  indicates,  for  the  coastal  strip,  the
waters limit when the satellite acquires the scene.
Extract of the SPOT 512/382 scene (12-09-95) Katiu atoll
This diagram represents the variations of the levels of radiometry of the three channels (blue for XS1,
green for XS2 and red for XS3) along the line (yellow color) of the small image.
Diagram analysis:
A  strong  variation  for  the  three  channels  is  to  be  noticed  on  the  left  part.  This  area
corresponds to the terrestrial part. It is composed of a barrier reef on which there is
some red, it corresponds to the vegetation (coconut trees). On this object, the response157
on the channel 3 remains strong. On the contrary, it is to be noticed that the response
on  the  two  other  channels  is  very  low  on  this  sector  of  vegetation  while  it  is  very
strong on both sides of the coral sands.
On the right part of the diagram, an important peak is to be noticed, it corresponds, on
the small image, to a well visible coral macro-structure, indeed the most influenced
are the channels 1 and 2.
On the central part of the diagram, peaks of variable intensity are to be noticed, they
correspond to isolated objects (coral pinnacles). At their level, variations more or less
marked are noticeable on channels 1 and 2; these variations correspond to the water
height above the shallows. The variation on channel 3 is very low, if not non-existent;
it  means  that  the  detected  pinnacles  were  awash  or  submerged  when  the  satellite
acquired the scene.
Different shades of blue are also noticeable on the small image (ranging from light to
dark blue); these shades show the variations of the water height (from shallow to deep
waters).  On  the  diagram  showing  the  different  variations  for  each  of  the  three
channels, the XS1 channel is the most influenced.
Modelization of the bathymetry:
The principle of the bathymetric measurement is based on a simple reflection model.
This model links the intensity of the radiometric signal measured by the satellite to the
depth. It can call on two methods:
* the physical method. It requires the knowledge of all the parameters which preside
to this model (water optical properties, seabed reflection coefficient, transparency of
atmosphere,  ...).  But,  most  of  time,  failing  anything  better,  these  parameters  are
valued and considered as constant on areas which are supposed to be homogeneous.
* the empirical method. It consists in adjusting the coefficients of the model on the in-
situ data taking, if possible, the nature of seabed into account.
The second method is used at the SHOM.
In order to avoid the bias due to the choice of the calibrate points of bathymetry, the
average values of the soundings, which correspond to homogeneous areas of constant
radiometry, are introduced into the model.
Some  twenty  values  (from  known  soundings  values  of  Ground  Control  Points)  are
necessary  for  an  adequate  gauging  of  this  model.  Its  mathematic  formula  is  as
Z = A * LN (R1 - R1inf) + B * LN (R2 - R2inf) + C
Z is for the depth calculated for the Ri radiometry pixel, the suffixes 1 and 2 refer to
the XS1 and XS2 channels, Riinf represents the radiometry in deep waters for the i
channel, the coefficients A, B and C are obtained with the least squares method from
the calibrate measures.
The  calculated  model  provides  a  single-channel  image,  on  256  levels,  each  pixel  of
the  maritime  area  is  represented  by  a  calculated  depth  and  not  by  a  measured
radiometry as it used to be.158
7.5 Vessel detection
§     Vachon, P.W., J.W.M. Campbell, C. Bjerkelund, F.W. Dobson, and M.T. Rey, 'Validation of Ship
Detection by the RADARSAT SAR' submitted to Proc. Pacific Ocean Remote Sensing Conference
(PORSEC'96), 13-16 Aug. 1996, Victoria, Canada.
The  use  of  remote  sensing  techniques  for  observing  ships  has  gained  substantial
momentum  over  the  past  decade.  Several  data  sources  exist:  satellite  and  airborne
radar  and  optical  instruments,  as  well  as  ground-based  HF  radar.  Here  we  focus  on
spaceborne  synthetic  aperture  radar  (SAR).  The  ERS  and  Radarsat  satellites  have
provided the first civilian long term sources of space based image data that present an
efficient  means  of  wide  area  monitoring  of  ocean  traffic,  both  day  and  night,
independent  of  visibility.  Developments  in  computer  power,  storage  space  and
communications  technology  have  made  it  possible  to  utilize  the  observations  in  a
matter  of  minutes  or  hours  if  necessary,  by  processing  raw  data  into  imagery,
analyzing   and   disseminating   images   and/or   analysis   results   to   operational
organisations both ashore and at sea.
While  the  vessel  detection  in  SAR  data  problem  initially  could  be  regarded  as
primarily a military area of interest, a number of civilian issues have emerged where
such  observations  can  be  very  useful:  fisheries  management  and  enforcement  (see
figure), oil pollution monitoring, search and rescue operations, monitoring of illegal
transportation  of  goods  and  people,  and  also  collection  of  statistical  information  to
assess requirements for navigational aids and insurance risks.
Radarsat  image  acquired  in  ScanSAR  Narrow  Far,  19  Aug.  1996  (Copyright  CSA  1996.)  Fishing
vessels (black dots are lined up along the border between the "Loop Hole" and Norwegian waters (the
colors are inverted).
These  issues  have  spurred  the  establishment  of  R&D  and  pilot  programs  in  several
countries, aimed at developing efficient methods of exploiting satellite SAR data for
vessel  traffic  monitoring:  Australia,  Canada,  Japan,  The  Netherlands,  Norway,  the
United  Kingdom  and  the  USA  to  mention  a  few.  Some  of  these  programs  have
transitioned  to  operational  status.  Certainly,  it  has  come  to  be  recognized  that  SAR
observations, when used in conjunction with other sources of information, can provide159
important support in making go/no go decisions for deployment of patrol aircraft or
Key questions from the operational agencies are:
1)  What information is available, and how often and how rapidly is it available?
2)  What is the minimum detectable vessel size?
3)  How reliable is it? i.e. what is the false alarm rate, and what about the ones that
are missed?
In  the  mid  1990’s,  several  algorithms  for  vessel  detection  in  SAR  images  were
published. This established some baseline estimates, that together with an analysis of
satellite orbits and ground station locations, provide some site specific answers to the
first  question.  In  addition  to  vessel  position,  methods  for  estimating  vessel  size,
orientation, velocity and type have been developed.
The last four or five years have seen a large effort towards operationalizing algorithms
and answering questions two and three. However, validation of marine observations is
expensive  and  effort  intensive,  and  remains  a  key  issue.  While  significant  progress
has  been  made  in  verifying  detection  rates  and  size  estimates  for  large  vessels,  we
have not seen any published results extending to vessels smaller than 20 m under well
documented environmental conditions. Other parameters remain mostly unvalidated,
at least in the open literature (the last four www sites listed at the top of this section,
give examples).
A  statistical  model  for  RADARSAT's  expected  ship  detection  performance  was
developed based upon experience with ERS-1 SAR data. The model includes ocean
clutter, SAR image probability density function, and ship cross section elements. Ship
detectibility predictions from the model are shown in the following figure.160
Plot of predictions of minimum detectable ship length as a function of beam mode and incidence angle
subject to the assumptions of the model described by Vachon Paris (from the CCRS) et al., 1996. The
minimum detectable ship length is expressed as a figure of merit for a 10m/s wind speed and allows for
comparison between ERS-1 SAR and RADARSAT's various beam modes.
Key  issues  for  the  future  include  further  validation  of  existing  algorithms,  and
continued access to suitable SAR data. Envisat and Radarsat-2, with their abilities to
collect co- and cross-polarized data, will represent significant improvements in vessel
detection reliability.
Plans  for  other  commercial  SAR  missions  are  also  emerging,  some  with  a  focus  on
higher  spatial  resolution  (one  to  three  metres)  at  the  expense  of  area  coverage,
possibly at frequencies other than C-band. While this may support extraction of more
information  about  observed  vessels,  it  will  not  have  a  similar  operational  impact  as
multi-polarized data.
The AMRS Forum on vessel detection can help address some of these issues through
facilitating co-ordination of validation efforts and exchange of results and experience.
Hopefully, we can initiate exchange of views between people working with different
instrument types. Also, there is a role to play in providing market driven requirements
to the commercial consortia planning new missions.161
8. Digital Video Remote Sensing
8.1 A little history
The use of video as a remote sensing tool started in the late 1960s and early 1970s.
The sustained interest of video in the remote sensing is illustrated by the increasing
number  of  papers  dealing  with  video  as  remote  sensing  tool  and  the  first  1988
international workshop dealing only with videography. Airborne video systems have
been applied in a wide range of topics like forestry, biology, agriculture, coastal zone
monitoring, landscape analysis and also in urbanised areas.
The increasing interest is partly due to new technologies such as the use of solid state
Charged-Coupled Devices (CCDs) sensors instead of tubes and the introduction of the
Super-VHS  (S-VHS)  recorder.  The  use  of  CCDs  improved  the  geometric  accuracy
and  the  radiometric  performance  of  the  video  camera.  The  development  of  S-VHS
increased the image to a 400-lines resolution instead of the typical VHS (colour) 230-
lines resolution. New technologies such as digital video and the advances in computer
hardware and processing techniques make the use of video as a survey tool even more
8.2 General advantages of using video
Today, a number of airborne video systems exists. They all have their specific pros
and  cons  but  the  general  characteristics  making  video  to  a  valuable  sensor  can  be
summarised as follows:
•    low cost due to off-the shelf systems,
•    real-time  and  near  real-time  availability  of  imagery  for  visual  assessment  or
image analysis,
•    immediate potential for digital processing on the signal,
•    acquired data can be checked in real-time making real-time surveys possible,
•    high  sensitivity  of  CCD  enabling  the  collection  of  spectral  data  in  narrow
bands (10 nm),
•    linking  of  video  data  with  ground  control  points  and  Global  Positioning
System (GPS) information,
•    wide spectral sensitivity of modern silicon detectors,
•    data  redundancy  in  which  images  are  acquired  every  1/25  second  producing
multiple views of a target or scene.
Basic video technical aspects are itemised as follows :
8.3 The video image
This part discusses the principles of the video image, what is its structure and what
kind of information does it contain?
Video signals are subjected to very strict norms. In Europe the so called PAL (Phase
Alternate Line) standard defined in the 1950's, is used. Other countries like the USA
use  NTSC  as  standard.  The  different  video  standards  are  not  compatible  with  each
other. Throughout these pages we will discuss the European PAL video standard.162
The PAL standard uses the principle of interlacing as illustrated in the figure. A video
image  (also  called  video  frame)  consists  of  two  fields.  The  even  field  contains  the
even  line  numbers  of  a  frame,  the  odd  field  the  odd  line  numbers.  With  the  PAL
standard 25 frames per second (25 Hz) are built up. Consequently fields are produced
with a rate of 50 fields per sec (50 Hz).
Two  aspects  are  of  importance  to  build  up  video  images;  the  radiometric  and  the
geometric information.
The video image radiometry
In  the  camera  photosensitive  CCD  elements  build  up  electronic  charges  caused  by
incoming light. The more light 'hits' the CCD element the higher the electronic charge
with a maximum of 0.7 Volt - the peak white level - and a minimum of 0 Volt - the
blanking level (= black). This information is read out by registers in the camera line
by line for each field. Synchronisation pulses (see below) are placed after each line of
radiometric information.
The video image geometry
Video  image  geometry  is  very  important.  Each  subsequent  field  should  be  aligned
properly   as   well   as   the   lines   with   radiometric   information   within   a   field.
Synchronisation pulses take care of that. In contrast to the radiometric information the
pulses  are  characterised  by  negative  voltages.  This  way  the  recording  and  display
devices  can  separate  geometric  information  form  radiometric  information.  Roughly
there      are      two      types      of      synchronisation      pulses      of      importance:
•     VSync : The start of each field is indicated by a field blanking period or Vertical
Synchronisation  pulse  (VSync);  a  pulse  used  to  trigger  the  vertical  retrace  of  a
scanning electron gun from the bottom of a field back to the top left.163
•     Hsync : Each line within a field is separated by a negative voltage of -0.3 Volt
called  the  Horizontal  Synchronisation  pulse  ;  a  pulse  used  to  trigger  the
horizontal retrace of a scanning electron gun from right to left and one line down.
The following figure shows the information a video line contains:
8.4 Charge-Coupled Devices
There  are  a  lot  of  video  cameras  available  but  the  modern  ones  have  one  thing  in
common:  Charge-Coupled  Devices  function  as  photosensitive  plates  and  have
replaced  the  old-fashioned  tube  cameras.  The  physics  of  CCDs  is  based  on  the
principle of a Metal-Oxide Semiconductor (MOS). Such MOS capacitor is formed by
positioning  a  metal  electrode,  insulated  by  a  film  of  silicon  dioxide,  onto  a  silicon
Example of an MOS capacitor generating a depletion region which is proportional to the amount of
incoming light.
Semi-conductor  light  sensors  and  the  use  of  charge-coupling  techniques  to  move
charge  around  within  the  device  structure  are  applied  in  video  and  photography.
Incident light is proportional converted by these sensors to an electronic charge and
trapped  in  the  depletion  region.  The  ability  to  store  a  quantity  of  charge  is
fundamental to the operation of CCD devices. It corresponds to a memory device, but
one that stores analogue quantities. A camera requires an Area Imaging Device that
can sense the whole picture focused on it by the lens and the optical system: a CCD
array. The CCD array analyses the picture brightness with the charge value on each
photo sensing element representing the brightness at that point.164
The dimension of an CCD array is standardised and is usually given in inches of its
diagonal (see Table). These figures you can usually find in the camera specifications
as well as the number of CCD elements present in an array.
CCD array size specifications
CCD array (inch)          Horizontal (mm) Vertical (mm)          Diagonal (mm)
1" CCD 12.6 9.5 15.8
2/3" CCD 8.8 6.6 11.0
1/2" CCD 6.4 4.8 8.0
Each array is made up of thousands of CCD elements. The more elements used in a
given picture size, the smaller each will become increasing the geometric resolution
but reducing its light sensitivity. The more common cameras usually contain up to 1
CCD  array,  while  the  professional  (and  more  costly)  cameras  may  have  3  CCDs
arrays. In the latest case a CCD array is available for each of the RGB (Red, Green
and Blue) video channels.
The time to which the CCD array is exposed to the light source is determined by an
electronic shutter. As CCD elements are more sensible to light then photographic film
very  high  shutter  speeds  can  be  used  usually  ranging  from  1/50  sec.  to  1/1000  sec.
Nowadays even cameras with shutter speed up to 1/10,000 sec exist.
Dependent  on  the  material  the  CCDs  are  made  of  they  have  a  wide  radiometric
spectrum ranging from the Blue Band (ca. 400 nm) up to Mid Infra-red (ca. 2400 nm).
This  makes  that  video  can  be  applied  in  many  different  surveys  from  true  colour
observations to thermal surveys of the earth surface.
8.5 The geometric resolution of video
One  should  use  common  sense  and  apply  video  for  what  it  is  suited  for.  One  can
obtain centimetre geometric resolution but this will go at the expense of the area one
video frame is covering, also called field of view (=fov).
If tens of kilometres have to be recorded you can consider to apply airborne video for
giving a fast overview of the area under study. If you are interested in a small specific
area some kilometres wide you can use video at a higher geometric scale.165
To  give  you  some  ideas  about  the  geometric  resolution  of  the  airborne  CIR  video
system (where CIR stands for Colour Infra-Red) we use you can have a look at the
following table based on a camera focal length of 7 mm.
Flying altitude (mtr)       Field of view (mtr) Geometric resolution (cm)
100 90 x 70 12
200 140 x 105 25
300 275 x 210 38
400 370 x 275 51
500 460 x 350 64
1000 900 x700 125
1500 1350 x 1050 190
2000 1750 x 1300 255
2500 2250 x 1700 315
3000 2750 x 2100 385
8.6 Airborne video data acquisition
Airborne  video  data  acquisition  is  quite  easy.  For  vertical  recording  you  can  either
mount  the  video  camera  above  a  hatch  present  in  the  bottom  of  the  aeroplane  or
outside the aeroplane. The latter option requires that the camera should be relatively
small and that changes in the camera settings can be set by means of remote control.
In any case the camera as well as the videorecorder (=VCR) should be stabilised in
the mounting.
Another  possibility  is  of  course  recording  out  of  hand.  In  most  cases  this  is  not
recommended as it may cause severe camera movements. On the other hand it enables
freedom of camera handling independent of unexpected movements of the airplane.
After  the  camera  system  is  set  up  and  various  devices  are  connected  the  airborne
survey can start. During the survey the analogue video signal is stored on. One can
also perform a real time survey of the area by looking at the monitor that is connected
to the camera. If the camera is linked to a GPS (Global Positioning System) receiver
every video frame is not only time coded but also GPS coded.
For  digital  video  image  analysis  a  so  called  frame  grabber  is  required.  This  device
converts the analogue video signal on the tape into a digital format. This process is
also called A/D conversion.
A/D  conversion  is  simple.  A  VCR  is  linked  to  a  framegrabber  present  in  a  PC  or
Workstation. Running the tape one can see what has been recorded including GPS and
/or  time  code  information.  This  GPS  information  is  very  useful  for  mapping  large
areas or areas with little points of recognition. If the area of interest has been reached
one pushes the button and the analogue video information is converted into a digital
video frame ready to be analysed
By means of this technique the areas of interest can be selected in a fast and easy way
out of several Gigabytes of potential material. This 'newspaper approach' is very time
and cost effective.166
8.7 Types of airborne video systems
In the airborne video remote sensing basically two types of video systems exist when
we  consider  multispectral  information;  multiplex  camera  systems  and  single  camera
systems. On this part we look at some of the characteristics of these two systems.
8.7.1 Multiband information
In  the  spectral  analysis  of  images  we  are  especially  interested  in  multiband
information  as  the  use  of  more  bands  makes  discrimination  between  objects  more
easy. Especially the use of a NIR band (700 nm - 1300 nm) in the system makes that
different vegetation types can be discerned from each other.
In  the  USA  initially  single  band  systems  were  used  using  black  &  white  (B&W)
cameras. These cameras had visible (400 - 700 nm) or NIR sensitivity (700 - 900 nm).
The  systems  were  portable,  cheap  and  easy  in  operation.  Due  to  the  single  band
information these systems were not very versatile and limited in their application.
8.7.2 Multiplex systems
In  a  following  stage  multiband  systems  has  been  developed  by  combining  several
B&W video cameras. Each camera represents one spectral band by placing a narrow
band filter in front of the camera. Each camera is connected to a VCR and a monitor.
An encoder is integrated to provide false colour information out of the single bands.
These so called multiplex video systems have the following advantages with respect
to multiband single camera systems:
•   Flexible spectral band choice
•   Narrow spectral bands (10 nm)
The disadvantages are:167
•   Bulky/heavy
•   Difficult camera alignment
•   Moderate to high costs (c. USD 20,000)
•   No oblique recording possible
8.7.3 Single camera systems
Another line was set out by modifying standard true colour cameras into Colour Infra-
Red  (CIR)  video  systems.  One  camera  yields  three  band  information  in  the  Green,
Red and NIR band. The camera is connected to a VCR and a monitor.
Synoptics uses - when required - an industrial S-VHS 3 CCD camera called Silvacam.
A single multiband camera system has the following advantages:
•   Very flexible in use
•   No camera alignment required
•   Oblique recording possible
•   Very high image quality
•   Real time survey in the False Colour spectrum
The disadvantages are:
•   Fixed spectral bands
•   Relative broad spectral bands
•   Moderate to high costs (c. USD 20,000)
8.8 The Silvacam camera
Synoptics uses a single camera Colour Infra-Red (CIR) video system called Silvacam
for  surveys  requiring  false  colour  data  The  system  is  very  simple  and  flexible  and
looks like a true colour industrial camera.168
The camera is converted from a true colour system into a false colour system by using
the same principles as CIR photography; the green spectral information is diverted to
the Blue video channel, the Red spectral information to the Green video channel and
the NIR information to the red video channel. Each video channel (Red, Green, Blue)
is  represented  by  a  CCD  array.  The  incoming  light  is  split  into  the  three  respective
colours   and   lead   to   the   respective   channels   by   means   of   dichroic   mirrors.
By  placing  internal  filters  in  the  camera  it  obtains  its  spectral  False  Colour
characteristics;  a  passband  in  the  Green  (505  -  580  nm),  one  in  the  Red  (580  -  680
nm) and one in the NIR (770 - 845 nm).169
The video information is stored on tape and can be analysed by running the tape by
means of a VCR which is connected to a framegrabber. Once the area of interest is
visible, a push of the button is sufficient to convert the analogue video signal into a
digital image.
8.8 Applications
8.8.1 Coastal monitoring in the Netherlands
In  1995  the  Survey  Department  of  the  Netherlands  launched  a  proposal  for  the
introduction  of  new  techniques  for  the  acquisition  of  elevation  data  and  thematic
imagery  of  the  coastal  zone.  The  techniques  consist  of  airborne  laser  scanning  for
elevation measurement and airborne video for thematic data.
The laser scanning (see in the chapter about Airborne Laser Scanner) is to replace, in
due course, the traditional photogrammetric approach to obtain profiles of the beach
and  foredunes.  With  the  introduction  of  laser  scanning  techniques  and  video  the
Survey Department aims to improve efficiency and performance. As of 1996, a DTM
of the beach and foredunes of the Dutch coast will be provided yearly, based on laser
scanning data.
Also, once every three year the complete coastal zone (incorporating dunes) will be
surveyed. Video data will provide information on e.g. vegetative cover, development
and  migration  of  blow-  outs.  By  means  of  draping  the  video  images  over  the  laser-
DTM a good visual impression of the coastal area is obtained. With the provision of a
laser-DTM in combination with video as basic data sets the Survey Department aims
to contribute to integrated management of the coastal zone.170
8.8.2 The Southwest Washington Coastal Erosion Study, and the ARGUS
The  Southwest  Washington  Coastal  Erosion  Study  is  a  United-States  Federal-State-
Local  cooperative  research  program  to  address  the  coastal  geology,  processes,  and
natural hazards of the Southwest Washington coast.
Remote Video Monitoring (RVM) sytems provide a means of automatically acquiring
video data from remote locations and returning them to a central laboratory computer
for processing. Since 1991, the U.S. Geological Survey's Center for Coastal Geology
has  been  building  RVM  capabilities  through  a  cooperative  agreement  with  Oregon
State  Univerisity  where  video  data  acquisition  and  processing  tecniques  have  been
under development for 10 years.
Two  video  cameras  have  been  positioned  on  Willapa  Bay  near  North  Cove,
Washington to monitor coastal erosion and wave processes on the northern shore of
the  bay.  Video  images  are  taken  hourly  by  two  cameras.  The  data  are  down-loaded
daily  by  the  U.S.  Geological  Survey.  The  images  are  stored  in  the  archive  by  year,
day, camera, and hour.
This project began as a result of the ARGUS program, developed under the guidance
of  Professor  Rob  Holman  of  the  Coastal  Imaging  Lab,  Oregon  State  University.
Video-based monitoring stations are being placed at important coastal sites around
the world. The initial information that they have provided has proved as surprising as
the early satellite shots
Critical  to  progress  in  understanding  nearshore  dynamics  is  the  ability  to  make
physical  measurements  under  natural  conditions.  This  is  traditionally  accomplished
with surveys of beach response and with arrays of fixed-point measurements of fluid
motions. To carry out even a single survey of the underwater bathymetry of the beach
is  expensive  and  often  impossible  under  any  but  the  mildest  wave  conditions.  A
current meter for measuring waves and currents at a single point in space costs about
$7,500. To make sense of the measurements requires installing an array of at least ten
current meters and attempting to maintain them in a hostile environment. Manpower
costs are high and specialized equipment is usually needed. Thus, our knowledge of
the nearshore has been limited by the difficulties and cost of measurement.
In contrast, a single Argus station costs approximately $10,000. From a station, useful
data  can  be  collected  over  miles  of  beach  and,  in  time,  every  hour  for  years.
Stations  are  unmanned,  data  are  transferred  over  internet  and  maintenance  costs  are
virtually  zero.  While  a  measurement  taken  with  an  Argus  station  is  usually  not  as
accurate  as  that  from  in-situ  measurements,  the  collection  of  data  over  years  has
revealed characteristics about beaches that were never suspected.
The  key  part  of  an  Argus  station  is  one  or  more  cameras  pointed  obliquely  along  a
beach. The camera is connected to an image processor contained in a small personal
computer. An Argus station needs a home, with power, protection from the elements,171
and a phone line to communicate with the lab computers. Once installed, the station
begins  to  collect  images.  Currently,  stations  are  typically  programmed  tot  take  one
snapshot  and  one  time  exposure  image  every  hour.  This  occurs  automatically.  The
data are downloaded daily, usually during the middle of the night when no images
are being taken and phone rates are low. Once in the lab, the data are processed and
added to the growing database from different Argus sites. Support data from close-
by  tide  gages  and  wave-measuring  instruments  are  also  incorporated  into  the
Currently  Argus  stations  are  installed  in  the  U.S.,  Hawaii,  Australia,  New  Zealand,
England  and  the  Netherlands,  and  the  images  from  these  stations  are  available  at:
Beach camera, North of Noordwijk, the Netherlands
At  this  point  the  primary  use  of  Argus  data  lies  in  time  exposure  images.  By
averaging  the  pattern  of  wave  breaking  over  about  ten  minutes,  the  locations  of
submerged  sand  bars  are  clearly  seen.  From  each  image,  measurements  can  be
extracted  of  the  sand  bar  scales  and  the  behaviour  of  the  beach  system  can  be
compared to the varying wave and tide forces.
A  second  application  concerns  the  extraction  of  beach  contours.  From  the  time
exposure  images  the  10  minute-mean  shorebreak  or  waterline  can  be  located.  If  the
tide level is known, the contour of the beach at that level can be determined. By doing
this at different tide levels, a contour map of the beach can be composed.
The use of time exposures is only one of many techniques that are now or have been
developed  and  tested.  Pilot  studies  show  the  ability  to  measure  the  strength  of
longshore currents as well as the period and angle at which incident waves approach
the beach.172
9. Altimetry
Both  aircraft-  and  space-based  altimeters  yield  direct  measurements  of  elevations
along narrow path footprints. These are readily plotted as profiles; a series of parallel
profiles  serve  as  a  framework  for  contouring.  Altimeters  work  by  sending  self-
generated  (active)  signals  to  reflecting  surfaces  and  measuring  total  roundtrip  times
from  the  targets  (solid  land,  tree  tops,  ice,  or  water)  over  which  they  move.  The
signals can be either radio pulses (radar included) or light pulses (laser).
9.1 Laser Altimety
Laser is a shortened term for light amplification by stimulated emission of radiation.
A  typical  solid  laser  device  is  a  chromium-doped  ruby  (Al 2 O 3 ).  When  an  external
source  of  radiation  (as  from  an  enclosing  flash  tube)  acts  on  a  shaped  (usually
cylindrical) ruby crystal, one end of which is silvered to act as a mirror, the chromium
(Cr) ions dispersed in its lattice are excited to a new energy state (electrons raised to
some new orbital level), The laser state is generated by stimulated emission when the
electrons  drop  back  to  a  lower  energy  level  (remember,  from  the  Introduction,  the
formula:  E  =  hc/).  Much  of  the  light  passes  sidewards  out  of  the  crystal  but  light
moving along the axial zone of the cylinder encounters other Cr ions as it undergoes
repeated  reflections,  further  exciting  them  and  causing  a  buildup  or  amplification,
called optical pumping, of light energy until discharged as a pulse (controlled by the
flash  lamp)  of  intense  coherent  radiation  at  some  discrete  wavelength  (actually  a
narrow frequency range; for the ruby laser the light is pinkish) that is collimated to
form a unidirectional beam which can be aimed. Other laser materials include gallium
arsenide  and  excited  gases  such  as  neon  or  helium.  Both  visible  and  infra-red
wavelength light can be generated in this way.
Laser altimetry has been conducted from aircraft platforms for several decades. This
type  of  Remote  Sensing  is  sometimes  referred  as  lidar  (for  LIght  Detection  And
Ranging). Timing devices allow extremely precise determination of transit times, so
that accuracies of a few centimeters in determining elevations (and their differences or
relief)  along  the  traverse  are  attainable.  Aircraft  can  be  scheduled  to  fly  in  good
weather,  offsetting  the  main  disadvantage  of  using  lasers  coming  from  cloud
interference with the light beam. Lidar instruments can be operated both as profiliers
and as scanners, and can, of course, be operated both day and night. Lidar can serve
either as a ranging device to determine altitudes (topography mapping) or to conduct
particle analysis in air.
Laser  altimeters  are  beginning  to  be  flown  on  space  vehicles.  The  Orbital  Profiling
Laser  Altimeter  is  a  pulsed,  time-of-flight  optical  (1.024  µm)  sensor  that  sends  10
pulses per second (pps) in a narrow beam (footprint 30-150 m diameter; sampling in
150 to 700 m [492-2296 ft] intervals). Operated on the Shuttle Endeavor in January,
1996, it achieved a vertical precision of 0.75 m [29.5 inches]. Each laser shot fired has
a dwell time of only 2-10 nanoseconds, within the 1-10 nsec resolution of the timing
electronics;  with  this  rapid  return  rate,  ground  positions  are  readily  determinable
provided space vehicle position is known.173
Essentially  the  same  sensor  has  been  launched  as  the  Mars  Global  Altimeter  to  the
Red Planet at the end of 1996. A dedicated satellite (ALT) in the EOS, scheduled for
launch in the year 2003, will mount the Geoscience Laser Altimeter System (GLAS)
designed   to   measure   ice-sheet   topography   as   well   as   cloud   and   atmospheric
properties,  but  will  also  survey  selected  land  and  water  surfaces.  The  40  pulse/sec
beam  is  generated  from  a  neodymium:yttrium-aluminum-  garnet  (Nd:YAG)  crystal
producing a two level emergy output :1.064 µm (infrared) for surface surveys and a
0.532 µm signal for atmospheric (clouds and aerosols) measurements. Each pulse will
spread  over  a  ground  spot  of  70  meters  (230  ft),  with  separation  between  pulses
amounting  to  170  m  (558  ft).  From  its  705  km  (440  miles)  orbital  altitude,  the
instrument can measure height differences on the ice of 10 cm (4 inches) precision;
characteristics of the altered return pulse indicate surface roughness.
Lasers are also reliable and very accurate tools for determining distances to satellites
in orbit. Retroreflectors are small prisms (typically quartz) mounted on the spacecraft.
Even at their high orbital speeds, these spacecraft can be targeted to intercept a laser
beam pulse directed at them from Earth-based stations which is then bounced back by
the  reflectors.  Along  with  GPS,  this  is  one  way  in  which  the  orbital  position  of  a
satellite  can  be  precisely  fixed.  The  astronauts  in  later  Apollo  missions  left
retroreflectors  on  the  Moon's  surface  as  a  means  of  learning  more  about  its  orbital
motions, including its recession from Earth.
9.2 Radar Altimetry
From  space,  real  aperture  microwave  (radar)  altimeters  have  served  to  measure
various aspects of sea state, such as heights relative to the geoid, wave geometries, sea
ice, and, indirectly, circulation characteristics such as currents and eddies. Like lasers,
the signals are dispatched as short pulses, but at the longer radar wavelengths these
signals  can  penetrate  cloud  cover.  A  typical  radar  altimeter  operates  in  the  13.5
Gigahertz  (radio/microwave)  portion  of  the  Electromagnetic  Frequency  Spectrum.
Again, using roundtrip times the instrument acts as a ranger to determine distance to
target.  Radar  altimeters  designed  to  secure  data  from  the  ocean  surface  use  small
antennas that dispatch long wavelength pulses through a wide beam that produces a
broad  footprint  whose  size  is  determined  by  pulse  length.  This  pulse-limited  type
operates  best  on  smooth  surfaces  but  analysis  of  the  degree  of  "stretching"  of
backscattered pulses (echoes) yields information on surface wave heights (roughness).
For  land  measurements,  especially  on  surfaces  of  high  relief,  the  beam-limited
altimeter  requires  a  larger  antenna  (a  practical  limitation)  capable  of  generating  a
narrow beam (hence, smaller footprint that better discriminates changes in slope) and
shorter wavelengths. The orbital position of the sensor platform must be determined
with  high  precision  and  accuracy  to  establish  the  position  of  the  geoid  and  local
departures of the surface from it; modifications of the signal by the atmosphere need
to  be  accounted  for,  usually  through  corrections  made  from  data  acquired  by  an
accompanying radiometer.
The  first  spaceborne  altimeter  experiment  was  conducted  during  the  1973  Skylab
mission.  Next  was  GEOS-3  (Geodynamics  Experimental  Ocean  Satellite)  in  1975
which measured height differences greater than 60 cm. This was followed in 1978 by
JPL's Seasat (which failed after 99 days but in the interim returned extremely valuable
data  over  sea  and  land).  Its  altimeter,  one  of  five  instruments  including  a  SAR,
performed at 10 cm range precision. Among its remarkable products was a map of the
morphology of the ocean floor, reproduced here:174
The actual data source for this was the global map of sea surface elevation, which was
digitally adjusted to remove effects of the lunar and solar tides and further corrected
for bathymetric variations. What appears to be a topographic expression of the floor is
actually the distortion of the surface gravitational field, expressed as broad variations
in  sea  surface  heights,  as  controlled  by  subsurface  mass  variations  and  their
distribution under variable floor relief. Discernible are the many transform fault zones
in the ocean plates, along with spreading ridges, trenches, and seamounts.
9.3 Radar Altimetry over the oceans
§     Chelton, Dudley B., WOCE/NASA Altimeter Algorithm Workshop, U.S. WOCE Technical Report
No. 2, 70 pp., U.S. Planning Office for WOCE, College Station, TX 1988.
9.3.1 Measuring Ocean Topography for Understanding and Predicting Climate
Most of the heat stored in the earth's hydrosphere resides in the ocean. The upper 3
meters  of  the  ocean  contains  the  same  amount  of  heat  as  that  stored  in  the  entire
atmosphere.  This  enormous  reservoir  is  moved  around  the  world  by  ocean  currents,
affecting  the  earth's  climate  and  its  variability  in  a  profoundly  complicated  manner.
The   ocean   circulation   is   fundamentally   turbulent,   always   changing,   and   its
observation, especially on a global scale for addressing its climatic impact, has posed
a tantalizing challenge to oceanographers for a long time.
Surface geostrophic currents, strongly linked to the circulation of the deep ocean, can
be  determined  from  the  ocean  topography,  defined  as  the  height  of  ocean  surface
relative to one of the earth's equi-geopotential surfaces, called the geoid. In addition,
ocean  topography  directly  reflects  the  heat  content  of  the  ocean  and  its  changes.
Knowledge of the ocean topography is thus very useful to the determination of ocean
circulation and its heat transport.
The concept of using a spaceborne radar altimeter to measure ocean topography was
formulated  in  the  late  1960s.  The  concept  was  first  demonstrated  by  Seasat  (1978),
followed by Geosat (1985-89) and reached its current state-of-the-art via the Joint US/
France  TOPEX/POSEIDON  Mission  (1992-present).  The  measurement  principle  is
straightforward  but  the  challenge  is  to  reach  the  exceptionally  demanding  accuracy
and  precision  at  the  level  of  one  centimeter  for  adequately  determining  the  oceanic175
transport of mass, heat, freshwater, and chemicals, to which the earth's climate system
is extremely sensitive.
9.3.2 Data sources
Information about the different altimeters from which the altimetry data is gathered.
These include: Geosat Follow-On
The Geosat Follow-On satellite data is sent directly to the Payload Operation Center
(POC) at the Naval Oceanographic Office (NAVO). Orbit solutions are produced by
both  the  Naval  Satellite  Operations  Center  (NAVSOC)  and  at  the  Jet  Propulsion
Laboratory  (JPL)  based  on  the  Global  Positioning  System  receivers  onboard  the
satellite.  The  orbit  solutions  are  merged  with  the  satellite  data  and  additional
geophysical corrections (atmospheric pressure, tides, ...) at the Altimeter Data Fusion
Center (ADFC) in NAVO within 48 hours of measurement. The GFO satellite has an
exact repeat period of approximately 17 days. TOPEX/Poseidon
The  TOPEX/POSEIDON  mission  was  designed  to  provide  information  about  the
changing  topography  of  the  world's  oceans  which,  in  turn,  helps  scientists  to
understand  the  ocean's  role  in  the  global  climate.  The  TOPEX/POSEIDON  satellite
was launched in August 1992 and was expected to operate through September, 1998.
The satellite is currently in operation. TOPEX/POSEIDON measures the global ocean
topography every 10 days.
TOPEX/POSEIDON is jointly conducted by the United States' National Aeronautics
and  Space  Administration  (NASA)  and  the  French  Space  Agency,  Centre  National
d'Etudes Spatiales (CNES), for studying the global circulation from space.
The specific goals of the TOPEX/POSEIDON mission are:
1) Measure sea level in a way that allows the study of ocean dynamics, including the
calculation of the mean and variable surface geostrophic currents and the tides of the
world's oceans.
2) Process, verify and distribute the data in a timely manner, with other geophysical
data, to science investigators.
3) Lay the foundation for a continuing program to provide long-term observations of
the oceanic circulation and its variability. Mission requirements
To    ensure    that    science    and    mission    goals    are    accomplished    by    the
TOPEX/POSEIDON Mission, the following requirements were established:
Accuracy of Sea-level Measurements
Each measurement of sea level shall have a precision of +2.4 cm and an accuracy of
+14   cm   (1   standard   deviation)   for   typical   oceanic   conditions,   with   small
geographically correlated errors. In this context, precision is the ability to determine
changes in sea level over distances of 20 km, and accuracy is the uncertainty of each
measurement of sea level when expressed in geocentric coordinates.
Sampling Strategy
Sea level shall be measured along a fixed grid of subsatellite tracks such that it will be
possible  to  investigate  and  minimize  the  spatial  and  temporal  aliases  of  surface176
geostrophic currents and to minimize the influence of the geoid on measurements of
the time-varying topography.
Tidal Aliases
Sea level shall be measured such that tidal signals will not be aliased into semiannual,
annual,  or  zero  frequencies  (which  influences  the  calculation  of  the  permanent
circulation) or frequencies close to these.
Duration and coverage
Sea level shall be measured for a minimum of three years, with the potential to extend
this period for an additional two years. Sensors on board the TOPEX/POSEIDON
The  science  and  mission  goals  are  carried  out  with  a  satellite  carrying  six  science
instruments, four from NASA and two from CNES. They are divided into operational
and experimental sensors as follows:
(A) 4 operational sensors
(1) Dual-frequency Ku/C band Radar Altimeter (NRA) (NASA)
The NRA, operating at 13.6 GHz (Ku band) and 5.3 GHz (C band) simultaneously, is
the primary sensor for the TOPEX/POSEIDON mission. The measurements made at
the two frequencies are combined to obtain altimeter height of the satellite above the
sea (satellite range), the wind speed, wave height and the ionospheric correction. This
instrument  is  the  first  spaceborne  altimeter  that  uses  two-channel  measurements  to
compute the effect of ionospheric free electrons in the satellite range measurements. It
is redundant except for the microwave transmission unit and the antenna.
(2) Three-frequency TOPEX Microwave Radiometer (TMR) (NASA)
The  TMR  measures  the  sea  surface  microwave  brightness  temperatures  at  three
frequencies (18 GHz, 21 GHz and 37 GHz) to provide the total water-vapor content in
the troposphere along the altimeter beam. The 21 GHz channel is the primary channel
for  water-vapor  measurement.  It  is  redundant  (21A  and  21B).  The  18  GHz  and  37
GHz  channels  are  used  to  remove  the  effects  of  wind  speed  and  cloud  cover,
respectively  in  the  water-vapor  measurement.  TMR  data  are  sent  to  CNES  for
processing along with their altimeter data. The measurements are combined to obtain
the error in the satellite range measurements caused by pulse delay due to the water
(3) Laser Retroreflector Array (LRA) (NASA)
The LRA reflects signals from network of 10 to 15 satellite laser tracking stations to
calibrate NRA bias and to provide the baseline tracking data for NASA precise orbit
(4) Dual-frequency Doppler tracking system receiver (DORIS) (CNES)
The DORIS system uses a two-channels receiver (1401.25 MHz and 2036.25 MHz)
on  the  satellite  to  observe  the  Doppler  signals  from  a  network  of  40  to  50  ground
transmitting stations. It provides all-weather global tracking of the satellite for CNES
precise  orbit  determination  and  an  accurate  correction  for  the  influence  of  the
ionosphere on both the Doppler signal and altimeter signals.
(B) 2 experimental sensors
The two experimental instruments are intended to demonstrate new technology.
(1) Single frequency Ku band Solid State ALTimeter (SSALT) (CNES)
The SSALT, operating at a single frequency of 13.65 GHz (Ku band), will validate
the  technology  of  a  low-power,  light-weight  altimeter  for  future  Earth-observing
missions. It shares the antenna used by the NRA; thus only one altimeter operates at
any  given  time.  Measurements  give  the  same  geophysical  information  as  NRA's.177
However,  since  this  sensor  uses  a  single  frequency,  an  external  correction  for  the
ionosphere must be supplied.
(2) Global Positioning System Demonstration Receiver (GPSDR) (NASA)
The  GPSDR,  operating  at  1227.6  MHz  and  1575.4  MHz,  uses  a  new  technique  of
GPS differential ranging for precise orbit determination. Orbit of the TOPEX/POSEIDON
The  orbit  chosen  for  the  TOPEX/POSEIDON  mission  is  a  compromise  among  a
number of conflicting requirements. It provides broad coverage of the ice free oceans
as frequently as possible without aliasing the tides to unacceptable frequencies, and it
is high enough to ease the precision of the orbit determination process in minimizing
the atmospheric drag. The reference (equatorial) altitude of the satellite is 1,336 km,
the equatorial cross-track seperation being 315 km. Data retrieval
The  TOPEX  data  is  retrieved  from  a  NAVO  computer  set  up  in  the  TOPEX  data
stream at JPL. This computer  converts  the  raw  satellite  telemetry  data  into  physical
units. The data is then passed on to the ADFC at NAVO. Orbit solutions produced at
JPL are also passed on to the ADFC. At NAVO, additional data fields are added to the
satellite data. These include estimates of the atmospheric pressure. The orbit solutions
from JPL are merged with the altimeter data and the geophysical corrections, and the
sea surface height measurements are made available within 48 hours.
The POSEIDON instrument is a solid state altimeter built by the French space agency
CNES.  The  data  from  POSEIDON  are  processed  at  the  French  Processing  and
Archival Facility (PAF). Jason
Jason is an oceanography mission to monitor global ocean circulation, discover the tie
between the oceans and atmosphere, improve global climate predictions, and monitor
events such as El Niño conditions and ocean eddies.
The  Jason-1  satellite  carries  a  radar  altimeter  and  it  is  a  follow-on  mission  to  the
highly  successful  TOPEX/Poseidon  mission.  It  is  joint  mission  between  France  and
USA. The satellite will be launched in May 2000.
The  Sea  Level  measurement  accuracy  of  Jason  is  required  to  be  better  than  4.2cm,
with 2.5cm set as the goal. Data coverage will be between 66 deg N and 66 deg S. ERS-2
The  ERS-2  data  is  provided  by  the  European  Space  Agency  (ESA)  through  the
National Oceanic and Atmospheric Administration (NOAA) Laboratory for Satellite
Altimetry. Orbit solutions are provided by Delft University. The data are passed from
NOAA to the ADFC, and are available within 48 hours. The ERS-2 satellite has an
exact repeat period of approximately 35 days.
On-board systems of the ERS-2:
AMI  -  active  microwave  instrument  consisting  of  a  synthetic  aperture  radar  (SAR)
and a wind scatterometer (both in the C-band).
The purpose of the Wind Scatterometer is to obtain information on wind speed and
direction  at  the  sea  surface  for  incorporation  into  models,  global  statistics  and
climatological datasets. It operates by recording the change in radar reflectivity of the178
sea due to the perturbation of small ripples by the wind close to the surface. This is
possible  because  the  radar  backscatter  returned  to  the  satellite  is  modified  by  wind-
driven  ripples  on  the  ocean  surface  and,  since  the  energy  in  these  ripples  increases
with wind velocity, backscatter increases with wind velocity.
The  three  antennae  generate  adar  beams  looking  45deg.  forward,  sideways,  and
45deg.  backwards  with  respect  to  the  satellite's  flight  direction.  These  beams
continuously  illuminate  a  500  km  wide  swath  as  the  satellite  moves  along  its  orbit.
Thus  three  backscatter  measurements  of  each  grid  point  are  obtained  at  different
viewing  angles  and  separated  by  a  short  time  delay.  These  `triplets'  are  input  into  a
mathematical model to calculate surface wind speed and direction.
Spatial resolution: >=45 km (along and across track)
Localisation accuracy: +-5 km (along and across track)
Wind direction range/accuracy: 0 - 360deg. / +-20deg.
Wind speed range/accuracy: 4 m/s - 24 m/s / 2 m/s or 10 %
RA-  radar  altimeter:  takes  precise  measurements  of  the  distance  from  the  ocean
surface  and  of  wave  heights.  The  Radar  Altimeter  is  a  Ku-band  (13.8  GHz)  nadir-
pointing  active  microwave  sensor  designed  to  measure  the  time  return  echoes  from
ocean and ice surfaces. Functioning in one of two operational modes (ocean or ice)
the  Radar  Altimeter  provides  information  on  significant  wave  height;  surface  wind
speed;  sea  surface  elevation,  which  relates  to  ocean  currents,  the  surface  geoid  and
tides; and various parameters over sea ice and ice sheets.
The Radar Altimeter operates by timing the two-way delay for a short duration radio
frequency pulse, transmitted vertically downwards. The required level of accuracy in
range  measurement  (better  than  10  cm)  calls  for  a  pulse  compression  (chirp)
technique. In ocean mode a chirped pulse of 20 micro-s duration is generated with a
band width of 330 MHz. For tracking in ice mode an increased dynamic range is used,
obtained  by  reducing  the  chirp  bandwidth  by  a  factor  of  four  to  82.5  MHz,  though
resulting in a coarsr resolution.
ATSR  -  along-track  scanning  radiometer  (operating  in  the  infrared  and  visible
ranges): measures sea surface temperatures and the vegetation cover of land surfaces.
The  ATSR  consists  of  two  instruments,  an  Infra-Red  Radiometer  (IRR)  and  a
Microwave Sounder (MWS). Both are nationally funded experiments resulting from
an ESA Announcement of Opportunity for a scientific add-on package.
The IRR is a four-channel infra-red radiometer (at wavelengths of 1.6, 3.7, 10.8 and
12   mm)   used   for   measuring   sea-surface   temperatures   (SST)   and   cloud-top
temperatures  .  It  was  designed  and  constructed  by  a  consortium,  consisting  of
Rutherford  Appleton  Laboratory  ,  Oxford  University  ,  Mullard  Space  Science
Laboratory , UK Meteorological Office and CSIRO in Australia.
The MWS is a two channel passive radiometer (at 23.8 and 36.5 GHz) designed and
built under the responsibility of Centre de Recherche en Physique de l'Environnement
(CRPE). The MWS is physically attached to the IRR and its data is merged with that
of the IRR prior to transmission to the ground.
GOME  -  global  ozone  monitoring  experiment,  an  absorption  spectrometer  which
measures  the  presence  of  ozone,  trace  gases  and  aerosols  in  the  stratosphere  and
troposphere. GOME, a nadir-scanning ultraviolet and visible spectrometer for global
monitoring of atmospheric Ozone, was launched on-board ERS-2 in April 1995. Since
summer 1996, ESA has been delivering to users three-day GOME global observations
of  total  ozone,  nitrogen  dioxide  and  related  cloud  information,  via  CD-ROM  and
internet.  A  key  feature  of  GOME  is  its  ability  to  detect  other  chemically  active
atmospheric trace-gases as well as aerosol distribution.179
MS - microwave sounder: supplies data on atmospheric humidity.
PRARE  -  precise  range  and  range  rate  equipment:  ERS  orbit  and  trajectory
LRR - laser reflector: determines satellite position using ground-based laser stations.
IDHT - instrument data handling and transmission: temporary on-board data storage
by means of two 6.5 GBit tape recorders, equivalent to the data volume acquired in
one orbit. Recording, formatting and transmission of data at 105 Mbit/s (transmission
of SAR imaging data in real time), or 15 Mbit/s (transfer from tape recorder). NRL Layered Ocean Model
The  NRL  Layered  Ocean  Model  (NLOM)  is  a  primitive  equation  numerical  ocean
model  with  the  vertical  coordinate  system  provided  through  set  of  layers  with
differing densities. The horizontal resolution used here is 1/4 degree, though present
manifestations are run at 1/16 degree resolution globally. The bottom topography is
also  included.  The  model  is  implemented  at  the  Fleet  Numerical  Meteorology  and
Oceanography Center (FNMOC). The altimeter data produced daily at NRL from the
ADFC are sent to FNMOC where they are assimilated into the NLOM. The model is
forced by 6-hourly wind stress fields produced at FNMOC. Modular Ocean Data Assimilation System (MODAS)
MODAS performs quality checking and optimum interpolation of ocean observations,
including  temperature,  salinity  and  Sea  Surface  Height  (SSH).  The  SSH  and  SST
appearing  in  the  plots  were  gridded  by  optimum  interpolation  using  altimeter  and
AVHRR  MCSST  products  from  the  Naval  Oceanographic  Office.  The  covariance
functions used in the interpolations were derived from several years of satellite-based
SST and SSH observations. MODAS also computes synthetic three-dimensional grids
of  ocean  temperature  and  salinity  based  on  the  gridded  SST  and  SSH  using
climatological  relationships  between  subsurface  temperature  and  SST  and  SSH
derived  from  historical  profile  observations.  Salinity  is  computed  from  the  derived
temperature using local climatological relationships between temperature and salinity.
9.3.3 Data Processing
After data aquisition, the data from each altimeter is processed in a similar manner: Initial Data Processing
The  Data  processing  begins  with  the  Geophysical  Data  Records  (GDRs)  of  each
satellite. These are the data records distributed by the ADFC, and they contain all the
information needed to correct the satellite SSH measurement for atmospheric effects
as well as information to remove processes that are not of interest (inverse barometer
pressure loading, solid earth tides, and ocean tides).
The following geophysical corrections are made to the altimeter data:
•   dry troposphere path delay
•   wet troposphere path dely
•   ionosphere path delay
•   electromagnetic bias
•   static inverse barometer correction
Significant wave height and wind speed are also extracted from the GDRs.
The  data  are  quality  controlled  at  this  step  by  verifying  the  quality  flags,  the  RMS
variability   of   the   high   rate   measurements   about   the   one   sample   per   second
measurements,  detecting  exceptionally  high  AGC  quantities,  detecting  questionable180
water  vapor  corrections  based  on  onboard  instruments,  and  detecting  questionable
ionosphere corrections based on onboard instruments.
The sea surface height resulting from the GDRs is a measurement of the position of
the sea surface relative to the reference ellipsoid. Corrections are made to ERS-2 so
that the data are relative to the same reference ellipsoid as TOPEX and GFO.
A schematic summary of altimeter measurements and the corrections that must be applied to obtain the
dynamic sea surface elevation h d  . The altimeter range measurement is h, and H and h g  are the orbit
height  and  geoid  height,  respectively,  relative  to  a  reference  ellipsoid  approximation  to  the  earth’s
surface. Interpolation
The measurement positions of the altimeter change from one repeat pass to another.
Thus  the  SSH  data  must  be  interpolated  along  the  ground  tracks  to  fixed  reference
points.  These  reference  points  were  produced  by  the  University  of  Colorado  Center
for  Astrodynamics  Research  (CCAR).  The  points  are  derived  from  past  altimeter
missions and are spaced by one second intervals along ground track solutions.
A quality control is made prior to interpolation to detect gross outliers. First, for each
SSH measurement, other surrounding SSH values are used to interpolate to the point
using  a  second  order  quadratic.  The  RMS  difference  between  the  interpolated  and
actual  values  is  then  computed,  and  values  that  are  significantly  different  from  the
interpolated  value  are  removed.  The  remaining  SSH  are  then  interpolated  to  the  the
reference ground tracks. Tide removal
A global numerical model for the dynamic ocean tide, the tidal loading, and the long
period tides is then removed from the altimeter data. The tide solution is the Grenoble
95.2.1. We use this to insure a consistent tide model is removed from all data. One181
drawback  to  this  is  that  the  model  solution  is  provided  on  a  ¼  degree  grid.  Thus,
points near land do not have a tide correction made. This is not a problem for deep
water regions. Orbit error removal
One of the largest error sources associated with real time altimeter data is errors in the
orbit  solution.  These  errors  are  so  large  that  the  altimeter  data  usefulness  would  be
questionable if they were not removed. Fortunately, the structure of the orbit error is
very long wavelength. Thus if we remove the long wavelength component of the SSH
(40,000  km,  or  once  cycle  per  satellite  orbit  revolution),  we  remove  the  orbit  error.
However, there is significant annual variability that resembles this long wavelength.
The  out  of  phase  annual  heating  of  the  northern  and  southern  hemispheres  causes  a
large steric anomaly in one hemisphere of the opposite sign to the steric anomaly in
the other hemisphere. Removing this steric signal can produce errors in the resulting
interpretation of the satellite SSH.
To minimize the damage to the oceanography when removing the orbit error, we take
into  account  the  expected  seasonal  variations  in  SSH  based  on  climatological  data.
For  the  seasonal  dynamic  height  variations  we  use  the  Navy  Generalized  Digital
Environmental     Model     (GDEM).     This     model     based     on     climatological
bathythermograph  data  allows  us  to  maintain  the  seasonal  steric  signal  in  the  SSH
data  while  removing  orbit  errors.  The  orbit  error  removal  is  done  on  one  satellite
revolution's  worth  of  data  at  a  time  using  a  weighted  least  squares  technique.  The
weighting  is  based  on  the  variability  observed  by  prior  altimeter  missions.  This
minimizes the impact of large amplitude features in the areas dominated by mesoscale
variability. Referencing to a consistent mean
The most difficult problem in using multiple altimeter data sets is referencing all the
data  sets  to  the  same  mean  sea  level.  Each  satellite  measures  variability  about  the
mean sea level along its own ground tracks. From previous altimeter missions we can
produce  a  mean  sea  level  along  the  ground  tracks,  and  use  present  data  relative  to
these  prior  means.  However,  the  time  periods  of  previous  missions  do  not  overlap,
and significant changes in sea level occur due to ocean circulation. We use the time
period   from   Jan   1,   1993   through   Jan   1,   1998   (the   first   full   5   years   of
TOPEX/POSEIDON) as the time period to which we reference all the altimeter data.
Two  different  methods  are  used  to  reference  ERS-2  and  GFO  to  this  time  period.
Because there is an overlap between the ERS and TOPEX/POSEIDON satellites, the
TOPEX anomalies about the 5 year mean may be used in conjunction with the ERS
data. First the T/P SSH anomalies about the 5 year mean are interpolated in space in
time.  These  T/P  deviations  from  the  5-year  mean  are  interpolated  to  the  ERS
measurement positions at the exact measurement times and then subtracted from the
ERS data. This transforms the ERS data into measurements of the 5 year mean, and
the  average  of  these  data  minimizes  interpolation  errors.  The  resulting  ERS  mean
allows use of the real time ERS data to provide anomalies relative to the 5-year mean
time period.
GFO presents a much more difficult challenge. There is no time overlap between the
original Geosat-Exact Repeat Mission and other altimeters. The only place where the
SSH may be compared to present altimeters is at the points where the Geosat ground
track  crosses  other  ground  tracks  since  the  geoid  is  the  same  at  these  points.  By
examining the mean SSH from the Geosat-ERM and from T/P at the crossover points,182
we  can  determine  the  SSH  change  between  the  two  time  periods.  This  provides  a
correction to the original Geosat-ERM mean to allow the GFO data to be referenced
to the common 5 year mean.
9.3.4 Altimetry products and derived products
Through  signal  processing  and  interpretation,  researchers  have  identified  several
different  types  of  geophysical  parameters  which  can  be  derived  from  the  altimeter
data  provided  by  TOPEX/Poseidon.  These  same  parameters  can  then  be  directly
associated with various oceanographic products as detailed in the table below:
Measurement Geophysical parameter          Oceanographic product
Elapsed Time of Return Altitude Marine Geoid, Current Location
& Velocity and Ice Topography
Waveform Leading Edge Sea/Surface Height, Distribution
& Standard Deviation
Significant Wave Height
Waveform Leading Edge Skewness Dominant Wavelength,
Significant Slope and
Thermocline Information
Amplitude Backscatter Coefficient: Sigma
Sea/Ice Boundaries and Sea
Surface Windspeed
Trailing Edge Satellite Pointing and
Sea/Surface Tilt
Ice Slope
Figure  of  the  average  profile  of  an  Ocean  ReturnWaveform  (taken  from  ERS-2  site).  Over  ocean
surfaces, the distribution of the heights of reflecting facets is gaussian or near-gaussian, and the echo-
waveform has a characteristic shape that can be described anayltically, as a function of the standard
deviation of the distribution, which is closely related to the ocean wave height.
The key principle behind the altimeter is that the information required is in the shape
and timing of the returned radar pulse.
The figure shows a pulse being reflected from a flat surface. As the pulse advances,
the illuminated area grows rapidly from a point to a disk, as does the returned power.
Eventually,  an  annulus  is  formed  and  the  geometry  is  such  that  the  annulus  area
remains  constant  as  the  diameter  increases.  The  returned  signal  strength,  which
depends  on  the  reflecting  area,  grows  rapidly  until  the  annulus  is  formed,  remains
constant until the growing annulus reaches the edge of the radar beam, where it starts
to diminish.183
Irreguarities on the surface, larger than the pulse width, cause the returned pulse to be
distorted  and  stretched.  The  effect  of  this  is  to  impose  an  additional  slope  on  the
leading  edge  of  the  returned  signal  strength  curve  (  see  the  figure  ).  This  slope  is
related  to  the  ocean  wave  height  and  the  mid-point  of  this  leading  edge  slope  is
equivalent to the reflection from the average position of the surface (i.e. the mean sea
surface). By measuring the total area under the curve, the average reflectivity of the
surface may be obtained.
Schematic  representation  of  a  wide  beamwidth,  short  pulse  propagating  from  the  satellite  to  the  sea
surface  (upper  row).  The  antenna  footprint  on  the  sea  surface  is  shown  as  a  function  of  time  in  the
middle row. The area of the footprint is shown as a function of time in the bottom panel. For a calm sea
surface,  the  area  rise  time  is  equal  to  the  pulse  duration  t.  For  a  rough  sea  surface  with  significant
wave height H 1/3 , this rise time increases by amount 2c-1H 1/3 .
Clearly the altimeter requires a very short pulse width and full analysis of the pulse
shape must be carried out quickly since the pulse shown in the figure is so short. Both
of these time requirements have been avoided by translating from time to frequency,
using   a   technique   called   full-deramp   .   The   altimeter   therefore   operates   with
frequency-modulated  rather  than  time-modulated  signals  and  the  analysis  of  pulse
shape is a spectrum analysis , performed by the processor. The characteristics of this
processor and the control of the radar are varied according to the mode of the Radar
Altimeter. The most important is the acquisition mode, during which the radar finds
the  approximate  distance  to  the  surface  and  then  switches  to  one  of  the  tracking
modes - ocean or ice (as is done in ERS-2). Dynamic Sea Surface Topography - (from Altitude)
A  measure  of  sea  level  relative  to  earth's  geoid  from  which  oceanographers  can
calculate the speed and direction of winds. Two very precise distance measurements
must  be  established  in  order  to  acquire  reliable  topographical  ocean  surface  maps.
First, the height of the satellite above the reference ellipsoid is measured by tracking
the  satellite  in  orbit  from  a  globally-distributed  network  of  lasers  and/or  Doppler
stations.  The  trajectory  and  height  of  the  satellite  are  further  refined  by  using  orbit
dynamic equations. Second, the height of the satellite above the closest ocean surface184
is  measured  with  a  microwave  radar  altimeter.  The  difference  between  the  height
above   the   reference   ellipsoid   and   the   altitude   above   the   ocean   surface   is
approximately  equal  to  the  geoid  height.  Depicted  in  the  image  below*  is  the
difference between the geoid height and mean sea level, commonly referred to as sea
surface  topography.  This  sea  surface  height  has  two  main  components,  the  geoid
height, which reflects the gravitational field of the earth and the dynamic sea surface
height, which reflects the ocean currents and tides.
*Image is from Jet Propulsion Lab's Physical Oceanography Distributed Active Archive Center
Date of Image: January 20, 1996 Sea Surface Variability –
Illustration  of  Dynamic  Sea  Surface  Topography  via  time-dependent  anomalies
relative to a mean sea surface. Wind Speed - (from backscatter coefficient)
By measuring global sea-surface wind speed and direction, ocean scatterometer data
can  help  meteorologists  more  accurately  predict  the  marine  phenomena  that  affect
human life on a daily basis. This is important for weather forecasting, storm detection,
ship routing, safer end efficient off shore oil and gas drilling operations, and more.
The wind speed is measured by examining the radar signal generated by the satellite
and  reflected  from  the  ocean  surface.  The  wind  on  the  ocean  surface  generates
roughness that affects the scattering of the radar signal. The radar backscatter is used
with  laboratory  measurements  to  infer  the  wind  speed.  There  is  no  direction
associated with this speed (however, wind direction can be derived from scatterometer185
data  –  see  in  the  details  about  ERS-2  above).  A  calm  sea  is  a  good  reflector  and
returns  a  strong  pulse,  while  rough  seas  scatter  the  signal  thereby  returning  a  weak
pulse. In general there is a high degree of correlation between wind speed and wave
height. The amount of scattered power can be used to observe the sea-ice extent, the
boundaries of currents, oceanic wind speed, and the rate of accumulation of snow on
Surface winds in the Pacific Ocean observed by the NSCAT scatterometer during September 1997.
Zoom in on the central area of the figure above:186 Scatterometry
A scatterometer is a high frequency microwave radar designed specifically to measure
ocean near-surface wind speed and direction. As the wind blows over the ocean, the
surface  is  roughened  by  the  generation  of  cat's  paws  (centimeter  scale  capillary
waves).  These,  in  turn,  modify  the  surface  backscatter  (reflected  signal  or  echo)
Scatterometery  has  its  origin  in  early  radar  used  in  World  War  II.  Early  radar
measurements over oceans were corrupted by sea clutter (noise) and it was not known
at that time that the clutter was the radar response to the winds over the oceans. Radar
response was first related to wind in the late 1960's. The first scatterometer flew as
part  of  the  Skylab  missions  in  1973  and  1974,  demonstrating  that  spaceborne
scatterometers  were  indeed  feasible.  The  Seasat-A  Satellite  Scatterometer  (SASS)
operated  from  June  to  October  1978  and  proved  that  accurate  wind  velocity
measurements  could  be  made  from  space.  A  single-swath  scatterometer  flew  on  the
European Space Agency's Remote Sensing Satellite-1 (ERS-1) mission.
The NASA Scatterometer (NSCAT) which launched aboard Japan's ADEOS-Midori
Satellite in August, 1996, was the first dual-swath, Ku-band scatterometer to fly since
Seasat.  From  September  1996  when  the  instrument  was  first  turned  on,  until
premature termination of the mission due to satellite power loss in June 1997, NSCAT
performed  flawlessly  and  returned  a  continuous  stream  of  global  sea  surface  wind
vector  measurements.  Unprecedented  for  coverage,  resolution,  and  accuracy  in  the
determination  of  ocean  wind  speed  and  direction,  NSCAT  data  has  already  been
applied  to  a  wide  variety  of  scientific  and  operational  problems.  These  applications
include such diverse areas as weather forecasting and the estimation of tropical rain
forest  reduction.  Because  of  the  success  of  the  short-lived  NSCAT  mission,  future
Ku-band  scatterometer  instruments  are  now  greatly  anticipated  by  the  ocean  winds
user community. The NSCAT mission proved so successful, that plans for a follow-on
mission  have  been  accelerated  to  minimize  the  gap  in  the  scatterometer  wind
Winds  over  the  ocean  modulate  air-sea  changes  in  heat,  moisture,  gases  and
particulates  (matter  in  the  form  of  small  liquid  or  solid  particles),  regulating  the
crucial bond between atmosphere and ocean that establishes and maintains global and
regional  weather  and  climate.  Data  derived  from  ocean  scatterometers  is  vital  to
researchers in the their studies of air-sea interaction and ocean circulation, and their
effects  on  weather  patterns  and  global  climate.  In  the  past,  weather  data  could  be
acquired over land, but our only knowledge of surface winds over oceans came from
infrequent, and sometimes inaccurate, reports from ships and buoys. This data are also
useful  in  the  study  of  unusual  weather  phenomena  such  as  El  Niño,  the  long-term
effects of deforestation on our rain forests, and changes in the sea-ice masses around
the polar regions. These all play a central role in regulating global climate.
Wind  affects  the  full  range  of  oceanic  motion  –  from  individual  surface  waves  to
complete current systems. The tropical Pacific Ocean and overlying atmosphere react
to,  and  influence  each  other.  Easterly  surface  winds  along  the  equator  control  the
amount  and  temperature  of  the  water  that  upwells  (moves  or  flows  upward)  to  the
surface. This upwelling of cold water determines sea-surface temperature distribution,187
which affects rainfall distribution. This in turn determines the strength of the easterly
winds – a continuous cycle. Ocean circulation
§     Marth,   Paul   C.,   et   al.,   Prelaunch    Performance    of    the    NASA    Altimeter    for    the
TOPEX/Poseidon    Project,    IEEE    Transactions    on    Geoscience    and    Remote    Sensing,
Volume 31, Number 2, Pages 315-332, March 1993.
The  way  in  which  global  mean  sea  level  is  related  to  ocean  circulation  is  best
described by Marth, et al (1993): “Movement of water in the sea on spatial scales
exceeding  ~30  km  and  temporal  scales  exceeding  ~1  day  are  manifested  by
deflections of the sea surface; that is, by changes in mean sea level associated with
the strength and direction of the flow. This change in topography varies from the 1
meter  increase  in  mean  sea  level  in  100  km  over  the  Gulf  Stream  to  a  10  cm
change  over  1000  km  over  an  El  Nino  event  in  the  tropical  Pacific.  Given  the
strength of the sea surface topography signature, one can infer the magnitude and
direction of the oceanic water movement.”
Ocean currents are mapped by studying the "hills" and "valleys" in maps of the height
of  the  sea  surface  relative  to  the  geoid.  This  height  is  called  "Dynamic  Ocean
Topography." Currents move around ocean dynamic topography "hills" and "valleys"
in a predictable way. Note that a clockwise sense of rotation is found around "hills" in
the Northern Hemisphere and "valleys" in the Southern Hemisphere. This is because
of  the  "Coriolis  effect."  Conversely,  a  counterclockwise  sense  of  rotation  is  found
around "valleys" in the Northern Hemisphere and "hills" in the Southern Hemisphere.
In  general,  our  major  ocean  currents  are  stable  and  so  maps  of  dynamic  ocean
topography change very little over time. Before satellites, it took many years of ship
data to create a map of global oceans. Now, an updated map of ocean circulation is
available every 10 days.
The following figure shows altimeter field heights of the California current:188
The  altimeter  height  fields  show  the  poleward  Davidson  Current  in  winter  and  the  equatorward
California Current (CC) jet that begins near the coast in spring and moves steadily offshore until fall.
Shading shows that water from the west wind drift only affects the region offshore of the CC jet core
during most of the year, except in the Southern California Bight. Significant Wave Height - (from Wave Form Leading Edge)
Determined from the shape of the return pulse of the radar altimeter; a calm sea with
low  waves  returns  a  condensed  pulse,  while  rough  seas  with  high  waves  return  a
stretched pulse. Watervapor
Watervapor  in  the  atmosphere  delays  the  return  of  the  radar  altimeter  pulse  and
thereby  produces  a  false  reading  of  sea  level.  Onboard  microwave  radiometers
measure the content of water vapor and allow for adjustments.189 Marine gravity and sea-floor topography
§     Chelton, Dudley B., WOCE/NASA Altimeter Algorithm Workshop, U.S. WOCE Technical Report
No. 2, 70 pp., U.S. Planning Office for WOCE, College Station, TX 1988.
§     Global Bathymetric Prediction for Ocean Modelling and Marine Geophysics, David T. Sandwell
and Walter H. F. Smith
§     David T. Sandwell, Walter H. F. Smith, Chapter 11 - BATHYMETRIC ESTIMATION, (To appear
in Satellite Altimetry and Earth Sciences, Academic Press, 1999)
Variations  in  gravitational  acceleration  over  the  earth’s  surface  result  in  an  uneven
distribution  of  water  mass  in  the  oceans.  There  is  a  latitudinal  variation  associated
with  the  oblateness  of  the  earth.  In  addition,  there  are  gravity  anomalies  associated
with topographic features on the earth’s surface. The gravitational acceleration at the
sea surface is slightly stronger over bumps on the ocean bottom and slightly weaker
over  depressions  in  the  bathymetry.  In  the  absence  of  other  forcing  (e.g.,  pressure
gradients,  wind  forcing,  or  tides),  the  sea  surface  would  be  a  surface  of  constant
gravitational potential (the marine geoid).
Schematic  diagram  of  a  bump  and  a  depression  on  the  ocean  bottom  and  the  corresponding  marine
geoid.  The  vectors  indicate  the  gravitational  acceleration  along  the  geoid.  The  gravitational
acceleration  is  locally  deflected  toward  the  bump  and  away  from  the  depression  and  is  tangentially
perpendicular to the geoid.
To first order then, the marine geoid is a low-pass filtered image of the bathymetry.
Global  estimates  of  the  marine  geoid  are  obtained  from  combined  satellite  tracking
data  and  ship-based  gravity  measurements.  The  long  wavelength  components  are
obtained from measurements of the perturbed motion of near-earth satellites and short
wavelength components are determined from ship-based gravity measurements. These
gravity measurements are unevenly distributed globally, so there is large uncertainty
of approximately 1 m in the global geoid. Except in limited geographical regions of
densely-sampled shipboard gravity surveys (e.g., the northwest Atlantic), uncertainty
in the geoid height h g  is the largest source of error in altimeter estimates of absolute
dynamic sea surface topography h d  . Globally, the height h g  of the marine geoid has a
range of about ±100 m about a reference ellipsoid approximation to the surface of the
earth.  By  comparison,  the  sea  surface  topography  h  d  from  dynamic  ocean  currents
has a range of at most a few meters globally. Thus, to a very close approximation, the
mean sea surface measured by an altimeter coincides with the marine geoid. Because
the  marine  geoid  is  a  low-pass  filtered  version  of  the  bathymetry  (as  noted  above),
altimetric mean sea surfaces closely resemble bathymetry maps.190 Global Bathymetric Prediction for Ocean Modelling and Marine Geophysics
David T. Sandwell and Walter H. F. Smith report about the following work:
We  are  constructing  a  complete  bathymetric  map  of  the  oceans  at  a  3-10  km
resolution by combining all of the available depth soundings collected over the past
30  years  with  high  resolution  marine  gravity  information  provided  by  the  Geosat,
ERS-1/2,   and   Topex/Poseidon   altimeters.   Detailed   bathymetry   is   essential   for
understanding physical oceanography and marine geophysics. Currents and tides are
controlled by the overall shapes of the ocean basins as well as the smaller sharp ocean
ridges  and  seamounts.  Because  erosion  rates  are  low  in  the  deep  oceans,  detailed
bathymetry   reveals   the   mantle   convection   patterns,   the   plate   boundaries,   the
cooling/subsidence   of   the   oceanic   lithosphere,   the   oceanic   plateaus,   and   the
distribution  of  off-ridge  volcanoes.  Current  global  digital  bathymetry  maps  (e.g.
ETOPO-5)  lack  many  important  details  such  as  a  400  km-long  ridge  that  rises  to
within 135 m of sea level in the sub-antarctic front. Moreover, they are contaminated
by  long-wavelength  errors  (~2000  km)  which  prevent  accurate  identification  of
seafloor swells associated with mantle plumes.
The tasks include:
•   Accumulate  all  available  depth  soundings  collected  over  the  past  30  years.
(funded by NSF)
•   Use  the  short  wavelength  (<  160  km)  satellite  gravity  information  to
interpolate between sparse ship soundings.
•   Improve  the  resolution  of  the  marine  gravity  field  using  enhanced  estimates
along repeat altimeter profiles together with the dense altimeter measurements.
•   Refine/Improve bathymetric predictions using the improved resolution gravity
field  and  also  by  investigating  computer-intensive  methods  for  bathymetric
prediction such as inverse theory.
•   Produce a Globe of the Earth  similar  to  the  globe  of  Venus  prepared  by  the
NASA Magellan investigation. This will also include the best available digital
land data.
In  the  wavelength  band  15  to  200  km,  variations  in  gravity  anomaly  are  highly
correlated  with  seafloor  topography.  Since  many  southern  ocean  areas  and  some
northern  ocean  areas  are  sparsely  surveyed,  these  new  satellite  altimeter  data  reveal
many  previously  unsurveyed  features  such  as  ridge  axes,  seamounts  and  fracture
Fundamental limitations of topographic recovery from gravity anomaly measurements are illustrated
by seamounts (left) and plateau (right). The gravity signatures of the closely-spaced seamounts (4 km
apart and 1 km tall) are strong and distinct when the average ocean depth is 2 km or less but their
signatures  combine  and  become  weak  when  the  ocean  depth  is  4  km.  The  isostatically  compensated
step in depth produces a local gravitational edge effect that is strongly attenuated at a distance of 150
km from the step; thus the gravity far from the step does not provide information on the overall depth
offset across the step.
The conceptual approach is to use the sparse depth soundings to constrain the long-
wavelength  depth  while  the  shorter-wavelength  topography  is  predicted  from  the
downward-continued satellite gravity measurement. Over the short wavelength band,
the topography/gravity ratio is regionally calibrated using available soundings.192
Data quality is the most important aspect of bathymetric prediction. High resolution
satellite gravity data is needed not only to interpolate among the sparse soundings but
also to identify which soundings are bad. Many soundings from remote areas of the
oceans  were  collected  before  shipboard  computers  and  satellite  navigation  were
available.  Moreover,  all  of  the  ship  sounding  data  were  collected  over  a  30  year
period  on  a  variety  of  ships  from  many  countries  and  with  many  different  chief
scientists  aboard.  Thus  the  quality  of  the  data  is  highly  variable  and  many  entire
cruises  or  sections  of  cruises  are  bad;  only  the  most  recent  (since  ~1987)  GPS-
navigated  multibeam  data  are  reliable.  Typical  errors  include:  navigation  errors,
digitizing errors, typographical errors due to hand entry of older sounding, reporting
the  data  in  fathoms  instead  of  meters,  incorrect  sound  velocity  measurements  and
even computer errors in reading punch cards One bad section of a cruise in an isolated
region will introduce a seafloor topographic feature that does not exist.
The  high  resolution  gravity  fields  provides  the  information  needed  to  assess  the
accuracy  of  the  ship  sounding  data.  Our  approach  is  to  identify  the  bad  cruises
through  a  comparison  with  an  initial  prediction  based  on  the  gravity  and  either
eliminate  them  or  attempt  to  fix  problem  areas  (data  rescue);  rescue  is  especially
important for soundings that fill a large data gap.
Data preparation and assembly is an ongoing process; the current data are sufficiently
good to construct a global bathymetric grid. Here is one recipe (Nettleton's Method)
that we are developing.
NETTLETON'S METHOD (to understand the concepts mentioned below, review the
chapter about filters):
1) Grid available bathymetric soundings on a 2 minute Mercator grid that matches our
gravity  anomaly  grid.  To  avoid  seams,  all  work  is  done  on  a  global  grid  between
latitudes of +72deg. Coastline points from GMT provide the zero-depth estimates. A
finite-difference,  minimum-curvature  routine  is  used  to  interpolate  the  global  grid.
This gridding program requires at least 256 Mbytes of computer memory.
2) Separate the grid into low-pass and high-pass components using a Gaussian filter
(0.5  gain  at  160  km).  Filtering  and  downward  continuation  are  performed  with  a
multiple  strip,  2-D  FFT  that  spans  0-360deg  longitude  to  avoid  Greenwich  edge
3) Form high-pass filtered gravity using the same Gaussian filter.
4)  Downward  continue  the  high-pass  filtered  gravity  to  the  low-pass  filtered
bathymetry  assuming  Laplace's  equation  is  appropriate.  A  depth-dependent  Wiener
filter is used to stabilize the downward continuation.
5)    Accumulate    high-pass    filtered    soundings    and    corresponding    high-pass
filtered/downward-continued  gravity  into  small  (160  km)  overlapping  areas  and
perform  a  robust  regression  analysis.  In  sediment-free  areas,  the  topography/gravity
transfer  function  should  be  flat  and  equal  to  1/2[pi]G[Delta][rho]  so  in  the  space
domain, a linear regression is appropriate. This works well on young seafloor but not
on  old  seafloor  where  sediment  cover  destroys  the  correlation  between  topography193
and   gravity.   In   these   cases   we   assume   the   seafloor   is   flat   and   set   the
topography/gravity   ratio   to   zero.   Finally   there   are   intermediate   cases   where
topographic  depressions  will  be  sediment  filled  while  the  highs  protrude  above  the
sediments  so  the  topography/gravity  relationship  is  non-linear.  It  is  these  partially
sedimented  areas  that  make  the  bathymetric  problem  difficult  and  inherently  non-
linear. Continental margins and shelves pose similar problems.
6)  Regional  topography/gravity  ratio  estimates  are  gridded  and  multiplied  by  the
high-pass  filtered/downward-continued  gravity  to  form  high-pass  filtered  predicted
7)  The  total  predicted  bathymetry  is  equal  to  the  sum  of  the  high-pass  filtered
predicted bathymetry and the low-pass filtered bathymetry.
8) Finally, the pixels constrained by ship soundings or coastline data are reset to the
measured  values  and  the  finite-difference,  minimum  curvature  routine  is  used  to
perturb the predicted values toward the measured values. Measured depths are flagged
so they can be extracted separately. This final step dramatically increases the accuracy
and resolution of the bathymetric grid in well surveyed areas so it agrees with the best
hand-contoured bathymetric charts.
Comparison of measured depth along ship track (middle) with ETOPO-5 (upper) and our predicted depth (lower).
These ship data were acquired after the V6.2 of the bathymetric grid was completed.
We have assessed the accuracy of the prediction through a comparison with soundings
from  a  recent  cruise  to  the  Foundation  Seamount  Chain  in  the  South  Pacific.  This
poorly  charted  area  contains  a  1600-km  long  volcanic  chain  as  well  as  topography
associated  with  microplate  tectonics.  Based  on  depth  predictions  from  our  earlier
study  [1994],  an  initial  mapping  and  sampling  expedition  was  carried  out  in  1995
aboard the R/V Sonne where they charted 44 volcanoes with height ranging from 1500194
m to 4000 m; eleven of the uncharted volcanoes come to within 500 m of the ocean
surface.  These  Sonne-100  sounding  data  were  included  in  our  global  seafloor
topography  map  and  provide  good  definition  of  the  summits  of  the  seamounts.  In
January  and  February  of  1997,  one  of  us  (Sandwell)  participated  in  a  second
expedition to the Foundation Seamounts area aboard R/V L'Atalante. The cruise track
covers very high relief topography areas that were not surveyed during the Sonne-100
cruise and thus offers an excellent test of the accuracy of the predicted seafloor depth.
The  results  are  shown  in  above  figure  where  the  center  beam  of  the  Simrad  12D
multibeam  echo  sounder  is  plotted  versus  distance  from  Tahiti  (center  profile).  A
number of large seamounts and ridges were surveyed between distances of 2500 and
4000 km while a 6500-m deep trench was surveyed at a distance of 5800 km. These
measurements  are  compared  with  depths  sampled  from  the  ETOPO-5  grid  (upper
profile)  and  our  version  6.2  predicted  depths  (lower  profile).  The  depths  from
ETOPO-5 show a poor correlation with the measured depths and have an rms misfit
of 590 m in an area where the rms signal is large 747 m. The poor fit is due to a lack
of  ship  soundings  in  the  area.  The  prediction  offers  a  much  better  match  to  the
observed  depths  (259  m  rms)  because  the  high-resolution  gravity  field  information
from   the   Geosat   and   ERS-1   satellite   altimeters   provides   most   of   the   depth
information.  In  other  well  charted  areas  of  the  northern  oceans,  the  grid  is  more
tightly  constrained  by  ship  soundings.  On  average,  12%  of  the  grid  cells  are
constrained by ship measurements while the remaining 88% are estimated.
The global map of sea floor topography can be accessed in the following site:
9.4 Airborne Laser Scanning (ALS):
§     Crandall Clifford J. (1976), Coastal Aerial Photo-Laser Survey (CAPS) – A Near Shore Charting
System, International Hydroraphic Review, LIII (1),  pp. 53-64
§     Nairn  R.  (1994),  Royal  Australian  Navy  Laser  Airborne  Depth  Sounder  –  the  first  year  of
operations, International Hydrographic Review, LXXI (1), pp. 109-119
§     Koppari Kurt, Karlsson Ulf and Steinvall Ove (1994), Airborne Laser Depth Sounding in Sweden,
International Hydrographic Review, LXXI (2), pp. 69-90
§     Aloysius  Wehr,  Uwe  Lohr,  Airborne  laser  scanning—an  introduction  and  overview,  ISPRS  Journal  of
Photogrammetry & Remote Sensing 54 1999 68–82
§     Emmanuel  P.  Baltsavias,  A  comparison  between  photogrammetry  and  laser  scanning,  ISPRS  Journal  of
Photogrammetry & Remote Sensing 54 1999 83–94
§     Jennifer L. Irish, W. Jeff Lillycrop, Scanning laser mapping of the coastal zone: the SHOALS system, ISPRS
Journal of Photogrammetry & Remote Sensing 54 1999 123–129
§     Peter  Axelsson,  Processing  of  laser  scanner  data  –  algorithms  and  applications,  ISPRS  Journal  of
Photogrammetry & Remote Sensing 54 1999 138-147
§     IHO standards for hydrographic surveys, 4 th  edition, April 1998, Special Publication no. 44
European Naval Oceanography, by David B. St.Pierre
§     Axelsson Rune and Alfredsson Mats, Capacity and Capability for Hydrographic Missions
of Commerce National Oceanic and Atmospheric Administration Coastal Services Center, Coastal
Services Center Technical Report CSC/9-98/001 Version 1.0 September 1998
9.4.1 Introduction – laser principles
Laser scanners utilise opto-mechanical scanning assemblies just as many multispectral
scanners  like  the  spaceborne  scanner  on  Skylab,  the  LANDSAT  MSS  and  the
Thematic Mapper TM. However, they are active sensing systems using a laser beam
as  the  sensing  carrier.  Therefore,  two  optical  beams  -  All  laser  systems  measure  by
some means the distance between the sensor and the illuminated spot on ground.
The  narrow  divergence  of  the  laser  beam  defines  the  instantaneous  field  of  view
IFOV. Typically, the IFOV ranges from 0.3 mrad to 2 mrad. The theoretical physical
limit of the IFOV is determined by diffraction of light, which causes image blurring.
Therefore, the IFOV is a function of the transmitting aperture D and the wavelength
of the laser light l. For spatially coherent light, the diffraction-limited IFOV is given
IFOV diff  = 2.44 * l / D
The IFOV of the receiving optics must not be smaller than that of the transmitted laser beam.
Due to the very narrow IFOV of the laser, the optical beam has to be moved across the flight
direction  in  order  to  obtain  area  coverage  required  for  surveying.  The  second  dimension  is
realised by the forward motion of the airplane. Thus, laser scanning means deflecting a ranging
beam in a certain pattern so that an object surface is sampled with a high point density. Pulse lasers
Laser  light  is  very  special.  The  acronym  laser  stands  for  Light  Amplification  by
Stimulated Emission of Radiation. In current ranging laser systems, mostly pulsed lasers are
The  most  direct  ranging  measurement  is  determining  the  time-of-flight  of  a  light
pulse, i.e., by measuring the travelling time between the emitted and received pulse.
According to the following Figure, the travelling time (t L )of a light pulse is:
t L  = 2R/c
with R the distance between the ranging unit and the object surface and c the speed of
Figure: measuring principles of pulse lasers. The upper figure shows the transmitted signal, the lower
the received one. A T  and A R  are the amplitudes of the transmitted and received signal, respectively. Wavelength
In current ranging ALS, semiconductor diode lasers and Nd:YAG lasers pumped by
semiconductor lasers are used, covering the optical band between 800 nm and 1600
nm.  As  current  laser  scanners  work  by  the  direct  energy  detection  principle,  a
coherent  light  source  for  ranging  is  not  required.  Therefore,  the  following  physical
laser properties are used in laser scanning: high power, short pulses, high collimation,
and narrow optical spectrum in which lasers emit. The narrow optical spectrum or in
other  words,  the  narrow  spectral  laser  line  is  advantageous,  because  narrow  optical
interference filters (usually with 10 nm bandwidth) can be mounted in the receiving
path  to  suppress  disturbing  background  radiation,  caused  e.g.,  by  backscattered
The selection of the optical wavelength of the laser is dependent on the overall laser
scanning system design. The most sensitive detectors are available between 800 nm
and 1000 nm. Therefore, the first laser scanners worked with a wavelength of 900 nm.
However, at this wavelength of the optical spectrum, eye safety is still a concern. If
higher laser pulse energy is required, wavelengths have to be considered in which the
eye is less sensitive.
When  discussing  laser  wavelength,  one  should  also  consider  the  backscattering
properties of the target, e.g., the object surface. The reflectivity of a target, for a given
wavelength,  also  influences  the  maximum  range.  The  following  table  shows  the
typical reflectivity of various materials for 900 nm wavelength:
Material  Reflectivity %
Snow  80–90
White masonry  85
Limestone, clay  Up to 75
Deciduous trees  Typ. 60197
Coniferous trees  Typ. 30
Carbonate sand dry  57
Carbonate sand wet  41
Beach sands, bare areas in dessert  Typ. 50
Concrete, smooth  24
Asphalt with pebbles  17
Lava  8
Black neoprene synthetic rubber  5 Scanning
For a given flying height above ground h, the laser footprint mainly depends on the
divergence of the laser beam g, and the swath width on the scan angle Q, which is also
called the FOV.
Scans  can  be  uni-  or  bidirectional.  Typical  scanning  mechanisms  that  are  used  for
airborne  surveying  are  shown  in  the  following  figure.  Oscillating  mirrors  usually
produce   a   zigzag-line   bidirectional   scan   or   with   two-axis   galvanometers   a
bidirectional  meander-type  scan  of  parallel  lines  or  arcs,  rotating  polygon  and
multifaceted  mirror  scanners  produce  parallel  lines  (unidirectional  scan)  ,  nutating
mirrors (Palmer scan) produce an elliptical pattern and the fiber scanner produces a
parallel line scan. The scan pattern on the ground depends not only on the laser scan
pattern but also the flying direction and speed and the terrain topography. The points
along a line are usually scanned in equal angle steps, i.e., their spacing on the ground
is not constant. Due to acceleration or slow down of the scan mechanism, the points at
the swath borders exhibit other characteristics and are sometimes removed from the
raw data set.
Figure: Scanning mechanisms from top left, clockwise : oscillating mirror, Palmer scan, fiber scanner, rotating
polygon. Position and orientation system
The laser scanner measures only the line-of-site vector from the laser scanner aperture
to a point on the earth surface. The 3-D position of this point can only be computed, if
at any time, the position and orientation of the laser system is known with respect to a
coordinate  system,  e.g.,  WGS84.  Thus,  to  obtain  accurate  range  measurements  in  a
given coordinate system, a laser scanner system must be supported by a POS. As laser
scanners  have  a  potential  range  accuracy  of  better  than  1  dm,  POS  should  allow  at
least the same accuracy. Such an accuracy can be achieved only by an integrated POS198
consisting of a DGPS and an IMU. Geocoding of laser scanner measurements requires
an exact synchronisation of all systems: IMU, DGPS and laser scanner data. Typical processing steps
After a surveying flight, basically two data sets are available: the POS data and the
laser  ranges  with  the  instantaneous  scanning  angles.  Assuming  that  the  accuracy  of
POS data is better than 1 dm in position and 0.02º in orientation, already very precise
laser  measurement  points  in  an  earth-fixed  coordinate  system  can  be  calculated.
However,  some  systematic  parameters  must  be  considered.  These  are  e.g.,  the  three
mounting angles of the laser scanner frame, described by the Euler angles roll, pitch
and yaw, with respect to the platform-fixed coordinate system (usually with origin at
the IMU), the position of the laser scanner with respect to the IMU, and the position
of  the  IMU  with  respect  to  the  GPS.  This  so-called  calibration  data  can  be  derived
from  laser  scanner  surveys,  whereby  certain  reference  areas  are  flown-over  in
different directions. Reference areas are e.g., flat terrain like large sport fields or stadi-
ums, buildings and corner of buildings. From the relative orientation and position of
the  different  surveys  and  their  absolute  orientation  and  position  with  respect  to  an
earth-fixed coordinate system, calibration data can be derived.
The  main  objective  of  the  further  processing  is  the  calculation  of  digital  elevation
models. First, the laser range data must be transformed from WGS84 to the desired
local map coordinate system. The result is a cloud of randomly distributed laser points
in elevation and position. The distribution of the measurement points depends on the
scanning pattern of the laser scanner system.
Now, the elevation measurements are sorted with respect to their position. After the
sorting  process,  ground  points  must  be  separated  from  non-ground  points  like
buildings and vegetation. By classifying the different points, one can create a Digital
Terrain Model - DTM, from the Digital Surface Model – DSM (all visible object top
surfaces), obtained by the laser. For this task, different filter algorithms and procedures are
Since  the  generated  data  amount  is  very  high  for  certain  applications  and  commercial
programmes  for  DTM  interpolation,  visualisation,  etc.,  can  handle  only  a  limited  number  of
points, a  thin-out is often applied. This is (or at least should be performed after filtering and
DTM interpolation. Some extended laser capabilities
Apart  from  measuring  ranges,  some  lasers  can  also  record  the  intensity  of  the  backscattered
laser light (with intensity measured as the maximum of the returned pulse, or signal integration
over the returned pulse width). The intensity can be used for visualisation of the scene
but also to improve filtering r removal and classification
/  separation  of  objects  in  combination  with  the  range  and  other  information,  if
Some lasers can also record multiple echoes for each pulse sent, i.e., the range (and
possibly  intensity)  of  various  objects  along  the  pulse  path  and  within  the  laser
footprint. Such systems usually record both the first and last echo of a pulse, although
there is one commercial system that can record up to four echoes per pulse. There are
also  experimental  systems  that  record  the  whole  form  of  the  returned  signal  (called
waveform)  with  waveform  digitisers  sampling  the  incoming  signal  with  very  high
frequency, e.g., 250 MHz. Lasers that record only one echo, do so for the first or last
one  (some  systems  allow  switching  to  first  or  last,  depending  on  the  application).199
Recording  of  multiple  echoes  can  be  useful,  when  the  vertical  profile  of  multiple
objects  within  the  laser  footprint  (or  at  least  the  highest  and  lowest  object),  as  for
example, in the case of trees, is needed. Finally, there are bathymetric lasers, that are
based on the same principles as the topographic lasers, but emit in two wavelengths,
usually  1064  nm  and  532  nm.  The  infrared  wave-length  is  reflected  on  the  water
surface,  while  the  green  one  penetrates  the  water  and  is  reflected  by  the  bottom
surface or other objects in the water. A short overview of applications
Some of the major applications of ALS include:
•   Mapping of corridors, e.g., roads, railway tracks, pipelines, waterway landscapes
•   Mapping  of  electrical  transmission  lines  and  towers  including  ground  /  tree
•   DTM  generation,  especially  in  forested  areas  in  forests  also  for  road  and  path
planning, study of drainage patterns, etc.
•   Measurement  of  coastal  areas,  including  dunes  and  tidal  flats,  determination  of
coastal change and erosion
•   High  accuracy  and  very  dense  measurement  applications,  e.g.,  flood  mapping,
DTM  generation  and  volume  calculation  in  open  pit  mines,  road  design  and
•   DTM  and  DSM  generation  in  urban  areas,  automated  building  extraction,
generation of 3-D city models for planning of relay antenna locations in wireless
telecommunication,  urban  planning,  microclimate  models,  propagation  of  noise
and pollutants
•   Rapid   mapping   and   damage   assessment   after   natural   disasters,   e.g.,   after
hurricanes, earthquakes, landslides, etc.
•   Measurement of snow- and ice-covered areas, including glacier monitoring
•   Measurement of wetlands
•   Derivation  of  vegetation  parameters,  e.g.,  tree  height,  crown  diameter,  tree
density, biomass estimation, determination of forest borders
•   Hydrographic surveys in depths up to 70 m. Airborne Laser Scanning vs Photogrammetry – a comparison
Common aspects between photogrammetry and ALS include:
1.  use of GPS, and with digital photogrammetric sensors, especially linear ones, GPS
/ INS;
2.  methods for processing of raw data, like filtering of large errors, removal of non-
DTM  objects  like  buildings,  data  reduction  (thin-out)  and  compression,  and
detection of breaklines, are shared between ALS and image matching for DSM /
DTM generation;
3.  furthermore,  when  laser  data  are  regularly  interpolated,  they  can  be  treated  as
images and various image analysis / processing techniques can be applied to them.
Thus, sensor integration and image (or digital signal) processing and analysis are two
important topics that unify the two technologies.
The main common application, and competition field, between photogrammetry and ALS is the
3D  measurement  of  surfaces  and  single  objects.  The  reason  is  that  classification  and200
identification of objects with ALS, without use of additional optical sensors, is very difficult to
However, ALS has some strengths which can be favourably exploited in certain applications,
the most important of which are listed below:
•   Mapping of surfaces with very little / no texture or poor definition. There, image
matching  delivers  very  poor  results,  and  manual  measurements  are  also  poor  or
slow  /  cumbersome.  Examples  include  ice  /  snow  surfaces,  sand  (coasts,  dunes,
desserts), swamps and wetlands.
•   Mapping of forests and vegetated areas. ALS systems can provide measurements
on the ground. The penetration rate mainly depends on type of trees (deciduous or
coniferous)  and  season.  Useful  results,  depending  also  on  the  terrain  roughness,
can  be  achieved  even  with  penetration  rates  of  20–30%.  In  addition,  through
appropriate data process-ing, both ground and tree height can be determined.
•   Mapping  of  long,  narrow  features.  This  includes  road  mapping,  planning  and
design, powerline corridor planning and tower design, coastal erosion monitoring,
coastal zone management, traffic and transport, riverways and water resources and
traffic  management,  mapping  of  railway  lines,  fiber-optic  corridors,  pipelines,
dikes,  etc.  Since  ALS  systems  have  a  narrower  swath  in  comparison  to  optical
sensors, they are more cost-effective in capturing the information needed for such
•   DSM generation of urban regions for urban planning, roof-top heights for communication
antennas, etc.
•   Mapping of very small objects, e.g., power lines (probably THE killer application
of  ALS),  which  are  hardly  visible  in  optical  images,  or  whose  measurement
cannot be automated.
•   Fast   response   applications.   Since   ALS   provides   digital   range   measurements,   this
information  can  be  quickly  converted  to  3D  coordinates.  This  can  be  important  in  some
cases, e.g., involving natural disasters.201
9.4.2 Laser remote sensing of forest structure
§     M.A. Lefsky, W.B. Cohen, S. A. Acker, T.A. Spies, G.G. Parker, D. Harding, LIDAR REMOTE SENSING OF
EXPERIMENTAL  FOREST,  OREGON,  USA,  Pages  79-91  in:  Greer,  J.D.  ed.  1997  Natural  Resources
Management using Remote Sensing and GIS. ASPRS, Washington D.C.
Scanning lidar remote sensing systems have recently become generally available for
use  in  ecological  applications.  Unlike  microwave  and  conventional  optical  sensors,
lidar sensors directly measure the distribution of vegetation material along a vertical
axis  and  can  be  used  to  provide  three-dimensional  characterizations  of  vegetation
structure.  Ecological  applications  of  scanning  lidar  have  previously  used  uni-
dimensional   indices   of   canopy   height.   A   new   three-dimensional   approach   to
interpreting  lidar  waveforms  was  developed  to  characterize  the  total  volume  of
vegetation  and  empty  space  within  the  forest  canopy,  and  their  spatial  organization.
These aspects of the physical structure of canopies have been infrequently measured,
either from field or remote methods.
Characterization  of  forest  structure  in  moderate  to  high  biomass  systems  is  one  of  the  key
challenges in remote sensing. The ability to remotely sense both the total biomass and physical
structure  of  these  forests  would  provide  one  way  to  meet  the  need  for  forest  inventory  in
support of research and management of both carbon balance and habitat conditions.
The SLICER (Scanning Lidar Imager Of Canopies By Echo Recovery) instrument is
one of a new generation of lidar remote sensing systems that augment traditional first-
return  laser  altimetry  with  a  surface  lidar  capability.  Laser  altimeters  measure  the
distance between the sensor and a target through the precise measurement of the time
between  the  emission  of  a  pulse  of  laser  light  from  the  sensor,  and  the  time  of
detection  of  light  reflected  from  the  target.  In  surface  lidar,  the  power  of  the  entire
returning  laser  signal  is  digitized,  resulting  in  a  waveform  that  records  the  vertical
distribution  of  the  backscatter  of  laser  illumination  from  all  canopy  elements  (foliar
and woody) and the ground reflection, at the wavelength of the transmitted pulse (1064
nm, in the near-infrared). The use of relatively large footprints (5-25 m) is optimized to recover
returns from the top of the canopy and the ground in the same waveform, yet be small enough
to be sensitive to the contribution of individual crowns.202
Most  work  in  remote  sensing  of  forested  systems  has  been  analysis  in  the  spectral  domain,
specifically the analysis of multi-spectral images. The spectral qualities of forest stands are due
both  to  the  electromagnetic  properties  of  the  elements  that  are  present  in  each  pixel  (foliage,
soil, woody debris, etc.) and to their three-dimensional organization, which determines the total
cover  of  each  element  in  the  pixel  and  the  distribution  of  illumination  and  shadow  on  those
elements.  While  these  type  of  images  have  been  extremely  valuable  for  general  vegetation
mapping,  it  has  not  been  possible  to  derive  from  them  detailed  biophysical  information  for
moderately-high to high-biomass. A second approach to the use of
multi-spectral images is analysis in the spatial domain. This approach uses the two-dimensional
spatial pattern of spectral variability to infer aspects of the physical organization of forests, and
has  had  considerable  success  in  predicting  forest  structure.  However,  these  approaches  have
been constrained by the requirement that they infer three dimensional structure from its two-
dimensional projection, in the form of a remotely sensed image.
The failure of optical sensors to adequately measure the structural properties of forests is not
incidental; it is a consequence of the imaging process itself. These sensors integrate energy and
matter  fluxes  along  the  axis  between  the  sensor  and  scene.  Integration  along  this  dimension,
roughly  corresponding  to  the  "z-axis"  or  height  of  vegetation,  combines  the  overstory,
understory, and soil measurements into a single measurement per pixel. This results in a two-
dimensional  remotely-sensed  image,  with  the  spectral  data  for  each  pixel  most  influenced  by
cover  of  canopy  structure.  More  detailed  structural  information  from  closed  canopies  is  only
possible when there are associated changes in the horizontal dimension (both within and among
pixels), such as varying proportions of tree and shadow with changing tree density, or by the
increased presence of new scene components (such as lichen) in the forest canopy. Lidar does
not  suffer  from  this  limitation,  because  it  directly  measures  the  vertical  distribution  of
aboveground plant biomass.
The first generation lidar sensors for vegetation studies were designed to record distance to the
first  reflective  surface  intercepted  by  a  laser  pulse  over  a  relatively  small  sampling  area,  or
footprint, approximately 1 m in diameter. Returns from the top surface of a forest canopy were
combined with subsequent measurements of distance to the forest floor, obtained through gaps
in  the  forest  canopy,  to  infer  the  height  of  the  dominant  trees.  A  later,  more  sophisticated
technique  involved  recording  the  distance  to  the  first  and  last  reflective  surface  for  each
footprint,  giving  a  direct  height  measurement  for  each  observation.  Such  techniques  have
proven  useful  for  predicting  canopy  height,  timber  volume  and  forest  biomass,  and  percent
canopy  cover.  However,  the  relatively  small  geographic  area  covered  in  these  data  sets,
challenges  in  analyzing  the  data,  and  the  lack  of  standardized  methods  for  their  geolocation
have limited the use of conventional lidar sensors within the ecological community.203
The new generation of lidar instruments developed at NASA's Goddard Space Flight
Center  and  elsewhere  have  minimized  these  barriers  to  a  wider  application  of  the
technology. Whereas earlier devices used a small footprint and most often measured
the  distance  to  the  first  reflective  surface,  the  newer  devices  send  out  a  laser  pulse
over an approximately 5-25 m diameter footprint, and record the timing and power of
backscattered  light  over  the  full  height  profile.  Although  the  power  of  the  return
signal  falls  as  the  signal  is  intercepted  by  canopy  structure,  return  energy  from  the
ground  is  recorded  in  nearly  all  waveforms,  which  allows  an  estimate  of  the  total
height  of  the  stand,  and  indicates  that  some  energy  is  available  for  the  detection  of
understory foliage, where present. Using an algorithm developed by Drs. D. Harding
and  M.  Lefsky,  the  lidar  waveform  can  be  transformed  to  estimate  the  bulk  canopy
transmittance and the vertical distribution of reflective canopy surfaces.
The canopy colume method of M. Lefsky and D. Harding – transforming the waveform into an estimate
of the canopy height profile (CHP), the relative distribution of the canopy as a function of height. A
threshold  value  was  then  used  to  classify  each  element  of  the  CHP  into  either  “filled”  or  “empty”
volume,  depending  on  the  presence  or  absence  (in  the  profile)  of  canopy  material.  A  second  step
classified the filled elements of the matrix into an “euphotic” zone which contains all filled elements of
the  profile  that  are  within  the  uppermost  65  %  of  canopy  closure,  and  an  “oligophotic”  zone,
consisting  of  the  balance  of  the  filled  elements  of  the  profile.  These  two  classifications  were  then
combined  to  form  three  classes;  empty  volume  beneath  the  canopy-  (ie.  closed  gap  space),  filled
volume within the euphotic zone, and filled volume within the oligophotic zone.
Using these methods, very good estimates of aboveground biomass, Leaf Area Index,
number of stems greater than 100cm, Diameter at Breast Height, and distinctions can
be made between young, mature and old-growth plots. Multi-Beam Laser Altimeter (MBLA) - The Lidar Instrument for the
Vegetation Canopy Lidar (VCL)
Motivation  for  work  relating  forest  attributes  to  lidar  sensed  canopy  structure  has  been
enhanced  by  the  announcement  that  VCL,  the  Vegetation  Canopy  Lidar  mission,  has  been204
funded  by  NASA’s  Earth  System  Science  Pathfinder  (ESSP)  program,  scheduled  to  be
launched in mid 2000.
The  Vegetation  Canopy  Lidar  (VCL)  Mission  utilizes  pulsed  laser  radar  (i.e.  lidar)
from a single-instrument, small spacecraft in a 65 degree inclination, 400 km altitude,
circular,  Earth  orbit  for  continuous,  global  remote  sensing  of  tree  canopy  height,
vertical  distribution  of  intercepted  surfaces  in  the  canopy,  and  ground  topography.
The  lidar  instrument  for  VCL  is  called  the  Multi-Beam  Laser  Altimeter  (MBLA).  It
has 5 laser transmitters, a large receiver telescope, a set of 5 detectors and laser-pulse
analysis  electronics,  computer  data/command  electronics,  and  a  pointing  angle
measuring  system.  The  primary  role  for  MBLA  in  measurement  of  the  Earth's
vegetation  canopy  is  enabled  by  the  related  techniques  of  laser  pulse  time-of-flight
measurement  and  laser  pulse  shape  analysis  that  constitute  a  lidar  capability  for
remote sensing of the Earth's surface.
The MBLA lidar data have a greater spatial extent than those of conventional space-
based nadir-profiling lidar systems due to increases in both coverage and sampling of
the  Earth's  vegetation  and  topography.  Spatial  coverage  is  increased  across-track  by
providing  information  with  multiple  beams,  each  of  which  samples  independently
varying  landforms.  Sampling  is  increased  by  using  higher  laser  pulse  rates  in  each
beam. In our MBLA measurement concept, 5 beams, each operating at the 1064 nm
fundamental wavelength of the Nd:YAG solid-state laser, are arranged in a pentagon
pattern  within  a  20  urad  circle  that  is  centered  on  nadir.  When  one  point  of  this
pentagon is oriented along the flight direction, VCL simultaneously produces 5 tracks
of  Earth  surface  information;  1  track  at  nadir  and  2  tracks  each  to  the  left  and  right
that are spaced apart by 5 urad. For the nominal VCL orbital altitude of 400 km the205
across-track separation between adjacent tracks is 2 km. The result is an 8 km wide
strip image of Earth surface vertical structure.
9.5 Scanning laser mapping of the coastal zone
9.5.1 Historic development
The utilization of an airborne laser as a depth “sounder” has been studied and tested at
the  U.S.  Naval  Oceanographic  Office  since  1966.  The  experimental  laser  system  at
the  early  1970’s  was  able  to  measure  depths  down  to  10m.  Since  then  much
development has taken place.
Today there are three fully operational airborne lidar bathymetry systems in operation:
SHOALS  (Scanning  Hydrographic  Operational  Airborne  Lidar  Survey)  ,  LADS
(Laser Airborne Depth Sounder), and Hawk Eye.
The  SHOALS  system,  developed  for  the  US  Army  Corps  of  Engineers  (USACE),
employs lidar technology to remotely collect accurate, high-density measurements of
both  bathymetry  and  topography  in  coastal  regions.  SHOALS  has  been  in  full
operation  since  March  1994  and  to  date  surveyed  more  than  230  projects  totalling
5000  km  2  .  These  surveys  cover  a  variety  of  project  types  including  coverage  of
maintained channels and harbours, coastal structures, and dredged material placement
areas as well as adjacent beaches. SHOALS data collected for the US Navy and for
the  National  Oceanic  and  Atmospheric  Administration  (NOAA)  were  used  for
creation of nautical charts. Other SHOALS surveys were for beach nourishment and
erosion monitoring and for emergency response to hurricanes and ship groundings.
Originally  developed  by  Australia's  Defence,  Science  and  Technology  Organisation
(DSTO) for the Royal Australian Navy, the Laser Airborne Depth Sounder (LADS),
fulfils a crucial need. Two hundred years after Cook, Flinders and other explorers of
the Southern Hemisphere started charting Australia's vast coastline, nearly half of the
continental  shelf,  or  around  one  million  square  kilometres,  remains  unsurveyed  or
incompletely surveyed. LADS was developed to help speed up surveying and charting
The LADS system has been providing valuable data for The Australian Hydrographic
Service over the last seven years (since 1993). 10,000 soundings per square kilometre,
spaced  10  metres  apart  in  a  swath  240  metres  wide;  LADS  provides  accurate,  high
density  digital  depth  and  positional  data  of  coastal  waters  up  to  50  metres  in  depth.
Flying  at  145  knots,  500  metres  above  the  sea,  unhindered  by  reefs  or  shallows,
LADS surveys the sea floor at a rate in excess of 50 square kilometres an hour.
The swedish helicopter borne lidar called FLASH (FOA Laser Airborne Sounder for
Hydrography  –  started  operating  in  1986)  has  been  further  developed  into  two
operational  systems  called  Hawk  Eye.  Saab  Instruments  is  the  main  contractor  and
Optech   Inc.   as   the   main   subcontractor.   FOA   (Swedish   Defence   Research
Establishment)  is  member  of  the  Hawk  Eye  project  team  together  with  the  Swedish
Hydrographic    Department,    the    Swedish    Navy    and    the    Swedish    Material
Administration. The first Hawk Eye unit was delivered in 1993.
A  Hydro  Optical  Transmission  Sensor  (HOTS)  was  developed  in  conjunction  with
Hawk Eye to obtain depth profiles of beam attenuation, for predicting the effect of the
optical  properties  of  water  on  Hawk  Eye  performance.  HOTS  is  used  in  the
preplanning  phase  of  Hawk  Eye  operations.  The  design  of  the  instrument  features206
easy  operation  by  one  man  from  any  surface  or  helicopter  platform.  The  system  is
lowered into the sea by hand and sinks at a rate of 0.5 m/sec. Reading are taken and
recorded every 0.5 m. When the bottom is touched the instrument is recovered by the
line. The stored data is transferred to a PC for evaluation.
9.5.2 Working principles
An airborne lidar bathymeter uses lidar technology to measure water depths. A laser
transmitter / receiver mounted on an aircraft transmits a laser pulse which travels to
the air–water interface where a portion of this energy reflects back to the receiver.
The  remaining  energy  propagates  through  the  water  column  and  reflects  off  the  sea
bottom.  The  water  depth  comes  directly  from  the  time  lapse  between  the  surface
return  and  bottom  return,  and  each  sounding  is  appropriately  corrected  for  surface
waves  and  water  level  fluctuations  (see  figure).  In  practical  application  of  this
technology,  laser  energy  is  lost  due  to  refraction,  scattering,  and  absorption  at  the
water  surface,  sea  bottom,  and  as  the  pulse  travels  through  the  water  column.  The
combination  of  these  effects  limits  the  strength  of  the  bottom  return  and  therefore
limits  the  maximum  detectable  depth.  Optical  water  clarity  and  bottom  type  are  the
two  most  limiting  factors  for  depth  detection.  Typically,  lidar  bathymeters  collect
through depths equal to three times the site’s Secchi (visible) depth.207
figure2: This is an example of a waveform that the Ground-Based Unit receives208
The  following  table  summarises  the  performance  of  ALS  used  for  bathymetry  (the
performance depends off course on flight height and speed):
ALS system LADS SHOALS Hawk Eye
Survey height 500m 200m 50 to 800m
Survey speed 145 knots (75m/s) 60 m/s
Wavelength (IR) 1,064nm 1,064nm 1,064nm
Wavelength (green)          532nm 532nm 532nm
Sounding rate 168 per second 200 per second
Swath width 240m (nominal) 110m programmable,
max 0.85*altitude
Sounding spacing 10m  (nominal)  –  thus,
providing full coverage
at    depths    shallower
than 12m
4m 1 to 15m
Depth range 2 to 50m Max 60m 1 to 40m
Area coverage 54 km 2 /hour 16 km 2 /h 18   km 2 /h   for   300m
altitude and 5m density
Depth sounding
better than 0.3m (one
standard deviation)
over depth range 2-
±15cm 0.3m
Positional  accuracy  of
15m ±3m (with DGPS),
±1m (with KGPS)
Diameter of laser
footprint at the bottom
Weather conditions can affect the data collected (as reported for the LADS):
1.  winds
•   strong  winds  increase  turbidity,  particularly  in  inshore  areas  and  the  rougher
sea increases beam refraction and scattering, degrading the overall quality of
bottom returns and adversly affecting sounding accuracy.
•   glass calm conditions on the other hand, cause loss of surface returns from the
extremity  laser  spots  with  consequent  degredation  of  the  surface  model  and
possible increased depth errors due to undetected platform tilt.
•   optimum  conditions  for  LADS  are  light  to  moderate  winds  with  sea  states
between 1 and 3.
2.  clouds
•   high  clouds,  particularly  stratus,  can  improve  performance  by  reducing
ambient light levels and sun glint.
•   low  clouds  (base  below  1640  feet)  prevents  data  collection,  yielding  bright
reflections and obscuring firstly the infra-red and then the green laser beam.
The accuracy of ALS soundings is dependent off course both on the accuracy of the
ALS depth measurement, and on the accuracy of tidal corrections applied.
9.5.3 Benefits
Airborne  data  collection  eliminates  the  difficulties  associated  with  shipborne  survey
of  shallow,  dangerous  or  complex  waters  such  as  reef  areas.  It  also  permits  the  fast
identification of key areas of navigational significance, such as undiscovered channels
and passages.209
Powerful analysis and conversion capabilities provide same day turn-around of data.
This enables efficient, flexible survey sortie operations.
The above figure (taken from the LADS pages) compares the coverage of soundings
of an LADS and a conventional echo-sounder technology, the productivity gains are
readily understood. However, ALS will never replace ships, due to its limitations.210
9.5.4 Case study: The ALACE project - Airborne LIDAR Assessment of Coastal
Erosion Project
Beaches are some of the earth’s most dynamic geologic features. Beach morphology
fluctuates over a wide range of time scales, varying from periods of hours associated
with  diurnal  tides  and  storm  events,  to  years  and  decades  associated  with  long-term
erosional trends. Human actions, especially during the last 100 years, have created a
situation in which beach erosion can have severe economic consequences.
Accurate and timely assessment of erosion conditions and storm impacts is needed to
assist  decision  making  on  land  use,  beach  renourishment,  erosion  calculations,
insurance   compensation,   and   property   value   estimation.   Proper   storm   damage
assessment  is  an  enormous  task  for  emergency  and  disaster  response  agencies  and
The ALACE project is a partnership between the National Oceanic and Atmospheric
Administration   (NOAA)   Coastal   Services   Center   (CSC)   in   Charleston,   South
Carolina;  NASA  Goddard  Space  Flight  Center,  Wallops  Flight  Facility  (WFF)  in
Wallops,  Virginia;  and  U.S.  Geological  Survey  (USGS)  Center  for  Coastal  Geology
in St. Petersburg, Florida. The project’s goal is to establish the capability of aircraft
laser swath mapping to provide highly accurate, cost-effective information on coastal
topography,  erosion,  and  shoreline  position.  In  working  toward  this  goal,  NOAA,
NASA,  and  USGS  have  conducted  several  mapping  missions  along  significant
portions of U.S. coast using the NASA Airborne Topographic Mapper (ATM) flown aboard
a NOAA Twin Otter aircraft.
The  ATM  is  currently  operated  with  a  Spectra  Physics  TFR  laser  transmitter  that
provides   a   7-nanosecond   wide,   250-micro-joule   pulse   at   a   frequency-doubled
wavelength of 523 nanometers (nm) in the blue-green spectral region. A scan mirror
with  an  off-nadir  angle  of  15  degrees  was  used  in  the  ALACE  beach  mapping
surveys,   producing   an   elliptical   scan   pattern   with   a   swath   width   equal   to
approximately 50 percent of the approximately 700-meter aircraft altitude. The usual
density of points on the ground is 30 per 100 square meters.
Figure: ATM Laser Mapping Scan Swath
For  the  fall  1997  mapping  missions,  a  passive  channel  sensor  was  added  to  the  ATM.  This
sensor collects geo-referenced panchromatic (excluding 523 nm) data along the same elliptical211
scan path as the active laser. Images created from the passive channel data help identify ground
features, and are used to assist in the delineation of the beach region.
Comparisons  were  made  between  same  day  ATM  surveys,  multiple  days  ATM  surveys,  and
with data collected by buggy trucks, Reiss total station survey, and others.
The statistical results from the ATM comparisons are summarized in the following tables ( m  =
mean,  s  = standard deviation).
It  can  be  seen  that  ATM  measurements  were  found  to  match  the  desired  accuracy
about  10  cm,  in  most  of  the  cases,  amounting  to  +/-  15  cm  over  the  beach.  The
horizontal accuracy is 1.5m in the worst case.212
The next figure shows a preliminary raw data image (note the elliptic pattern of the
The following figure presents a beach profile of Assateague Island, as surveyed by the
ATM-2 in october 1996:214
9.5.5 IHO and FEMA standards, and their relation to ALS technology IHO S-44
The  following  table  is  a  Summary  of  the  Minimum  Standards  for  Hydrographic
Surveys,  as  required  by  S-44  (special  publication  no.  44)  of  the  International
Hydrographic Organization (IHO):
Order Special 1 2 3
Examples of typical
berthing areas,
and associated
critical chanells
with minimum
Harbours, harbour
approach channels,
tracks and some
coastal areas with
depth up to 100m
Areas not
described in
Special Order and
Order 1, or areas
up to 200m water
Offshore areas
not described
in Special
Order, and
Orders 1 and 2
Horizontal accuracy
(95% confidence
2m 5m + 5% of depth        20m + 5% of
150m + 5% of
Depth accuracy for
reduced depths (95%
confidence level)
0.25m 0.5m 1.0m Same as order
100% bottom search      Compulsory Required in
selected areas
May be required
in selected areas
Not applicable
System detection
Cubic features >
Cubic features >
2m in depths up to
40m; 10% of depth
beyond 40m
Same as order 1        Not applicable
Maximum line
Not applicable,
as 100% search
3*average depth or
25m, whichever is
3-4*average depth
or 200m,
whichever is
ALS is in fact the only Remote Sensing technology included in the S-44, and I quote:
“Airborne laser sounding is a new technology which can offer substantial productivity
gains  for  surveys  in  shallow,  clear  water.  Airborne  laser  systems  are  capable  of
measuring depths to 50 m or more.”
“Airborne  laser  systems  are  capable  of  measuring  depths  to  50  m  or  more  provided
the  water  is  clear.  Hazards  to  navigation  detected  by  airborne  laser  should  be
examined  using  SBES  (Single  Beam  Echo  Sounder),  MBES  (Multi  Beam  Echo
Sounder) or high density airborne laser. All swaths should be crossed, at least once,
by  a  checkline  to  confirm,  by  this  method,  the  accuracy  of  positioning,  depth
measurement and depth reductions.”
A comparison analysis between MBES and ALS in performing hydrographic surveys
in  shallow  waters,  was  done  by  Axelsson  and  Alfredsson  (both  from  Saab).  Their
main conclusions are as follows:
•   The cost for MBES survey mainly depends on the depths of the waters surveyed,
with  very  high  costs  for  shallow  waters,  decreasing  a  lot  for  deeper  waters.  The
cost for surveying with ALS is mainly dependent on the IHO order and not on the
•   Configurations  using  combinations  of  MBES  platforms  and  ALS  systems  have
much lower annual costs, with the ALS covering most of the shallow waters.
•   ALS surveys offer the ideal complement to sonar surveys, for several reasons:215
Ø  Survey  efficiency,  ALS  provides  in  shallow  waters  outstanding  productivity
with  minimum  crew  requirements.  High  velocity  of  the  carrier  and  a  large
swath width give fast coverage of the sea bed. Narrow passages, archipelagos,
reefs and the coast line can be surveyed in just one mission.
Ø  Survey  accuracy,  ALS  gives  full  bottom  coverage  in  all  areas  even  in  very
shallow areas where other methods is very expensive to use. Local variations
in salinity and/or temperature present no problem to ALS.
Ø  Survey  safety,  the  survey  platforms  of  an  ALS  is  airborne  and  readily  clears
narrow passages, shoals and reefs. Unknown waters and areas subject to mine
hazards are no safety problems for the survey crew or the survey equipment.
Ø  Versatility,  an  ALS  may  be  adapted  for  environmental  control,  e.g.  erosion,
contamination    or    reef    growth/breakdown    periodical    surveys.    In    its
environmental control role, the ALS also benefits from the carrier’s short re-
deployment time. FEMA
The  United  States  Federal  Emergency  Management  Agency  (FEMA)  has  published
guidelines and specifications that must be used for the application of Airborne LIght
Detection  And  Ranging  (LIDAR)  systems  for  gathering  the  data  necessary  to  create
digital  elevation  models  (DEMs),  digital  terrain  maps,  and  other  National  Flood
Insurance Program (NFIP) products.
Given  below  are  the  main  guidelines  and  specifications,  to  give  the  reader  an  idea
about the complications involved AIRBORNE LIGHT DETECTION AND RANGING SYSTEMS - General Guidelines for Use
Two  important  factors  in  the  LIDAR  system  mission  planning  are  the  point  density  of  the  randomly
spaced  LIDAR  points  and  the  point  spacing  of  the  uniformly  spaced  DEM  points  derived  from  the
randomly  spaced  LIDAR  returns.  The  correct  point  density  necessary  to  accurately  represent  terrain
and  terrain  features  will  depend  on  flight  conditions,  mission  purpose,  and  other  variables.  As
discussed  later  in  this  Appendix,  DEM  point  spacing  of  5  meters  and  vertical  accuracy  of  30
centimeters are required.
Flight-path  planning  is  another  important  factor  in  the  LIDAR  system  mission.  The  flight  path  shall
include both parallel and cross flight lines to cover the study area satisfactorily.
Unlike aerial photogrammetry, LIDAR missions can be flown without regard to sun angle. Flights may
take place at night, if conditions otherwise allow.
Elevation and measurement information related to subsurface channel and hydraulic structure geometry
must  be  obtained  through  the  use  of  other  mapping  technologies  over  deep  or  turbid  water.  In  some
instances,  shallow  water  and  near-shore  coastal  surveys  can  be  accomplished  using  LIDAR  systems
equipped with lasers operating in portions of the light spectrum that allow transmission through water.
LIDAR system tolerance for inclement weather conditions (e.g., high winds, wet snow, rain, fog, high
humidity, low cloud cover) generally is higher than that of other photogrammetric methods. However,
such conditions have been known to degrade the accuracy of laser return data. Therefore, the contractor
shall generally avoid missions during inclement weather.
High point densities may allow satisfactory data collection in areas of dense foliage. Still, care must be
taken  in  planning  missions  with  regard  to  both  natural  (vegetative)  and  manmade  (structure)  ground
cover. Pulse width, beam divergence, first and last pulse return discrimination, and choice of the post-
processing algorithms may all affect the accuracy of LIDAR-derived data in areas of dense foliage. Performance Standards
The appropriate mapping standards in Appendix 4 of these Guidelines shall apply to NFIP maps and
map products derived from LIDAR systems. LIDAR-derived data must have the accuracy required to
produce  topographic  maps  and  products  that  meet  the  National  Standard  for  Spatial  Data  Accuracy
FEMA is not aware of any existing LIDAR system performance standards. Current information about216
LIDAR  systems  is  available  from  the  National  Oceanic  and  Atmospheric  Administration  (NOAA),
National  Aeronautic  and  Space  Administration,  U.S.  Army  Corps  of  Engineers,  LIDAR  system
manufacturers and vendors, and private firms that provide LIDAR system services. As professional or
trade associations issue specifications and standards, FEMA may adopt them and amend this Appendix.
A.   Performance Standards
The contractor must furnish all necessary materials and equipment. The contractor also must supply the
supervisory, professional, and technical services personnel required to manage, survey, document, and
process all data associated with LIDAR system mapping, scanning, and digital image processing. All
deliverables must be provided in accordance with the contract and the requirements in this Appendix.
B.   System Calibration
LIDAR system components are most effectively tested and calibrated by the equipment manufacturer.
Therefore,   the   contractor   must   provide   FEMA   with   evidence   of   manufacturer   calibration.
In addition to evidence of manufacturer calibration of system components, the contractor must submit
evidence  that  the  total  LIDAR  system  was  calibrated  prior  to  project  initiation  for  the  purposes  of
identifying and correcting systematic errors. Proper system calibration requires repetitive overflight of
terrain features of known and documented size and elevation using flight paths similar to those that will
be used in the study area.
C.   Flight Planning
Planning  a  flight  path  that  considers  all  aspects  of  data  collection  is  critical  to  the  success  of  the
mission. An analysis of the project area, project requirements, topography, proximity to restricted air
space,  and  other  factors  will  determine  the  flight  path  configuration.  The  mission  should  include
parallel flight lines and at least one cross flight line. The spacing between the flight lines will depend
on the desired amount of sidelap between swaths.
The density and accuracy of data generated by different equipment vary widely. The contractor shall
have  the  flexibility  of  providing  a  flight  path  to  create  the  necessary  point  density  to  minimize  the
occurrence of data voids.
The contractor must check the Position Dilution of Precision (PDOP) in the study area. The PDOP is an
indicator of the positional accuracy that can be derived from the current GPS satellite geometry, which
varies continuously; the smaller the PDOP number, the higher the data quality.
The contractor must document mission date, time, flight altitude, airspeed, scan angle, scan rate, laser
pulse rates, and other information deemed pertinent. For a sample mission data recordation checklist,
refer to Table A4B-1.
D.   GPS Base Stations
The contractor must select the GPS base station(s) carefully to ensure reliable differential processing of
airborne  GPS  data.  The  National  Geodetic  Survey  (NGS)  recommends  the  simultaneous  use  of  two
GPS  base  stations  during  the  mission.  (Note:  Either  public-  or  private-domain  GPS  base  stations  are
suitable for use for this purpose.) Where possible, GPS base stations shall have ellipsoid height to an
accuracy  of  2  centimeters  relative  to  the  Continuously  Operating  Reference  Stations  (CORS)  or  the
High  Accuracy  Reference  Network  (HARN),  both  operated  by  the  NGS.  The  contractor  must  use  a
high-quality, dual-frequency GPS receiver and associated antenna at the GPS base stations. GPS Control
Part 1, "Reporting Methodology (FGDC-STD-007.1)," and Part 2, "Standards for Geodetic Networks
(FGDC-STD-007.2),"  of  the  Geospatial  Positioning  Accuracy  Standards,  published  by  the  FGDC  in
1998,  provide  a  common  methodology  for  determining  and  reporting  the  accuracy  of  horizontal  and
vertical coordinates for geodetic control points (survey monuments). Additional guidance is included in
NOAA  Technical  Memorandum  NOS  NGS-58,  "Guidelines  for  Establishing  GPS-Derived  Ellipsoid
Heights  (Standards:  2  cm  and  5  cm),"  dated  November  1997.  The  GPS  control  guidance  in  FGDC-
STD-007.1  and  FGDC-STD-007.2  and  in  Appendix  4  of  these  Guidelines  shall  apply  to  LIDAR-
derived data submitted to FEMA. Post-Processing of Data
For  hydraulic  modeling,  the  contractor  must  provide  high-resolution,  high-accuracy,  "bare-earth"
ground elevation data. To restrict data to ground elevations only, the contractor must remove elevation
points on bridges, buildings, and other structures and on vegetation from the LIDAR-derived data. In
addition to randomly spaced LIDAR points, before and after removal of data associated with structures
and vegetation, the contractor must produce a bare-earth DEM, with regular 5-meter point spacing in
eastings  and  northings.  In  accordance  with  NSSDA,  the  contractor  must  use  Triangular  Irregular
Network (TIN) linear interpolation procedures when validating the vertical accuracy of the DEM.217
In addition to DEMs, the contractor shall produce breaklines for stream centerlines, drainage ditches,
tops  and  bottoms  of  streambanks,  ridge  lines,  road  crowns,  levees,  bulkheads,  road/highway
embankments,  and  selected  manmade  features  that  constrict  or  control  the  flow  of  water  (e.g.,  curb
lines). Quality Control/Quality Assurance
Quality Control/Quality Assurance (QC/QA) of the LIDAR-derived data is primarily the responsibility
of the contractor. This QC/QA process shall include reviews of flight alignments and completeness of
supporting data (e.g., cross sections, profiles). FEMA or its designee may perform additional QC/QA
NSSDA  uses  the  root  mean  square  error  (RMSE)  to  estimate  both  horizontal  and  vertical  accuracy.
RMSE  is  the  square  root  of  the  average  of  the  set  of  squared  differences  between  dataset  coordinate
values  and  coordinate  values  from  an  independent  source  of  higher  accuracy  for  identical  points.  If
those differences are normally distributed and average zero, 95 percent of any sufficiently large sample
should be less than 1.96 times the RMSE. Therefore 15-centimeter RMSE is often referred to as "30-
centimeter  accuracy  at  the  95-percent  confidence  level."  Following  that  convention,  the  vertical
accuracy  of  any  DEM  is  defined  as  1.96  times  the  RMSE  of  linearly  interpolated  elevations  in  the
DEM, as compared with known elevations from high-accuracy test points.
DEMs  should  have  a  maximum  RMSE  of  15  centimeters,  which  is  roughly  equivalent  to  1-foot
accuracy.  The  contractor  must  field  verify  the  vertical  accuracy  of  this  DEM  to  ensure  that  the  15-
centimeter RMSE requirement is satisfied for all major vegetation categories that predominate within
the floodplain being studied. The main categories of ground cover that the contractor must separately
evaluate and report on the DEM accuracy for shall be:
a)   Bare-earth and low grass (plowed fields, lawns, golf courses);
b)   High grass and crops (hay fields, corn fields, wheat fields);
c)   Brush lands and low trees (chaparrals, mesquite, swamps);
d)   Fully covered by trees (hardwoods, evergreens, mixed forests); and
e)   Urban areas (high, dense manmade structures).
The contractor shall evenly distribute sample points throughout each category area being evaluated and
not group the sample points in a small subarea.
The RMSE calculated from a sample of test points will not be the RMSE of the DEM. The calculated
value  may  be  higher  or  it  may  be  lower  than  that  of  the  DEM.  Confidence  in  the  calculated  value
increases with the number of test points. If the errors (lack of accuracy) associated with the DEM are
normally  distributed  and  unbiased,  the  confidence  in  the  calculated  RMSE  can  be  determined  as  a
function  of  sample  size.  Similarly,  the  sample  RMSE  necessary  to  obtain  95-percent  confidence  that
the DEM RMSE is less than 15 centimeters can also be determined as a function of sample size.
For each major vegetation category, the contractor must test a sample of points and show the test points
have an RMSE less than
where n is the number of test points in the sample.
The contractor must select a minimum of 20 test points for each major vegetation category identified.
Therefore,  a  minimum  of  60  test  points  must  be  selected  for  three  (minimum)  major  vegetation
categories,  80  test  points  for  four  major  categories,  and  so  on.  The  contractor  should  consider
establishing test points when planning field surveys to gather cross section data for hydraulic modeling.
If more than two test points are outside the range of two times the RMSE, the contractor must make the
appropriate adjustment using guidance in Appendix 4 of these Guidelines.
The  contractor  shall  select  the  test  points  carefully  in  areas  of  the  highest  PDOP  to  evaluate  DEM
accuracy  under  trees  and  in  vegetation  representative  of  the  study  area.  Test  points  on  sloping  or
irregular  terrain  would  be  unreasonably  affected  by  the  linear  interpolation  of  test  points  from
surrounding DEM points and, therefore, shall not be selected.
Because the definition and criterion for measuring accuracy are derived from the assumption that the
test point samples come from a uniformly distributed population with zero mean, the contractor must
calculate other statistics. In particular, the mean and the coefficient of skew must be calculated for each
sample. Values of the mean of the test points outside of the interval +/- 2 centimeters and/or values of
the  coefficient  of  skew  outside  of  the  interval  +/-  0.5  centimeter  may  indicate  systematic  error;  the
contractor should discuss such values with the FEMA Project Officer (PO).218 Deliverables
All data and products associated with contract deliverables must meet or exceed relevant NSSDA and
fully  comply  with  the  FGDC  metadata  format  standard  with  the  provisions  in  the  contract.  The
contractor  shall  use  Appendix  7,  "Digital  Product  Delivery  Specifications,"  of  these  Guidelines  as  a
guide for preparing and submitting deliverables in digital format.
A.   Pre-Project Deliverables
Prior to data collection, the contractor must submit:
1.  A  map  (typically,  U.S.  Geological  Survey  maps  are  desirable  for  this  purpose)  showing  the  study  area
boundaries     and     flight     path,     at     a     medium     scale     (1:50,000)     or     small     scale     (1:100,000);
2.  Documentation  specifying  altitude,  airspeed,  scan  angle,  scan  rate,  LIDAR  pulse  rates,  and  other  flight  and
equipment information deemed appropriate; and
3.  A  chart  of  areas  of  high  PDOP,  or  a  list  showing  the  time  of  the  beginning  and  end  of  high  PDOP.
B.   Post-Project Deliverables
Following project completion, the contractor must submit:
1. A LIDAR system data report;
2. A flight report;
3. A ground control report;
4. Data processing procedures for selection of postings, and all orthometric values of x, y, and z coordinates for
LIDAR returns; and
5. A system calibration report.
The  LIDAR  system  data  report  must  include  discussions  of:  data  processing  methods  used;  final
LIDAR pulse and scan rates; scan angle; capability for multiple returns from single pulses; accuracy
and precision of the LIDAR data acquired; accuracy of the topographic surface products; any other data
deemed appropriate; and companion imagery, if any.
The  flight  report  must  document  mission  date,  time,  flight  altitude,  airspeed,  and  other  information
deemed  pertinent.  The  report  must  include  information  about  GPS-derived  flight  tracks,  provide  a
detailed  description  of  final  flight  line  parameters  and  GPS  controls  (i.e.,  benchmarks),  and  include
ground truth and complementary reference data.
The  ground  control  report  must  include,  at  a  minimum,  all  pertinent  base  station  information  and
mission    notes,    including    information    on    GPS    station    monument    names    and    stability.
C. Delivery of Digital Data
In addition to the pre- and post-project deliverables described above, the contractor must submit
the following:
1.   All   raw   datasets,   bare-earth   DEM   data,   and   breaklines   in   separate   data   files;   and
2. Uniformly spaced DEM(s), on ISO 9660 standard CD-ROM media in a format specified in
Appendix 7 of these Guidelines.
The  contractor  must  deliver  raw  datasets  and  LIDAR  system  data,  including  orthometric  values  for
each point, in standard, comma-delimited ASCII files in x, y, and z format. The contractor also must
flag raw datasets from sidelap and overlap areas of separate flight lines. Breaklines must be produced,
and breakline files must contain a flag record that identifies them as breakline features. The contractor
must submit raw datasets in tiles or data models matching file delivery format.
All deliverables must conform to the projection, datum, and coordinate system specified in the contract.
File  sizes  cannot  exceed  1  gigabyte,  unless  otherwise  specified  by  the  FEMA  PO.  Each  file  must  be
organized to facilitate data manipulation and processing.219
10. Metadata
The  growing  mass  of  digital  data  in  the  GIS  marketplace  is  fuelling  a  demand  for
information  about  geographic  data,  i.e.  geospatial  metadata.  Recent  efforts  toward
standardization  of  digital  geographic  data  around  the  world  (particularly  the  US
Spatial Data Transfer Standard (SDTS)) constitute a key element in utilizing existing
spatial data. Through metadata clearinghouses, such as the National Geospatial Data
Clearinghouse (, spatial data users can find what data exists, its
quality and condition, and the terms for getting and using it.
10.1 GIS metadata
The major sections of a GIS metadata include the following (the relatively unfamiliar
terms are explained, the full standards can be found in the FGDC site):
1.  Identification Information - basic information about the data set.
data set title, area covered, keywords, purpose, abstract, access and use
2.  Data Quality Information - a general assessment of the quality of the data set.
horizontal and vertical accuracy assessment, data set completeness and lineage
Completeness    Report    --    information    about    omissions,    selection    criteria,
generalization, definitions used, and other rules used to derive the data set.
Lineage  --  information  about  the  events,  parameters,  and  source  data  which
constructed the data set, and information about the responsible parties.
3.  Spatial Data Organization Information - the mechanism used to represent
spatial information in the data set.
raster, vector, or an indirect (e.g. address) link to location
4.  Spatial Reference Information - the description of the reference frame for, and
the means to encode, coordinates in the data set.
lat/long, horizontal and vertical coordinate system, map projection, datum
5.  Entity and Attribute Information - details about the information content of the
data set, including the entity types, their attributes, and the domains from which
attribute values may be assigned.
definitions of the attributes of the data set
Attribute Domain Values -- the valid values that can be assigned for an attribute.
Attribute  Value  Accuracy  Information  --  an  assessment  of  the  accuracy  of  the
assignment of attribute values.
6.  Distribution Information - information about the distributor of and options for
obtaining the data set.
distributor, file format of data, off-line media types, on-line link to data, fees220
Offline Media -- name of the media on which the data set can be received (CD-ROM,
8 mm cartridge tape, etc)
7.  Metadata Reference Information - information on the currentness of the
metadata information, and the responsible party.
who created the metadata and when
Currentness Reference -- the basis on which the time period of content information is
10.1.1 GIS metadata worked example – Bathymetry of the Gulf of
Carpentaria and the Arafura Sea
The  following  worked  example  of  a  GIS  metadata  was  taken  from  ANZLIC
(Australia New Zealand Land Information Council),  and  is  constructed  according  to
their metada guidelines.
Dataset TITLE - Bathymetry of the Gulf of Carpentaria and the Arafura Sea, Edition 1
Dataset CUSTODIAN - Royal Australian Navy Hydrographic Service
Dataset JURISDICTION - Australia
Description ABSTRACT -
The  Bathymetry  of  the  Gulf  of  Carpentaria  and  the  Arafura  Sea,  Edition  1,  is  a  dataset  that  contains
digital bathymetric information of the Gulf of Carpentaria and the Arafura Sea, Australia, Papua New
Guinea, and Indonesia. The digital data in this data base are the latitude, longitude coordinates of the
end  points  of  vectors  that  represent  bathymetric  contours.  Contours  were  hand  drawn  using  digital
systematic-survey soundings from the Royal Australian Navy Hydrographic Office and from a variety
of other bathymetric contour maps.
Data accuracy varies. Older data were collected using techniques and navigation that were less accurate
than those used at the time this map was prepared. Although nautical charts are periodically updated,
some contain soundings that are over 100 years old. Accuracy estimates of the location and depth of
soundings  cannot  be  made  because  information  on  collection  and  processing  procedures  was  not
available  for  much  of  the  source  data.  These  data  have  been  published  in  analogue  form  as  USGS
Miscellaneous Investigations Series Map 1-2550.
Description SEARCH WORD(S) -
MARINE Geology & Geophysics
-3.0 130.0, -3.0 149.0, -18.0 149.0, -18.0 130.0, -3.0, 130.0
Data Currency BEGINNING DATE - Not Known
Data Currency ENDING DATE - 31DEC1994
Dataset Status PROGRESS - Complete
DIGITAL data are stored in separate files for each contour level. Each file has a header that describes
the format and the number of records in the file. The files are in ASCII, approximately 5.5 Mbytes, for
map scales 1:750,000 to 1:2,500,000
NONDIGITAL - Plotted Maps
Data Quality LINEAGE
Contours  were  hand  drawn  using  digital  systematic-survey  soundings  obtained  from  the  Royal
Australian Navy (RAN), Hydrographic Office (unpublished data 1992), and using a variety of nautical
charts  from  the  following  sources:  Australian  Department  of  the  Navy,  Hydrography  Branch  (1970);
[Great  Britain]  Admiralty  (1983,1990);  [Great  Britain]  Hydrographer  of  the  Navy)  (1988);  Indonesia
Angkatan  Laut,  Djawatan  Hidrografi  (1961);  and  the  Royal  Australian  Navy  Hydrographic  Service
(1968. 1974, 1979, 1991). Also contours were scanned from an unpublished map (1987) by T Chase, B.
Seakins, J. Young (all of the U.S. Geological Survey), and H. Prasentyo (Marine Geological Institute of
Indonesia);  they  were  modified  and  scanned  from  published  geological  studies  (Jongsma,  1974;
Torgensen  and  others,  1983);  and  they  were  modified  and  scanned  from  the  Australian  National
Bathymetric  Map  series  (Australian  Division  of  National  Mapping,  1981,  1983,  1984,  1986,  1989,
1990), which was compiled by the division of national Mapping (NATMAP) from the bathymetric data
that now comprise the RAN digital data set.
The older data were gathered using less accurate techniques and navigation. Some of the data are over
100 years old. Sea level corrections, when applied, were estimated to about 0.5 metres from values for
distant  tidal  stations.  The  hand  contoured  maps  were  scanned  using  an  auto-vectorising  scanner  and
vector  end-points  were  converted  to  latitude-longitude  values  using  the  USGS  software  package
MAPGEN which is also available as the vector part of the GIS, GRASS.
Accuracy estimates of the location and depth of soundings cannot be made because information on
collection and processing procedures were not available for much of the source data.
Data Quality ATTRIBUTE ACCURACY - As for positional accuracy
All lines were visually checked at 1:1 000 000 and 1:250 000 scale to verify that no lines crossed, that
there were no extraneous line segments and that all lines had the correct contour value. Multiple and
dangling lines were edited using in-house software.
All lines and polygons are complete where there was sufficient depth sounding data except for three
features where lines are not shown for clarity because of the steepness of the gradient.
Contact Information CONTACT ORGANISATION - Australian Oceanographic Data Center
Contact Information CONTACT POSITION - Data Manager
Contact Information MAIL ADDRESS 1 - Level 2, MHQ Annex
Contact Information MAIL ADDRESS 2 - Wylde St.
Contact Information SUBURB/PLACE/LOCALITY - Potts Point
Contact Information STATE/LOCALITY 2 -NSW
Contact Information COUNTRY - Australia
Contact Information POSTCODE - 2011
Contact Information TELEPHONE - 61 2 563 4812
Contact Information FACSIMILE - 61 2 563 4820
Contact Information CONTACT ORGANISATION - National Geophysical Data Center
Contact Information CONTACT POSITION - Data Manager, NOAA - NGDC E/GC3
Contact Information MAIL ADDRESS 1 - 325 Broadway
Contact Information SUBURB/PLACE/LOCALITY - Boulder
Contact Information STATE/LOCALITY 2 - Colorado
Contact Information COUNTRY - United States
Contact Information POSTCODE - 80303-3328
Contact Information TELEPHONE - 1 303 497 6945
Contact Information FACSIMILE - 1 303 497 6513
Metadata Date METADATA DATE - 01MAR1996
Additional Metadata ADDITIONAL METADATA -
Maps of the Australian National Bathymetric Map Series for the area referenced.
10.2 Remote Sensing metadata
A special case of spatial data is Remote Sensing (RS) data that require some unique
considerations,  though  the  fundamental  concepts  for  arranging  their  metadata  are
essentially similar to those of any other geospatial data.
There seems to be a lack in papers dealing with the theoretical and conceptual aspects
of remote sensing metadatabases. Most of the materials deal with general geospatial
The  following  text  describes  the  special  attributes  that  Remote  Sensing  metadata
should include, as there are yet no standards.
Proper use of remote sensing data requires an understanding of how those data were
obtained.  While  ground-based  data  are  often  compiled  from  existing  data  sources
without  change  of  form  or  are  obtained  by  direct  in  situ  measurement,  deriving
geospatial data from the measurements made by remote sensing instruments is often
much  less  direct.  To  do  so  may  require  knowledge  of  the  observing  geometry,  the
instrument  behavior,  and  the  processing  methods  and  history.  In  addition,  remote
sensing  measurements  produce  large  volumes  of  data,  and  users  typically  do  not
access the entire data set, only selected files or frames.
Information  about  the  viewing  geometry  and  the  properties  and  behavior  of  the
instrument in the FGDC Metadata Content Standard  is  limited  to  the  description  of
the  number  of  points  along  the  raster  axes.  The  draft  ISO  metadata  standard  also
includes solar elevation and azimuth angles and the angle of an image to the vertical.
However,  many  user  needs  a  more  detailed  viewing  geometry:  satellite  orbit  or
aircraft flight path, platform orientation, and orientation of instruments relative to the
platform. While the proposed ISO standard includes a number of items in the section223
of  spatial  data  representation  describing  instrumentation  not  present  in  the  FGDC
Metadata  Content  Standard,  the  only  calibration  information  is  whether  camera
calibration  information  is  available.  More  information  on  the  calibration  of  the
instrument, including its dependence on wavelength and time, is usually required. A
standard description of such metadata should be defined.
Processing  of  remote  sensing  data  passes  through  several  stages.  The  instrument
calibration  must  be  applied  to  the  readings  communicated  by  the  raw  telemetry  and
the resulting physical measurements located geographically. In some cases, what the
instruments  measure  is  not  the  final  product;  for  example,  radiation  measurements
may  be  used  to  infer  temperatures.  Maps  and  grids  may  be  generated  from  data  at
individual  points.  Information  on  the  algorithms  used  to  for  these  steps  should
accompany the data. In addition, information about the processing itself, such as what
stage a given processing represents, or which version of processing is represented, is
needed. The FGDC Metadata Content Standard  allows  for  this  information  an  entry
for lineage, which the draft ISO standard has expanded this item to an entire section
on  lineage  information,  but  in  both  cases  the  content  is  unspecified  free  text.  These
extensions will define the specific items that are needed in remote sensing metadata.
The   dataset   containing   results   from   a   remote   sensing   mission   is   large   and
heterogeneous.  Necessary  descriptive  metadata  may  not  apply  to  the  entire  dataset,
but only to individual pictures or files. While the FGDC Metadata Content Standard
has no specific provision for such granularity, the informative Appendix F to the ISO
draft  provides  but  does  not  define  granule-specific  metadata.  These  extensions  will
define the granule-level metadata appropriate to remote sensing.224
11. References and links
Following is a selected list of useful references for further reading:
11.1 Basic books about Remote Sensing:
§     Remote Sensing : Principles and Interpretation, by Floyd F. Sabins, 3rd edition (August 1996),
W H Freeman & Co.; ISBN: 0716724421
§     Remote  Sensing  and  Image  Interpretation,  by  Thomas  M.  Lillesand,  Ralph  W.  Kiefer,  4th
edition (October 1999) , John Wiley & Sons; ISBN: 0471255157
§     Remote Sensing : Models and Methods for Image Processing, by Robert A. Schowengerdt, 2nd
edition (July 1997), Academic Pr; ISBN: 0126289816
11.2 Remote Sensing journals:
§     ISPRS Journal of Photogrammetry and Remote Sensing
§     Remote sensing of environment
§     IEEE transactions on geoscience and remote sensing
§     International journal of remote sensing
§     Canadian journal of remote sensing
§     Backscatter – observing the marine environment
11.3 On-line tutorials:
Following  are  listed  some  of  the  tutorials  (and  university  courses)  I  have  used
preparing this text. Given is the main subject of each tutorial.
§     The Remote Sensing Tutorial – NASA
§     Canada Center for Remote Sensing tutorial
§     Remote Sensing Core Curriculum, volume 3, Introductory Digital Image Processing
§     Centre National d'Etudes Spatiales - The science of Remote Sensing
§     University of Tennessee - Astronomy course 162 - Radiation laws
§     Washington university in St Louis – EPSc course 407 – Remote Sensing
§     USGS – Spectroscopy and imaging spectroscopy
§     USGS - Surface Reflectance Calibration of Terrestrial Imaging Spectroscopy Data
§     Iowa State University - Remote Sensing in Precision Agriculture
§     Synoptics and the Dutch survey department - Airborne Digital Video
§     NOAA - Airborne Laser Beach Mapping
§     University of Texas - Radar altimetry
§     The NCGIA Core Curriculum in GIScience
§     GIS lessons (each html page is a lesson) – by Michael Frank Goodchild Remote Sensing softwares:
Among the leading Remote Sensing softwares, the following names can be given:
There are also some low-cost Remote Sensing-GIS raster softwares:
And, there are some Remote Sensing-GIS raster freewares (downloadable):
11.5 Other Remote Sensing links:
§     Airborne Laser Mapping
§     Canada Center for Remote Sensing
§     Landsat 7
§     Center for the Study of Earth from Space
§     Table of Fundamental Physical Constants
§     European Space Agency
§     TOPEX/Poseidon
§     IKONOS

About Author:

I am Thomas Britto here to share my experiences in the civil engineering field to all my readers.Today many students are struggling to buy books at high prices. So I decided to start a blog and share my experience and knowledge with all my readers.

Let's Get Connected: Twitter | Facebook | Google Plus

No comments