This document is part of the Satellite Imagery FAQ
Subject: Image Basics
Image Basics _Contributed by Wim Bakker (firstname.lastname@example.org)_
What is an image?
A digital image is a collection of digital samples.
The real world scene is measured at regular distances (=digital). One
such measurement is limited in
One sample covers only a very small area from the real scene.
The sensor needs some integration time for one measurement (which
is usually very short).
* Spectral coverage
The sensor is only sensitive for a certain spectral range.
Furthermore, the sample is quantized, which means that the physical
measure in the real world scene is represented by a limited number of
levels only. Usually 256 levels of "grey" are sufficient for digital
images; 256 levels can be represented by an eight bit unsigned Digital
Number (DN). "Unsigned" because the amount of light is always
positive. More levels will need more bits; the quantization determines
the amount of bits per pixel on the image storage.
Image samples are usually called _pixel_ or _pel_ after the
combination of "picture" and "element". A pixel is the smallest unit
of a digital image. The size of this unit determines the resolution of
an image. The term _resolution_ is used for the detail that can be
represented by a digital image. As discussed before the resolution is
limited in four ways:
* Spatial resolution.
If one pixel is a ground cell sample of 20 by 20 meter then no
objects smaller than 20 meter can be distinguished from their
background. This doesn't necessarily mean they cannot be
Note that if the spatial resolution doubles, the amount of image
data increases by a factor 4!
* Temporal resolution.
A distinction can be made between
+ Temporal resolution of one image.
Fast moving objects will appear blurred on one image. E.g.
the temporal resolution of one TV image is about 1/25 of a
+ Temporal resolution of a time series of images.
If the images are taken sparsely in time then the possibility
exists that some phenomena will be missed. The resolution of
Landsat is 16 days, of SPOT 26 days and of NOAA 4 hours. So
the latter satellite is said to have a _high_ temporal
resolution even though the spatial resolution is _low
_compared to the two other satellites! (1.1 km and 20-30 m)
* Spectral resolution.
Current imaging satellites usually have a broad band spectral
response. Some airborne spectrometers exist that have a high
spectral resolution; AVIRIS Airborne Visible/Infrared Imaging
Spectrometer (from NASA/JPL) has 224 bands, GERIS Geophysical and
Environmental Research Imaging Spectrometer has 63 bands.
E.g. if 100 Lux light gives DN 200 and 110 Lux yields DN 201 then
two samples from the original scene having 101 and 108 Lux will
both get the DN 200. Values from the range 100 up to 110 Lux can
not be distinguished.
======================== Image Formats (HTML) ======================
_Contributed by Wim Bakker (email@example.com)_
Subject: Image Formats
Image data on tape
Looking at the images stored on tape there's three types of
* Volume Directory, which is actually meta-information about the way
the headers/trailers and image data itself are stored
* Information about the images
This information can be stored in separate files or together with
the image data in one file.
This information can be virtually anything related to the image
+ Dimensions. Number of lines, pixels per line and bands etc.
+ Calibration data
+ Earth location data
+ Orbital elements from the satellite
+ Sun elevation and azimuth angle
+ Annotation text
+ Color Lookup tables
+ Etc. etc...
The information is often called a _header_, information _after_
the image data is called a _trailer_
* The pure image data itself
The image data can be arranged inside the files in many ways. Most
common ones are
* BIP, Band Interleaved by Pixel
* BIL, Band Interleaved by Line
* BSQ, Band SeQuential
If the pixels of the bands A, B, C and D are denoted a, b, c and d
respectively then _BIP_ is organized like
abcdabcdabcdabcdabcdabcdabcdabcdabcd... line 1
abcdabcdabcdabcdabcdabcdabcdabcdabcd... line 2
abcdabcdabcdabcdabcdabcdabcdabcdabcd... line 3
BIP can be read with the following pseudo-code program
FOR EACH line
FOR EACH pixel
FOR EACH band
I[pixel, line, band] = get_pixel(input);
_BIL_ looks like
aaaaaaaaaaaa... band 1, line 1
bbbbbbbbbbbb... band 2
cccccccccccc... band 3
dddddddddddd... band 4
aaaaaaaaaaaa... band 1, line 2
BIL can be read with the following pseudo-code program
FOR EACH line
FOR EACH band
FOR EACH pixel
I[pixel, line, band] = get_pixel(input);
aaaaaaaaaaaa... line 1, band 1
aaaaaaaaaaaa... line 2
aaaaaaaaaaaa... line 3
bbbbbbbbbbbb... line 1, band 2
bbbbbbbbbbbb... line 2
bbbbbbbbbbbb... line 3
cccccccccccc... line 1, band 3
cccccccccccc... line 2
cccccccccccc... line 3
dddddddddddd... line 1, band 4
dddddddddddd... line 2
dddddddddddd... line 3
BSQ can be read with the following pseudo-code program
FOR EACH band
FOR EACH line
FOR EACH pixel
I[pixel, line, band] = get_pixel(input);
Of course others are possible, like the old _EROS BIP2_ format (for
four band MSS images) where the image is first divided into four
strips. EROS BIP2 strips
Then each strip is stored like
aabbccddaabbccddaabbccddaabbccdd... line 1
aabbccddaabbccddaabbccddaabbccdd... line 2
To decode one strip the following pseudo-code can be used
/* The '%' character is the modulo operator */
/* Note that operations on 'i' are integer operations! */
/* Copyright 1994 by W.H. Bakker - ITC */
FOR EACH line
FOR i=0 TO BANDS*WIDTH
I[(i/8)*2+i%2, line, (i/2)%4] = get_pixel(input);
Subsequently, the strips must be glued back together.
Subject: Basic Processing Levels
What are the different types of image I can download/buy?
_Very brief - needs a proper entry_
Raw data (typically Level 0)
(as with other levels, annotated with appropriate metadata).
Only useful if you're studying the RS system itself, or data
Processed Images (typically Level 1, 2)
+ Radiometric correction - compensating for known
characterisitcs of the sensor.
+ Atmospheric correction - compensating for the distortion
(lens effect) of the atmosphere.
+ Geometric correction - referencing the image to Lat/Long on
the Earth's surface, based on the satellite's position and
viewing angle at the time of the acquisition. Uses either a
spheriod model of Earth or a detailed terrain model; the
latter enables higher precision in hills/mountains. Requires
Ground Control Points (GCPS: points in the image which can be
accurately located on Earth) for high precision.
The various part-processed levels are suitable for a image
processing studies. Most Remote Sensing and GIS applications
will benefit from the highest level of processing available,
Geocoded Projected Imagery (typically Level 3)
The image is mapped to a projection of the Earth, and in some
cases also composited (ie several images are mosaiced to show a
Images you can download from the net are likely to be browse
images. These are typically GIF or JPEG format, although a
number of others exist. Whilst providing a good idea of what is
in an image, they are not useful for serious applications. They
have the advantage of being a manageable size - typically of
the order of 100Kb-1Mb (compared to 100Mb for a full scene) and
are often available free. A browse version of any image (except
raw data) can be made.
Subject: Is there a non-proprietary format for geographical/RS images?
Is there a non-proprietary format for geographical/RS images?
The GeoTIFF format adds geographic metadata to the standard TIFF
format. Geographic data is embedded as tags within an image file.
For a detailed description, see the spec. at
Subject: Do I need geocoded imagery?
Do I need geocoded imagery?
In a recent discussion of mountain areas, John Berry
The problem that Frank has is that he is working in an area without
adequate maps: therefore, he cannot geocode his Landsat using a DTM, because
the data available is neither detailed enough or accurate enough to use as an
He can georegister the imagery using using one or two accurately
located ground control points and the corner-point positions given in the
image header: these are calculated from ephemeris data of, usually, unknown
accuracy (within +/- 1 km), but internal image geometry is good so an x,y
shift and a (usually) very small rotation can take care of everything to
better than the accuracy of his maps. Positions used should be
topographically low, and at the same elevation. GPS is the best solution, as
someone else pointed out, if Frank can get in the field.
The next problem is the parallax error introduced by the high relief.
In his situation, the only answer* is to get SPOT stereopairs and make a DTM or
DEM from them. Except in the case of very narrow gorges or slopes steeper
than 60 deg. there should be few problems with carefully chosen images (high
sun angles, etc). ERDAS has an excellent module for doing this. However, I
doubt that Frank has the budget. I believe ERDAS`s Ortho module would then
allow Frank to make an Ortho image that would be a perfectly good map.
*there may be some LFC or Russian stereo coverage in this area, which
would be a lot cheaper than SPOT but would require the use of analog stereo
Even if there were good topographic contour maps for all of Frank's
area, the cost of digitising these and turning them into a usable DTM would
probably be prohibitive (though there are outfits in Russia who might be able
to quote a price affordable to a large western company).
Subject: Imaging Instruments
How do Remote Sensing Instruments work?
If you put a camera into orbit and point it at the Earth, you will get
images. If it is a digital camera, you will get digital images.
Of course, this simplistic view is not the whole story.
Digital images comprise two-dimensional arrays of pixels. Each pixel
is a sensor's measurement of the albedo (brightness) of some point or
small area of the Earth's surface (or atmosphere, in the case of
clouds). Hence a two-dimensional array of sensors will yield a
two-dimensional image. However, this design philosophy presents
practical problems: a useful image size of 1000x1000 pixels requires
an array of one million sensors, along with the corresponding
circuitry and power supply, in an environment far from repair and
Such devices (charge coupled deices) do exist, and are essentially
similar to analogue film cameras. However, the more usual approach for
Earth Observation is the use of tracking instruments:
1. A tracking instrument may use a one-dimensional array of sensors -
one thousand rather than one million - perpendicular to the
direction of the satellite's motion. Such instruments, commonly
known as pushbroom sensors, instantaneously view a line. A
two-dimensional image is generated by the satellite's movement, as
each line is offset from its predecessor. If the sampling
frequency is equal to the satellite's velocity divided by the
sensor's field of view, lines scanned will be contiguous and
non-overlapping (although this is of course not an essential
_btw, would the above be better expressed in some ASCII
representation of mathematical notation?_
2. Another approach is to use just a single sensor. It is now not
sufficient to use the satellite's motion to generate an image:
cross-track scanning must also be synthesised. This is
accomplished by means of a rotating mirror, imaging a line
perpendicular to the satellite motion. These are known as scanning
instruments. This is somewhat analagous to the synthesis of
television pictures by CRT, although the rotating mirror is a
mechanical (as opposed to electromagnetic) device.
As the sensor now requires a large number of samples per line, the
sampling frequency necessary for unbroken coverage is
proportionally increased, to the extent that it becomes a design
constraint. A typical Earth Observation satellite moves at about
6.5 Km/sec, so a 100m footprint requires 65 lines per second, and
higher resolution imagery proportionally more. This in turn
implies a sampling rate of 65,000 per second for a 1000-pixel
swath. This may be alleviated by scanning several lines
Either design of scanning instrument may have colour vision (ie be
sensitive to more wavelength of light) by using multiple sensors
in parallel, each responding to one of the wavelengths required.
List of Imaging Spectrometers
Subject: What is a Sounding Instrument?
What is a Sounding Instrument?
_Answer posted by Wayne Boncyk (firstname.lastname@example.org) to
Satellite-borne remote sensing instruments may be used for more than
imaging; it is possible to derive information about the constituents
of the local atmosphere above a ground target, for example. One common
area of study is to observe atmospheric emissions in the spectral
neighborhood of the 183GHz water absorption line (millimeter-wave;
in-between microwave and thermal IR). These channels can be monitored
by an appropriate collection of narrow passband radiometers, and the
data that are returned can be analyzed to deduce the amount of water
vapor present at different levels (altitude layers) in the atmosphere.
The reference to "sounding" is an application of an old nautical term,
the investigation of the state of a medium at different depths
(original application: the ocean - specifically determination of the
depth of the ocean floor).
_Need a general entry here!_
Where can I learn about satellite orbits?
Wim Bakker has compiled a list of online references at
Wim adds the question _"When can *I* see a specific satellite"_, and
suggests the following pointers from his list:
* Visual Satellite Observer's Home Page:
* Satellite Observing Resources:
Satellite Orbital Elements
_Thanks to Peter Bolton (email@example.com) for this one!_
Jonathan's Space Report is at
http://hea-www.harvard.edu/QEDT/jcm/jsr.html. The introduction:
The Space Report ("JSR") is issued about once a week. It describes all
space launches, including both piloted missions and automated
satellites. Back issues are available by FTP from sao-ftp.harvard.edu
in directory pub/jcm/space/news. To receive the JSR each week by
direct email, send a message to the editor, Jonathan McDowell, at
firstname.lastname@example.org. Feel free to reproduce the JSR as long as
you're not doing it for profit. If you are doing so regularly, please
inform Jonathan by email. Comments, suggestions, and corrections are
How do I convert Landsat Path/Row to Lat/Long?
In response to this question, Wim Bakker wrote:
The SATCOV program is available by anonymous FTP from sun_01.itc.nl
(188.8.131.52). Here's how to get it:
$ ftp 184.108.40.206
ftp> idle 7200
ftp> cd /pub/satcov
ftp> mget *
If you can't use FTP, drop me a line and I will send a uuencoded version
Those of you who prefer a WWW interface can obtain it from the following URL:
Don't forget to set the "Load to local disk" option.
SATCOV is a PC program for converting Path/Row numbers of Landsat and
K/J of SPOT to Lat/Lon and vice versa. Furthermore it can predict the orbits
of the NOAA satellites, although I wouldn't recommend it for this purpose!
But that's an other can of worms....
Subject: Ground Stations
How is satellite data recieved on the ground?
_Intro to Ground Recieving Stations contributed by Peter Bolton
1. GROUND RECEIVING STATIONS
This document is an introduction to Ground Receiving Station (GRS)
acquisition and processing of remote sensing satellites data such as
SPOT, LANDSAT TM and ERS-1 SAR. Ground receiving stations regularly
receive data from various satellites so as to provide data over a
selected areas (a footprints approximately covers a radius of 2500 km
at an antennae elevation angle of 5 degrees.) on medium such as
computer tape, diskette or film, and/or at a specific scale on
photographic paper. GRS are normally operated on a commercial basis of
standard agreements between the satellite operators and the
Governments of the countries in which they are situated. Subject to
the operating agreements, local GRSs sell products adapted to end
users needs, and provide remote sensing training, cartography, and
2. GROUND RECEIVING STATION ARCHITECTURE
A Ground Receiving Station consists of a Data Acquisition System
(DAS), a Data Processing (DPS) and a Data Archive Center (DAC).
2.1. DATA ACQUISITION SYSTEM
DAS provides a complete capability to track and receive data from the
remote sensing satellite using an X/S-band receiving and autotracking
system on a 10 to 13meter antenna in cassegranian configuration. DAS
normally store fully demodulated image data and auxiliary data on High
Density Digital Tapes (HDDTs). However, in one small UNIX based
system, data storage can be stored directly on disk and/or
electronically transmitted to distant archives.
2.2. DATA PROCESSING SYSTEM
DPS keeps an inventory of each satellite pass, with quality assessment
and catalog archival, and by reading the raw data from HDDTs,
radiometrically and geometrically corrects the satellite image data.
2.3.DATA ARCHIVE CENTRE
The Data Archive closely related to DPS offers a catalog interrogation
system and image processing capabilities through an Image Processing
3. GROUND RECEIVING STATION PRODUCTS
The GRS products can either be standard or value added products. Both
are delivered on Computer Compatible Tapes (CCTs), CD ROM, cartridges,
photographic films or photographic paper prints at scales of 1:250
000, 1:100 000, 1:50 000 and 1:25000.
i. Standard products
- SPOT-1 and 2/HRV : data of CNES levels 0, 1A, 1B, 2A
- Landsat TM : data of LTWG levels 0, 5,
- ERS-1 SAR : Fast Delivery and Complex products.
ii. Value added products
- For SPOT
. P + XS : Panchromatic plus multi-spectral,
. SAT : a scene shifted along the track,
. RE : a product made of 2 consecutively acquired scenes,
. Bi-HRV : Digital mosaic produced by assembling 2 sets
2 scenes acquired in the twin-HRV configuration.
. Stereoscopy : Digital terrain model (DTM) generation,
. Levels 2B, S and level 3B using DTMs.
- For Landsat TM: levels 6, S and 7.
- For ERS-1 SAR : geocoded data.
- For any instrument:
. Image enhancement and thematic assistance,
. Geocoded products on an area of interest defined by the
customer (projection, scale, geocoding and mosaicking
according to the local map grid).
4. GROUND RECEIVING STATION OPERATION
Persons needing images for thematic applications in the field of
cartography, geology, oceanography or intelligence, etc, will refer to
the station catalog in order to find out if the data are available
over the area concerned.
There are two possibilities :
The data exists.
The customer fills in a purchase order and is then provided
with the product on a medium such as CCT, film or paper print.
If the data are available in the GRS catalog, a list of the
related scenes and their hardcopies (named "quick looks") are
The data does not exist.
a) For SPOT, the customer fills in a programming request form
which is sent by GRS to the Mission Control Centre (MCC) in
Toulouse, France. MCC returns a Programming Proposal to be
submitted for approval. Upon approval, the confirmation is
returned to MCC which in turn sends a programming order to the
satellite for emitting the data during its pass over the GRS
At the same time, MCC sends to GRS, the satellite ephemerides
for antenna pointing and satellite tracking.
In the case of SPOT, if the data does not exist within the
Station catalog but are listed in the SPOT IMAGE worldwide
catalog, GRS may request the level O product from SPOT IMAGE in
TOULOUSE in order to process it locally.
b) For other sensors, LANDSAT TM or ERS-1, the satellite
ephemerides are known at GRS and the antenna is pointed
accordingly in order to track all selected passes.
Within the GRS, the raw satellite data are received by the Data
Acquisition System (DAS), and recorded on High Density Digital Tapes
(HDDTs). HDDTs are then sent to the Data Processing System (DPS),
where an update of the Station catalog is made as well as a quick look
DPS is also in charge of automatic processing of selected raw data in
order to produce images of standard level.
Value added products with cartographic precision are produced within
DPS using interpretation workstations which must be part of an
operational Geographic Information System (GIS) combined to an Image
Processing System (IPS).
Once processed, the data, on CCT, are sent to the Data Archive Center
(DAC) where they are delivered to the customers after a quality
checking. At DAC, further processing may be applied to the data such
as image stretching, statistical analysis, DTM, or a conversion from
tape to film and paper prints in the photographic laboratory;
"customized services" may also be offered.
Subject: How can I assess my results?
How can I assess my results?
_(for basics, see Russell Congalton's review paper In Remote Sens.
Environ. 37:35-46 (1991). Think we should have a basics entry here
too!)_ Michael Joy (email@example.com) posted a question about
Contingency table statistics and coefficients, and subsequently
Second, a summary of responses to my posting about contingency table statistics
and coefficients. Basically, I need to come up with a single statistic for
an error matrix, along the lines of PCC or Kappa, but which takes into
account the fact that some miscalssifications are better or worse than others.
Tom Kompare suggested readings on errors of omission or commission.
Chris Hermenson suggested Spearman's rank correlation.
Nick Kew suggested information-theoretic measures.
Others expressed interest in the results; I'll keep them posted in future.
The responses are summarized below.
Your thinking is halfway there. Check out how to use an error matrix to get
of Omission and Commission.
Good texts that explain it are:
Introduction to Remote Sensing, James Campbell, 1987, Gulliford Press
start reading on page 342
Introductory Digital Image Processing, John Jensen, 1986, Prentice-Hall
start reading on page 228 or so.
These are the books where I learned how to use them. Sorry if you don't have
to them, I don't know how Canadian libraries are.
Illinois Natural History Survey
Champaign, Illinois, USA
Excerpt from my response to Tom Kompare (any comments welcome...)
These are useful readings describing error matrices and various measures we can
get from them, eg PCC, Kappa, omission/commission errors. But from these
I do not see a single statistic I can use to summarize the
whole matrix, which takes into account the idea that some misclassifications
are worse than others (at least for me). For example, if I have two error
matrices with the same PCC, but with tendencies to confuse different categories
I'd like to get a ststistic which selects the 'best' matrix (ie the best image)
One simple way I can think of to do this is to supply a matrix which gives
a 'score' for each classification or misclassification, and then multiply each
number in the error matrix by the corresponding number in the 'score' matrix.
So a very simple example of such a matrix might look like this:
Deciduous Conifer Water
Decid 1.0 0.5 0.0
Conifer 0.5 1.0 0.0
Water 0.0 0.0 1.0
In this notation, the 'score' matrix for a PCC statistic would be a diagonal
matrix of "1". Obviously there are a number of issues for me to think about
in using such a matrix, eg can you 'normalize' the score matrix? Can you
use it to compare different matrices with different numbers of categories?
An obvious extension to this would be to apply this idea to the Kappa
statistic as well.
Spearman's rank correlation is often used to test correlation in a situation
where you are scoring multiple test results. You might be able to adapt
it to your problem.
Chris Hermansen Timberline Forest Inventory Consultants
Voice: 1 604 733 0731 302 - 958 West 8th Avenue
FAX: 1 604 733 0634 Vancouver B.C. CANADA
firstname.lastname@example.org V5Z 1E5
C'est ma facon de parler.
Your question touches on precisely the field of research I'd like to be
pursuing, if only someone would fund it:)
> I'm comparing different datasets using contingency tables, and I would
> like to come up with summary statistics for each comparison. I am using
> the standard PCC and Kappa, but I'd also like to come up with a measure
> which somehow takes into account different 'degrees' of misclassification.
> For example, a deciduous stand misclassified as a mixed stand is not as
> bad as a deciduous stand misclassified as water.
I would strongly suggest you consider using information-theoretic measures.
The basic premise is to measure information (or entropy) in a confusion matrix.
I can send you a paper describing in some detail how I did this in the
not-totally-unrelated field of speech recognition.
This does not directly address the problem of 'degrees of misclassification' -
just how well it can be used to do so is one of the questions wanting further
research. However, there are several good reasons to use it:
1) It does address the problem to the extent that it reflects the statistical
distribution of misclassifications. Hence in two classifications with
the same percent correct, one in which all misclassifications are between
deciduous and mixed stands will score better than one in which
misclassifications are broadly distributed between all classes.
Relative Information is probably the best general purpose measure here.
2) By extension of (1), it will support detailed analysis of hierarchical
classification schemes. This may be less relevant to you than it was
to me, but consider two classifiers:
A: Your classifier - which for the sake of argument I'll assume has
deciduous, coniferous and mixed woodland classes.
B: A coarser version of A, having just a single woodland class.
Now using %correct, you will get a higher score for B than for A - the
comparison is meaningless. By contrast, using information (Absolute,
not Relative in this case), A will score higher than B. You can
directly measure the information in the refinement from B to A.
> In effect I guess I'm
> thinking that each type of misclassification would get a different 'score',
> maybe ranging from 0 (really bad misclassification) to 1 (correct
I've thought a little about this, as have many others. The main problem is,
you're going to end up with a lot of arbitrary numerical coefficients, and no
objective way to determine whether they are 'sensible'. Fuzzy measures can
be used, but these are not easy to work with, and have (AFAIK) produced
little in the way of results in statistical classification problems.
> I can invent my own 'statistic' to measure this, but if there are any such
> measures available I'd like to use them. Any ideas?
Take the above or leave it, but let me know what you end up doing!
Michael Joy email@example.com
University of British Columbia, Vancouver, B.C., Canada
Subject: Is there a program to compute Assessment measures, including Kappa coe
Is there a program to compute Assessment measures, including Kappa
Nick Kew's assess.c (ANSI C source code to compute several assessment
measures, including PCC, Kappa, entropy and Mutual and Relative
Information) is available for download from the WebThing site,
http://pobox.com/%7Esatfaq/ or from the satfaq autoresponder (mail to
firstname.lastname@example.org with subject line "send assess.c").
_Old reference to Dipak Ram Paudyal's kappa program deleted, as the
FTP server is apparently no longer available._
Subject: How good are classification results in practice?
How good are classification results in practice?
The following detailed commentary was posted by Chris Hermansen
Mike Joy posted a question regarding irregularities between two
classifications, one derived from manual interpretation of
large-scale aerial photography, the other from a supervised and
enhanced spectral classification of Landsat TM imagery.
I've read several of the responses, and I just thought it time
to kick in my $0.02 worth, since I am quite familiar with both
of the classifications with which Mike is working.
First, Peter Bolton rattles off his experience in tropical forests
and chastises Mike for discovering what should have been obvious.
Well, Peter, the boreal forest is a much different beast than
what you're used to in Malaysia (I can attest from firsthand
experience in both cases). Classification from remotely sensed
data is generally quite reliable in the boreal forest, especially
given the vegetative nature of the TM-derived classification
that is Mike's second dataset. Detecting predominantly deciduous
from predominantly coniferous stands is (spectrally speaking)
pretty straightforward. Problems arise in mixedwood stands,
however, since the nature of the classification of proportion
is not necessarily the same and in any case any aggregative
techniques applied to the TM image prior to classification (eg
smoothing) could significantly alter the proportional balance.
Also, depending on the proportion of deciduous in a predominantly
coniferous stand, and the spatial distribution of deciduous trees
within that stand, the classifier may have difficulty detecting
the differences between mixedwood and younger pure coniferous
types. Furthermore, deciduous stands with coniferous understory
are classified as deciduous in Mike's first dataset but may
easily be interpreted as mixedwood stands in the TM image.
Secondly, on the subject of incorporation of field data, Mike's
second dataset has some ground truthing incorporated in the
Thirdly, on the subject of large numbers of classes in some
people's TM-derived classifications, remember that in many cases
these additional classes are derived by incorporating other
datasets (field measurements, other digital map data, DEM
information, etc). The people I've seen most test this envelope
are the folks at Pacific Meridan Resources; their TM-derived
datasets form only the first step of several. As Vincent
Simonneaux points out, most people stop at the first step.
So, in response to Mike's original questions:
> 1) Is it reasonable to expect a TM-based classification to accurately
> distinguish Coniferous and Deciduous forest? The area I am dealing
> with is boreal mixedwood forest in northeren Alberta, Canada. I had
> expected that the classification should at least be able to do this.
On the face of it, yes. But! You must ensure that your definition of
Coniferous and Deciduous forest is exactly the same in both cases (and
the prevailing definitions in use in Alberta don't exactly help out in
> 2) Do people out there have similar experiences, i.e. the actual
> accuracy being very much lower than the reported results, or major
> differences when comparing with different source of information?
Of course, this is a possibility; the most unreliable classes may
interfere in a nasty way between to datasets. You really need to ensure
that you are sampling the same population in both cases; then you need
to examine the distribution of errors among classes in both cases. In
your first dataset, you don't really have error estimates with which to
> understand that an air-photo-based forest inventory and a TM satellite
> are measuring different things, and that I shouldnt expect perfect
> but I would have thought they could agree roughly on the overall area of
> Coniferous or Deciduous forest. Ditto for two similar TM-based
> + classifications.
Once more, not necessarily. See the points above on coniferous understory
in deciduous stands and the basic definitions of coniferous/deciduous
There are, of course, really obvious errors that can occur, like using
pre-leaf or post-leaf images when trying to locate deciduous stands...
Sorry to go on at such length about this; I hope that my comments are of
interest to some of you.
Subject: I need to classify a mosaic of several images. How best to do it?
I need to classify a mosaic of several images. How best to do it?
David Schaub (email@example.com) posted a question on this. Here
is his summary of replies:
Some time ago I posed a question to this list with regards to classification,
rectification, and mosaicking. My original question was as follows--
>We need to georectify, mosaic, and classify several (3 or 4) Landsat TM
>scenes using ERDAS Imagine. The classification will need to show major
>land cover categories, such as bare ground, grassland, shrubby range,
>built-up, coniferous forest, broad-leaf forest, water, etc. In the past
>when we have done this the seams between images are quite evident in the
>classification. We would like to minimize differences between images, yet
>be asaccurate as possible in the classification of each image.
>My main questions are these -- Should we classify each image separately
>and then mosaic them, or should we mosaic the images first and then
>classify them? Can georectifying the images effect the classification?
>You can assume that images along a path will have the same acquisition date,
>however scenes on adjacent paths will have different dates (at least by two
>weeks). I will post a summary. Thanks in advance for your opinions :-)
This quickly generated a flood of responses. While there wasn't complete
agreement, the majority of respondents believed that I should first classify
the images, then do the rectification and mosaicking. Nearest neighbor
should be used when rectifying the classified image (or if the image data
are rectified before classification). Thanks to all who responded!! Comments
are summarized below:
I have done the same things you are attempting to do for my thesis work.
I think the best course of action would be to classify the images first, then
rectify the images and then merge or mosaic the images. Rectifying the images
before you classify may distort the spectral characteristics of pixels and
thereby influence your classification. Furthermore, the smaller the area you
are classifying, the more accurate the classification will be, so if you
mosaic a large area and then attempt to classify the mosaiced image, there
will be more confusion possible based on the heterogeneity of a larger area. I
hope this helps, contact me if i can be of further assistance.
Here's my 2c for what it's worth...
I classify TM scenes separately and then mosaic the classifications. My
classifications almost never have a seam in them...If there is a seam
it is usually due to the difference in the date of the scene. You have
to be careful though... you need to use the same method of classification
(plotting out feature spaces and elipses helps) for overlapping scenes.
Sometimes this is why people use the other method...
If you're going to do this the other way round...i.e. mosaic and then classify
scenes you will have to calibrate the scenes to radiance and then use some
kind of atmospheric correction before mosaicking them. This should in theory
minimize the difference in the spectral information between scenes....I would
avoid using any kind of histogram equalization ...although it may look nice,
you are loosing the original pixel information.
\\\ /ccccccc x\ Fiona Renton, GIS and remote sensing analyst
>>Xccccccc( < CALMIT, Conservation and Survey Division
/// \ccccccc\_/ University of Nebraska-Lincoln
'' ~~~~ firstname.lastname@example.org
What sort of classification? Pixels? Clusters? Polygons?
Higher-level features? If your classification units are homogenous
and shape is not important, you should clearly do it before mosaicing.
If not, you have a genuinely interesting problem, and will probably have
to your own research (starting at your local academic library, assuming
there is one :-)
Geo-rectification will have a small effect on classification due to the
resampling process. I can't help to much on classification part,
because that is not my area, but my feeling is that mosaicking
non-classified images may be easier than trying to match features in a
Ok, this is my area. You can not assume that images on the same path
are imaged on the same day, However, they could be. You should be able
to check the meta-data to find out if they were. The next path west
could have been imaged 7 days after the path of interest or 9 days
before and the next path east could have been imaged 9 days after the
path of interest or 7 days before, again check the meta-data. The next
chances are to add 16 days on to those numbers (i.e. 7 + 16).
This is true for Landsat 4 and 5 only (will be true for Landsat 7).
Yes the georectification process will affect the classification
results. My suggestion is to classify each individual image first and
then mosaic them together. I have done this before and it works well.
If you mosaic first and then classify you have to calibrate the data,
apply radiometric corrections etc... Not worth the trouble in my
opinion, and you probably won't get any good results.
The resampling technique (convolution) will affect the radiometric value
of the image and may not be suitable for adequate identification
aftrewards. To avoid visible seams, just go around the areas, try to
contour the natural groupings (classes after classification)
To resume, in my opinion, if you want good accurate results: Classify first
and after mosaic.
You definitely want to mosiac the 4 images first (into one file) and
then run the classification on that. Because of subtle differences in
the radiometric characteristics of each image, the classes in separate
classifications will rarely "line up" perfectly when mosaicked afterward.
> Can georectifying the images effect the classification?
Yes it can, depending on the resampling technique you use. When
rectifying the images, use Nearest Neighbor resampling as that
will ensure that original pixel values are used to create the
new rectified dataset. (Bilinear or Cubic will average the
original data, resulting in slight degradation.) I would:
1) Rectify the four images (use Nearest Neighbor)
2) Contrast balance them, using for example Histogram
Matching or another technique.
3) Mosaic the four contrast balanced scenes into one file.
4) Run the classification.
I hope this helps.
Manager of Training Services
In general you can't depend on the DN values from one image to the next
to be related. You should classify before your mosaic - in other words
mosaic the classification, not the images. Otherwise you mix unrelated
DN values into a signal classification which would be wrong.
Classification may be affected by geo-rectification. If the
geo-rectified image has the same pixels and pixel values as the
original, the classification should not be affected. However, this is
an unreal assumption. A geo-rectified image will almost always have
resampling - which means that pixels are either dropped or replicated -
unless a filter is applied (like bilinear or cubic convolution) in
which case the pixel values change as well.
If the classifier is single pixel based (like isodata) then the
classification is only affected by the resampling as the sigatures are
affected by the replication or dropping of values. If the classifier is
regional or global (like multi-resolution/multi-scale classifiers, or
region linking) then the classifiers may be affected to a greater
You can classify before or after geo-rectifiction and the results
will not be vastly different.
But the bottom line to mosaic at the very end.
Michael Shapiro email@example.com
NCSA (217) 244-6642
605 E Springfield Ave. RM 152CAB fax: (217) 333-5973
Champaign, IL 61820
Re Michael Shapiro's posting,
There is no doubt that that you cannot depend on the DN values from one
image to the next (especially with images from adjacent paths which are
taken on different dates (see Chuck Wivell's posting).
However mosaicing images which have been classified seperately may produce
unusual results ie trying to match classes from different images.
A suggestion would be to first try some kind of atmospheric correction on
the images, mosaic them and then classify them together. Assuming
i) you can do a credible atmospheric condition (using Dark Pixel
Substraction, Band Regression etc) plus, perhaps, correct the images to a
constant solar elevation angle
ii) the images from different paths were not taken on widely different
iii) (linked to ii) the ground conditions are similar for the images
from different paths
then the DN values between images should be comparable.
We are currently doing a statewide land cover classification for Mississippi
using TM scenes (10 of them). My responses for your questions:
1. We classified each scene separately - mainly because the dates differed
and in the cases where we had adjoining scenes taken on the same day, it
was decided that classifying a full scene was a big enough task in both
computer and human resources. If you had subscenes, it would not be too
bad. I would advise against mosaicking scenes before classifying - your
signatures for the same landcover class in the other scene(s) would be
different and it would be a nightmare. Matching techniques that changed
image pixel values would change your original data and corrupt your
2. We also georeferenced each scene before classification for the following
- georeferenced ancillary data sources (roads, streams, NWI, etc) were
used - including leaf-off TM scenes already in-house.
- the need to have maps to take into the field for pre and post
We used nearest neighbor. This doesn't change pixel values but just
moves them to a different location. In our case the image statistics
were unchanged after georectification although it is probable that some
pixels may be dropped or replicated (but when you georeference the
classified image, those same pixels are going to be affected anyway).
Bottom line would be to classify each scene separately. I would georeference
each TM scene first - when the classiciations are completed, stitching is
Our lab has had luck using regression techniques to mosaic the three
bands together. Using ERDAS imagine, the steps are:
1) create an image where the two scenes overlap (this is best
done with modeller, not layerstack: layerstack only
uses the geographical boundaries, whereas you want to
have the area where there are values in both images
2) Use the Accuracy Assessment module to create random points
on the image and remove those points which lie in cloud
3) Export the X,Y coordinates from the random points and use
these as a point file in the Pixel-to-Table function.
Use the overlap image as the output image (make sure
you have all the bands you want to regress (ie. image
one's band 3,4,5 on top of image 2's 3,4,5
4) You now have a set of points that can be imported into
any standard statistical package. You need to have the
values from the "larger" or primary image be the Y values
and the other image be the X value (I'm told the correct
statistical term is that the Y is the master and the X
is the slave).
This should create a seamless image. Obviously, the closer the B
number in the Y= bx + constant equation is to 1, the less you are
transforming the values of your slave image. We have also tried doing
classifications of each image first, but the results have been
University of Maine