Path: senator-bedfellow.mit.edu!faqserv

Path: senator-bedfellow.mit.edu!faqserv
From: [email protected] (Nick Kew)
Newsgroups: sci.image.processing,sci.geo.meteorology,sci.geo.eos,sci.answers,news.answers
Subject: Satellite Imagery FAQ - 3/5
Supersedes: <sci/Satellite-Imagery-FAQ/[email protected]>
Followup-To: poster
Date: 17 Mar 1997 11:24:38 GMT
Organization: none
Lines: 1167
Approved: [email protected]
Expires: 30 Apr 1997 11:24:01 GMT
Message-ID: <sci/Satellite-Imagery-FAQ/[email protected]>
References: <sci/Satellite-Imagery-FAQ/[email protected]>
Reply-To: [email protected]
NNTP-Posting-Host: penguin-lust.mit.edu
Content-Type: text/plain
Summary: Satellite Imagery for Earth Observation
X-Last-Updated: 1996/12/17
Originator: [email protected]
Xref: senator-bedfellow.mit.edu sci.image.processing:26847 sci.geo.meteorology:33267 sci.geo.eos:2244 sci.answers:6111 news.answers:97369

Archive-name: sci/Satellite-Imagery-FAQ/part3

This document is part of the Satellite Imagery FAQ

------------------------------

Subject: Image Basics

Image Basics _Contributed by Wim Bakker ([email protected])_

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
* Space
One sample covers only a very small area from the real scene.
* Time
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:

------------------------------

Subject: Resolution

* 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
_detected_!
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
second.
+ 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.
* Quantization.
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 protected])_

------------------------------

Subject: Image Formats

Image data on tape

Looking at the images stored on tape there's three types of
information
* 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
data
+ 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
+ Histograms
+ 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
..
abcdabcdabcdabcdabcdabcdabcdabcdabcd...
abcdabcdabcdabcdabcdabcdabcdabcdabcd...

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);

_BSQ_ shows
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
processing systems

Processed Images (typically Level 1, 2)
Processing includes:

+ 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,
including geocoding.

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
larger scene).

Browse Images
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.

Stereopairs

Multitemporal Images

------------------------------

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
http://www-mipl.jpl.nasa.gov/cartlab/geotiff/geotiff.html

------------------------------

Subject: Do I need geocoded imagery?

Do I need geocoded imagery?

In a recent discussion of mountain areas, John Berry
([email protected]) wrote:

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
input.

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
comparators (probably).

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

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
maintenence!

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:

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
property).
_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
simultaneously.
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

http://www.geo.unizh.ch/~schaep/research/apex/is_list.html

------------------------------

Subject: What is a Sounding Instrument?

What is a Sounding Instrument?

_Answer posted by Wayne Boncyk ([email protected]) to
IMAGRS-L_

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).

------------------------------

Subject: Orbits

Orbits

_Need a general entry here!_

Where can I learn about satellite orbits?

Wim Bakker has compiled a list of online references at
http://www.itc.nl/~bakker/orbit.html.

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:
http://www.rzg.mpg.de/~bdp/vsohp/satintro.html
* Satellite Observing Resources:
http://www-leland.stanford.edu/~iburrell/sat/sattrack.html

Satellite Orbital Elements

_Thanks to Peter Bolton ([email protected]) 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
[email protected]. 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
encouraged.

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
(192.87.16.8). Here's how to get it:

$ ftp 192.87.16.8
Name: ftp
Password: your-email-address
ftp> bin
ftp> idle 7200
ftp> prompt
ftp> cd /pub/satcov
ftp> mget *
ftp> bye
$

If you can't use FTP, drop me a line and I will send a uuencoded version
by email.

Those of you who prefer a WWW interface can obtain it from the following URL:
http://www.itc.nl/~bakker/satcov
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
<[email protected]>_

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
thematic applications.

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
System (IPS).

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
of
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
provided.

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
antenna.
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
processing.

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.

_________________________________________________________________

Image Interpretation

------------------------------

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 protected]) posted a question about
Contingency table statistics and coefficients, and subsequently
summarised replies:

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.

===============================================================================
Michael:

Your thinking is halfway there. Check out how to use an error matrix to get
+ errors
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
+ access
to them, I don't know how Canadian libraries are.

Tom Kompare
GIS/RS Specialist
Illinois Natural History Survey
Champaign, Illinois, USA
email: [email protected]
WWW: http://www.inhs.uiuc.edu:70/
============================================================================

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
+ readings
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)