For generating an interferogram, two co-registered SAR images covering the same area are multiplied in a complex fashion. The result of this complex multiplication is the average of the two SAR images and the difference of their corresponding phase values.
A synthetic aperture radar is an active sensor transmitting and receiving microwave signals, i.e. measuring
distances between the sensor and the point on the Earth's surface, where the signal is backscattered,
perpendicular to the flight direction.
A digital elevation model (DEM) is an image consisting of heights above sea level for a certain location. It shows terrain, viz. mountains, valleys and plains.
This example was taken from ASF ERS-1 SAR image number 8154400, covering the Prudhoe Bay, Alaska area on January 21, 1992. It compares a strip of complex data, that data with each five lines averaged, and the corresponding image obtained via standard processing.
The coherence is a measure for the correlation of the phase information of two corresponding signals and varies
in the range of 0 to 1. The degree of coherence can be used as a quality measure because it significantly
influences the accuracy of phase differences and height measurements.
A baseline is a distance between two satellites that can be represented by its length and orientation or
separated into two cartesian components or expressed in polar form. The term "temporal baseline" refers
to the time interval between sucessive satellite observations.
The User Services Office is the main point of contact for the Alaska
Satellite Facility, providing a wide range of products, services
and information to the SAR user community.
Digital elevation model of the Delta Junction area. The mosaic was calculated from 15 pairs of ERS tandem mission SAR images with a ground resolution of 30 meters by the ASF Advanced Product Development group.
Due to its side looking geometry, a radar sensor measures the distance from the satellite to a point on the ground.
This distance is referred to as "slant range". In contrast to this perspective, the SAR image can also be projected
such as maps or optical images. In this case the distances are in "ground range".
Phase can only be detected between - and ,
but the actual phase shift between two waves is often more than this. Phase unwrapping is the process
of reconstructing the original phase shift from this "wrapped" representation.
A synthetic aperture radar (SAR) works by illuminating the Earth with a beam of coherent microwave
radiation such as a laser. This radiation can be thought of as a wave, such as a water wave or a sound wave.
As you may remember from trigonometry or physics class, a wave can be described by just three properties:
its wavelength, amplitude, and phase.
NOTE: ASF is capable of processing all RADARSAT-1 beam modes to level 0 products. At this time calibrated level 1 data products are only available for standard beams, ScanSAR Wide B, and Fine Beam1. The remaining ScanSAR beams are available as uncalibrated products. NASA approved U.S.
ASF's ERS-1 SAR data archive runs from September 1991 to June 1996; ERS-2 data coverage began in October 1995. The JERS-1 SAR data coverage began in May 1992, and ASF began archiving RADARSAT SAR data in June 1996.
The RADARSAT-1 imaging window has returned to its nominal orbit. There are no significant geolocation differences between pre-anomaly and post-recovery data over the Delta Junction targets.
Introduction
On April 25, 2004 during nominal orbit maneuvers, RADARSAT-1 experienced what was termed a “super-thrust” anomaly.
First the ice tracker algorithm must select two input images:
Reference Image - the earlier image of a particular region (usually three days before the target image, as the ERS-1 satellite retraces its path in that time)
The 14 weeks of my internship in Fairbanks have been the longest period of time that I have been away from "home" so far, and it was worth every single day!
The first thing I noticed was on my arrival at 10 pm that the sun was still way from setting. In fact it was so `bright' that I did not fall asleep until 3 o'clock in the morning...fortunately I had planned for a spare day before actually starting to work. When Jason Williams picked me up from the airport I learned that we would be four interns this summer, so that promised to be a good start for first contacts.
I was also in the fortunate position to stay on campus with a couple of other interns of the Geophysical Institute, and we did a lot of stuff together, most noteable evening activities plus a few weekend trips. It was great fun! For future interns I can highly recommend staying with GI interns.
The first highlight at work was the trip of us four interns north to the Toolik Lake Reflector Array all on our own, since both Jason Williams and Parker Martyn attended a conference in Nebraska. It was a great experience for me first of all to drive on the Haul Road and to be north of the Arctic Circle, eventually arriving as far north as Prudhoe Bay on the Arctic Ocean. Unfortunately I did not have a social security number at that time, so I could not join the other three guys to work on the oilfield to set up a reflector and hike to the shore; instead I had to pay sixty dollars to go on a tour to the oilfield and the Arctic Ocean. On that trip we saw a lot of wildlife: grizzly, moose, caribou, birds and more! When I was in the Brooks Range again two months later with the GI interns the only things we saw was a golden eagle and a couple of ground squirrels. Also the mosquitoes were not too bad in late May. In other words : Great!!
Working with everybody else at ASF was not boring a single minute. I can only agree with previous interns : It's a perfect atmosphere to work, "a big family"! Although we did hard work we did not lose our sense of humor. In fact it was in the first week that Jason Williams told me "You're having TOO much fun here!". Since he grinned heavily I supposed he did not really mean it. I was right.
Last but not least I have to express my gratitude to a couple of people.
First have to be mentioned Jason Williams from ASF and Werner Wiesbeck from the University of Karlsruhe, Germany. Due to their contact only was it possible for me to do my internship in Alaska! To Jason Williams, Parker Martyn, Ty Sullins, Yinghong Sun and Ylva Timner, who had to endure me for 14 weeks and live to tell, and gave me such a wonderful time at work. Furthermore all the GI interns at the dorm for having such a nice summer with me. I really enjoyed it! And of course everybody at ASF for giving me such an extremely good time!
Special thanks go to Jason Williams for letting me borrow his bike and Ty Sullins for a couple of rides he gave me in his car.
The Alaska SAR Facility is affiliated to the Geophysical Institute at the University of Alaska Fairbanks. Amongst its activities are downlinking, processing, archiving and distributing SAR data from ERS-1 (now decommissioned), ERS-2 (being calibrated), JERS, and RADARSAT (to be calibrated).
Available SAR products include:
-full resolution images (1 pixel represents 12.5 x 12.5 m2)
-low resolution images (1 pixel represents 100 x 100 m2)
-complex images (containing amplitude and phase information for each pixel)
ASF is one of several Distributed Active Archive Centers (DAACs) sponsored by NASA as part of the Earth Observing System initiative.
The Synthetic Aperture Radar is mainly used for remote sensing of the earth by aircraft and satellites. It allows a small resolution in range and along-track (azimuth). The radar waves scanning the earth's surface have the advantage of being little influenced by atmospheric changes. This permits observation of particular areas of the earth, e.g. rain forests and polar ices. The observation of those areas is nearly impossible in a visible or infrared spectrum because of fog and clouds covering humid areas most of the time.
To achieve a high resolution in high altitudes using conventional means would require a very large antenna to generate a narrow beam, which had to be carried into an orbit by spacecraft. This is of course impossible.
To increase the along-track resolution with antennas of only 10 meters length SAR was developed. It observes each point for a longer period of time, which requires an additional processor system, simulating a long antenna (hence "Synthetic"). The range resolution depends only on the time length of each impulse. A shorter impulse increases the resolution in range.
An interesting point is that the best resolution of SAR that is possible always equals half of the antenna aperture and is independent of all other parameters like wavelength or distance to the target.
SAR images are processed by using amplitude and phase information of the received signal. A large number of parameters can be influenced during this process. If they aren't set correctly, the image could be useless for further observations. Here are some examples of important parameters affecting SAR image processing:
Position, altitude and velocity of the SAR sensor
Movement of the earth
Ground profile (mountains)
Atmospheric refraction
Variable attenuation in the atmosphere
Perspective distortions
Some of these parameters can be calculated, but for others there are no equations available. Parameters that do not have equations can only be determined by detailed field measurements.
The ASF CalVal group works with two different kinds of calibration:
a)geometric calibration is used to assure that the location of a target in the SAR image matches the real (geographic) location, which is determined using GPS.
b) radiometric calibration is used to assure an accurate measuring of the backscattering property of the target. This knowledge about a certain target allows one to gain a lot of additional information.
Validation is basically the same as calibration, only that it is done while the sensor is up and running to assure and maintain the image quality after calibration is done.
For the calibration and validation of SAR satellites ASF uses Corner Reflectors. Their advantage is a high power output with a wide 3dB-width plus rather easy mathematics to calculate the radar cross section (rcs). These reflectors are used for geometric and radiometric calibration, although for this task only the radiometric performance is of importance.
ASF has two different reflector arrays, around Delta Junction and Toolik Lake. For radiometric only the array at Delta Junction is used since the reflectors need periodic maintenance in order to yield reliable radiometric results.
There are many different forms of Corner Reflectors. Those used by ASF can be generally classified as Triangular Trihedral Corner Reflectors (TTCR). For matters of simplicity I will refer to them hereafter simply as Corner Reflectors (CR).
Over the years ASF has used three different generations of corner reflectors: Gen I, Gen II, and Gen III.
These were the first ones used for calibration/validation purposes, dating back to 1992. As can be seen below it consists of three sides of triangular aluminum panels. Each panel has small holes in it that reduce the weight and allow wind and rain to penetrate it with reduced harm. The reflector is fixed in its position in that it cannot be set for different passes or beams. This meant that for ERS-1 there had to be two reflectors at a location to point for ascending and descending passes.
There were 5 reflectors pointed for descending and 3 for ascending passes.
For RADARSAT it became necessary to calibrate 20 different beams, so the old Gen I reflectors could not be used effectively. Thus Gen II was created. Apart from a base that allowed the reflector to be turned to any heading and inclination, the major difference to Gen I is its design: The panels were made of solid aluminum in a pentagonal shape; this promised to make it less susceptible to background noise [1]. Unfortunately there is a heavy warp in the panels which probably was created while they were cut to their shape. They were deployed in addition to the Gen I in August 1995.
When it was discovered that the Gen II reflectors would not perform as good as was hoped for, ASF decided to bring the Gen I reflectors back from Delta Junction, reinforce them by welding metal bars to the back of each panel to increase stability and reduce the curvature, and mount them on the same base as the Gen II to allow them the same maneuverability. Thus, some sites then needed only one reflector as opposed to two Gen I, which allowed to put the spare reflectors on some new sites. This was done in spring 1996, and all but one Gen II reflector were removed.
These are mentioned here just for the sake of completeness, as they are used only for geometric calibration. In summer 1996 we deployed some pentagonal shaped reflectors fitted to the base of Gen I (so it is neither a true Gen I nor a true Gen II) and smaller Gen I types (length 1.8 m instead of 2.4 m).
Delta Junction is located about 160 km southeast of Fairbanks at the confluence of Delta and Tanana River and at the official end of the Alaska Highway.
The old array consisted of finally 11 different reflectors: DJ1/2, 3/4, 5, 7, 9/10, 11, 13, 15. Currently (i.e. as of July 1996) there are 14 reflectors located within a radius of 50 km around Delta Junction, of which 9 are used for radiometric calibration. DJR1 to DJR9 are Gen III reflectors with the exception of DJR3, which is a Gen II.
To analyze images of the Delta Junction Array the following steps had to be taken: -search for passes of ERS-1 over Delta Junction -see if there is an image available for that pass -order the image -analyze the image -analyze the data For the first three steps I relied heavily upon Jason and Parker. The most common problem with an image was that not all reflectors
could be seen, either because the image did not cover the whole array or due to mispointing of the reflectors. The latter reason was of course true for the old DJ array (as most of them could only point in one fixed direction), while for the DJR array we could not point every reflector for every pass simply because it would have taken too much time. Each maintenance trip to Delta Junction takes at least 9 hours for all reflectors!
Some of the images were already analyzed by Parker, so I just had to update the database with the worksheet data. One measurement for DJ 3 (Day 1995:243:20:55:27) though had to be discarded because the analyzed object on the image proved to be not the reflector but some other structure!
The main problem though for the analysis was the fact that ERS-1 was turned off for good on June 2nd (which was beyond our control and responsibility). This is the reason for the small number of entries for the DJR array!
Analyzing SAR Images for Performance of Corner Reflectors
The Calibration/Validation (CalVal) team uses a software called `disa version 1.04' to analyze images. By entering the measurements obtained in the field, i.e. heading and inclination, disa can compute the theoretical rcs of a given CR. By specifying the coordinates of the reflector on the SAR image disa can then compare this result to the actual measured power by the following steps:
Each pixel on a SAR image represents a digital number (DN) of 0-255. The rcs is determined by using the formula
[sigma]0dB = 10*log10 [a2*(DN2-a1*n(r))+a3]
[sigma]0dB : normalized radar cross section in dB a1, a2, a3 : noise scaling/image scaling/absolute offset parameters that are obtained from the CEOS leader file
n(r) : noise of range, obtained from the CEOS leader file The power of a CR is calculated according to
Theo-rcsdBm2 : Theoretical rcs, which can be calculated by using a formula. Parameters include heading, inclination, and physical size of the
reflector Corr-factordB : Theoretical Correction Factor between theory and fact (for ERS-1/2, RADARSAT : -1.9 dB)
Proc-gaindB: The processor gain, currently -18.00 dB to prevent saturation CR powerdB : The measured reflected power of the CR minus the background noise, in dB
NormalizerdBm2 : A term to normalize K in respect to area, equals 10*log10[area of pixels]
= 21.9382 dBm2in this case The target value for K is -49.21 dB for ERS-1 data processed on the ASP. The K values for each reflector were entered into a worksheet and from there into a database; most of the entries for the DJs were
already entered by previous intern Volker Kaltenborn, so I could use his database with slight modifications.
All data was entered into a database using Microsoft Excel 5.0.
I entered for each measurement the image id #, kind of pass, CR #, K Value in dB, Revolution #, and Julian Overflight Day and Year. By using references the actual date was computed, and the K measurement listed in a column for each reflector converted to linear scale.
I used two different worksheets to enter data for the old DJ array and the new DJR array. Thereby it was easy to make a difference between for example DJ 1 and DJR 1.
In a separate worksheet I defined the range for `summer' and made three tables where I defined the dates in which a given reflector belonged to which Generation; DJ 7 for example was first a Gen I and later a Gen II, but for the time it was Gen II no summer data was available; the first summer data for that reflector was available after it was renamed as DJR 3.
By applying database functions the mean K values were calculated in linear scale. These were then converted back to dB. For standard deviation the K values in dB were compared to the mean K value in dB. The same method is used by JPL.
The charts were created on a worksheet of their own. I created more charts and graphs than I include in this report; the reason for not giving them here is mainly that they are neat to look at, but don't really allow you deduct any conclusions.
First there can be error sources which are independent of the reflectors : they can occur in the sensor subsystem, platform and downlink subsystem, and signal processing subsystem [2].
Since this analysis is based on the calibrated ERS-1 sensor, I assume these errors to be negligible.
Error sources for the reflector include manufacturing errors (e.g. bent panels), heavy wind and weather susceptibility, background noise, and false readings for reflector heading and inclination. Fortunately the CR have a 3 dB-width of approximately 40 degrees, so the latter source is again neglected for this purpose.
The weather has great impact on the K value; snow inside the CR yielded K values of -30 dB and higher! For this reason, I included only data from between julian day 90 and 259, which so far have proved to be snowfree.
What the Results of this Analysis tell us and what they do not
For the reasons mentioned above all factors are neglected except those based on manufacturing, wind and weather except snow, and background noise. Therefore the results give us an estimate for the performance of a given CR for a given site. To further discern between a bad reflector and a bad site (i.e. high background noise) I had to take into account the direct surrounding of the reflector: What kind of vegetation grows there? Is it highly reflective for the given SAR frequency (in this case 5.3 GHz)? Are there any structures close by that influence the reflected power, for example silos?
Unfortunately the effect of the surrounding cannot be quantified, so in most cases the only thing that can be said is "This reflector is (not) suitable for this site!" or "This is a bad site" (due to highly reflective crop).
To evaluate the different Gens of reflectors I had to look for a trend within each generation. One major problem there is if the results for some of them are acceptable and others just don't behave themselves at all. Is it simply due to a combination of good sites and bad sites? Substantial differences between the reflectors within the same design?
This could be solved if different generations were deployed on the same site, which is true for the pairs DJ5-DJR2, DJ7-DJR3, DJ11-DJR5, DJ13-DJR6, DJ15-DJR7.
These results were obtained using Microsoft Excel 5.0 on Macintosh.
A few remarks on the charts:
All of these charts are compiled from data between day 90 and day 259 of any given year, i.e. 1992-1995. In 1996 the data goes from day 90 to day 148; after that date there was no data available for my analysis.
For this reason, to compare different reflectors on the same site Figure 10 consists only of data between day 90 and day 148, to exclude too much influence from vegetational growth over a given year.
For calcualtion of the Mean K Value and Standard Deviation see section 2.4.
A blank entry in a table/chart means that there is no data available for that particular field. In the case of DJ4, which was a Gen I, there was no summer data available at all. Usually it means that that particular reflector did never belong to the given Gen.
Fig 6. The Old DJ Array
Fig 8. Deviation from the Target K Value ( = -49.21 dB)
First, I want to stress again that this analysis and the interpretation applies only to the radiometric performance of the reflectors. For geometric calibration almost anything will do, as long as it does not move!
As can be seen from Figure 6 and Table 1, there is sufficient data for each reflector to evaluate it.
Most importantly, all reflectors have a mean K value of -49.21 dB +/-1 dB, which is defined as acceptable according to the official calibration plan of ASF [3]. Moreover three reflectors (DJ2, 5, 7) show a value that is very close to the target value. Since each valid summer measurement for these has a deviation of less than 1 dB from the target K, these results are quite stable (Figure 8).
The worst results in both respects is yielded by DJ9, which was located near Donnelly Dome, close to the site of the current DJR4. On most of our maintenance trips we had a stiff breeze to heavy wind up there. DJ10 was located there too, which explains certain deviations of more than 2 dB.
Therefore my conclusion is that Gen I is a good choice if there is need only for one target heading and inclination, or if there is enough room to deploy more reflectors. Although this is not efficient, the results can be expected to be reliable and constant, provided the site is not too bad.
If however as in case of Donnelly Dome there is heavy wind on the site, the results should be treated with care, because one given measurement of K can easily fall below the deviation level of -48.21 dB.
The analysis of Generation II proved to be the most difficult. Because I limited the valid results to be between day 90 and 259 there was only one measurement for DJ 11/13/15. Thus no final evaluation of Gen II can be done, as there were only 5 datatakes for DJR3 as well, so its failure could be the fault of that particular reflector rather than the problem of the whole Generation.
Yet the one measurement for DJ11 gives the idea that Gen II could follow the theoretical advantage over Gen I, since it is lower than the target K value.
Looking at Figure 9 you will find DJ7 compared to DJR3; both were located at Shaw Creek Farm. It shows clearly that the problem is more the Gen II reflector than the site, because DJ7 produced acceptable results. DJR3 is completely useless for radiometric calibration (though it might be used for validation by looking on deviations from its mean value).
In other words, Gen II might be an option. If the manufacturing problems can be circumvented to substantially reduce the curvature of the panels I would give the Gen II concept at least another try!
As the current design of Gen II makes the panels of solid aluminum without holes, I would not recommend them for sites like Donnelly Dome, as they are far more susceptible to wind influence than Gen I.
The only major difference between the analysis of Gen II and Gen III is that there are more reflectors of which a few measurements were available. It is hardly enough to make a statement on the stability of the results.
Still, it is obvious that DJR7 and DJR8 are way off the target value. Partly for this reason DJR7 has already been moved to another site, close to where DJ3/4 used to be.
According to Figure 8 and Table 1 I suggest that DJR1 and DJR6 are not too heavily relied upon for radiometric calibration, as by now it is clear that deviations of more than 1 dB do happen!
The good news is that so far DJR2, DJR4 and especially DJR9 perform as desired. Therefore these should be used as primary means for current calibration/validation ; but I have to stress again that this evaluation is based on three to five measurements!
In summary I deem Generation III at least not inferior to Generation I. A part of the problems is the location of some DJRs. Yet comparing the two Generations on Greg Pippin's Farm (DJ5 / DJR2) we can see that Gen III can perform as desired (Figure 9). I suppose that one advantage over Gen I will be a lower standard deviation, although these few measurements do not really allow more than expectation.
"Distributed target calibration refers to external calibration using natural targets of large areas with homogeneous backscattering properties. A fundamental assumption is that the scattering properties of these areas are stable or that the variation is well characterized."[2]
In order to analyze targets for their possible use as distributed targets, a software had to be written that analyzes the backscattering properties of a given area.
Since nature almost never will do us the favor of rectangular or circular distributed targets the software has to be able to analyze a polygon of any shape.
First I thought : Hey, any drawing program can cut/fill polygons of any shape, so it should be easy once I have the coordinates of the corners!
So I started with coding the input of the corners and calculating the edges of the polygon. My intention was to define a mask where every TRUE meant `include this pixel' and FALSE `leave this pixel blank'. Then the problems started : How can the program find out which part is the inside of the polygon? One option might have been letting the user enter a coordinate of an inside pixel, but even then it would have taken much coding, trying, swearing and debugging. In short after the first day I decided that I do NOT have to invent the wheel a second time.
Therefore I had to find an algorithm that allows to cut any polygon from any given image. Jason suggested to follow the algorithm outlined in [4]. It is quite simple and has the only drawback that the bottom line/corner is omitted from the clipped part. As that is not of too much importance for this purpose I decided not to try to fix that problem.
When poly is run, it first displays the status of the switches (see below) followed by the version number, which at the moment should
read v1.61.
It then asks for the image filename which requires the FULL filename (e.g. "208361100.dat")!
Next thing is the width and height of the image. By popular belief these are for lo-res 1024x1024, hi-res 8192x8192, complex 2048x12800. Note that you have to enter these values seperated by a comma!
After that you have to enter some image-specific data concerning the image file format:
Length of header record in bytes : for lo-res 1036, hi-res/complex 8204
Length of row headers in bytes : 12 for all types
However, if you want to clip a raw image file like the lo_res.dat that disa uses, the headers are each 0 bytes of course.
Fig. 10
After specifying these variables poly requests the coordinates for each corner of the polygon. From the very start the number of corners was not to be limited by the program so the user has to enter a special `End of Corner' mark (-1,-1) when finished.
Then the data mask is created. For details on how this is done see the description of Filling Polygons in [4].
A field is then defined that is large enough to hold the clip. For this purpose the lowest and highest x- and y-coordinates are taken, thereby defining the maximum width and height of the clip that is possible (which is the case if the polygon is a horizontal or vertical rectangle). For visualization :
Fig. 11
Once this has been completed the image data file is opened for reading. The program skips the header bytes and copies only those pixels for which the mask entry reads TRUE, otherwise zero is stored. In order to save memory the image is not stored (currently a full-res SAR image takes 64 Mbytes !!).
Thereafter the mean value for the normalized rcs of the polygon is determined. This is done by using the calibration formula:
[sigma]0 = a2*(DN2-a1*n(r))+a3
The a1/a2/a3/n(r) are read from the CEOS leader file accompanying the image file. Again for the sake of memory the rcs value for each pixel is not stored. The sum of every pixel-rcs is divided by the number of pixels of the polygon, giving the mean value in linear scale, i.e. not in dB.
After the mean value the standard deviation is computed in linear scale. Since the rcs was not stored the calibration formula has to be used once again.
he formula used for the standard deviation is:
The clip is saved in raw image format, i.e. width*height bytes each representing one pixel with a brightness value between 0 and 255.
Finally poly generates an ascii output of the vital statistics of the clip on screen as well as in a file.
The default for the output image file is ALWAYS the input image filename that you specify plus the extension ".clip". The default for the ascii and histogram (see below) files are derived from the output image filename by appending ".ascii" and ".hist" respectively. In other words, if you don't accept the default for the output image filename but enter "Alex.is.a.moron", the default for the ascii file is "Alex.is.a.moron.ascii" and for the histogram file "Alex.is.a.moron.hist". A change in the ascii filename will NOT affect the default for the histogram file! To accept defaults press <return> when prompted.
Included in poly are a couple of switches. The general call to poly from a Unix shell should be:
poly [-c] [-f] [-h] [-i inputfile] [-l leaderfile] [-s]
-c : Image is a complex image. Make sure you use this ONLY for complex image data files!
-f : Turn on the fixed scaling. When used, poly saves the polygon you defined with exactly the same DN value as in the original file (exception : Complex data is converted by dividing the I^2+Q^2 by roughly 360, thus yielding an 8-bit value).
IMPORTANT : If this switch is NOT used, poly adjusts the DN values of the clip to make full use of the range of 256 (i.e. real images suffer multiplication of their DN values).
This means that the a1/a2/a3 and noise values are NOT (repeat N O T) valid for the values in the saved clip image !!!!!
-h : poly will create a histogramfile for the polygon. It will ask for a filename, again with a default which you can accept simply by pressing <return> when prompted.
This switch will be ignored if used in conjunction with -c.
-i : poly will (try to) get all its input data from the inputfile you specified instead of stdin
(=keyboard). If there is insufficient data in the file poly will probably quit with an
error! (So far -i is redundant to "poly <inputfile" in Unix)
The format for the inputfile should look like this:
line # description example
=========================================
0 image filename 208361100.dat
1 width,height 8192,8192
2 header record length 8204
3 row header length 12
4..n-1 Coordinates x,y 114,2093
388,2207
412,2593
.
.
.
n End of Coordinates -1,-1
n+1 Output Image Filename <return>
n+2 Output Ascii Filename <return>
{optional n+3 Output Histogram Fln. <return> }
-l : leaderfile is the name of a file that poly should use to determine the a1/a2/a3/noise
for that particular image you want to mutilate. If you don't use this switch, poly
takes the filename of the image and replaces .dat with .ldr
-s : save-only. When used, poly saves only the polygon and does not calculate mean
values/std deviation, nor does it need a CEOS leader file! In other words,
if you just want to cut out some weird shape out of an image, use this switch
...and don't worry about the LOG error reports you get at the end, they do
Figure 12 and 13 show sample clips from a lo-res image. Both were created using the -f switch; however this was necessary for Figure 12 in order not to get a pitch black clip. Still, due to the bright spot in the lower left corner the scaling was not very large.
Figure 13 shows a clip from an area that can be found in Figure 12.
Poly was 100% programmed in using the GNU C-Compiler v2.6. To extract the parameters from the leader file I implemented some functions from CalVal's rcal_ceos.c.
The program consists mainly of functions for reading and creating files, calculating, and switch retrieving.
One of the major obstacles was implementing the complex switch. In a complex image file, each pixel is represented by 4 bytes : two bytes for a signed int real value followed by two bytes for a signed int imaginary value. These were in hi-lo byte order. The problem was that the fgetc()-function would yield an unsigned byte value, whereas the first byte of the real/imaginary value would be signed and the second unsigned. That is where the union-structure in C came as a blessing, as I could thereby address each single byte to take the value from fgetc(), while later it would be interpreted as one signed int value! The rcs was calculated by replacing the DN with the magnitude of the complex figure. After scaling down the magnitude is saved to the image output file as a single unsigned byte.
Currently the clip is limited to a maximum height of 16384 lines. This is about the only hard-coded restriction on the clip size (to be found in the declaration of the variable et[]). The memory required for the actual clip and mask are reserved by using malloc(), so lack of memory will not become apparent until you define the corner coordinates.
This summer was supposed to be filled with trips to Delta Junction for the calibration of the 20 different beams of RADARSAT. In the last half of June we had an average of 5 trips per week to change the inclination and heading of each reflector to have all of them pointed for that beam. Then by the end of June we got word from the Canadian Space Agency saying that RADARSAT faced difficulties and would thus be turned of until the first week of July. Eventually this was delayed further and further back until late August.
Still we did some trips in the meantime to point the reflectors for ERS-2 passes over the reflector array.
The maintenance trip to Toolik Lake this year was scheduled from the 25th to the 28th of May. In preparation for that trip I obtained maps that cover the region from Toolik Lake to Prudhoe Bay o give us an idea where some of the reflectors would be. After 350 miles and the crossing of Yukon and the Arctic Circle we arrived in Toolik Lake camp, which is a scientific research station of the Institute of Arctic Biology of UAF.
We maintained a couple of reflectors and moved some to a new location. One of those included a helicopter ride for me because the reflector was to be located at "West Kuparuk" about 60 miles west of Dalton Highway (in other words completely unaccessible by road). During the transport, the reflector was in a net with additional watertanks to provide enough weight, so that the payload does not swing back and forth whilst flying. Somehow these watertanks wreaked havoc on one of the panels: it was bent inside and fractured on its edge; we did not find out before landing. We tried to fix it by using a rebar as means of bracing, that is we fixed it with wire to the bottom side of the front edge. To provide extra pull on that edge we drove an additional stake in the ground which we linked with a cable to the horizontal rebar mentioned above.
There were two more reflectors removed that we took further north: One near Sagwon and the other one to Prudhoe Bay. The latter were mainly put there for other people from UAF who use them for geometric means (similar to CalVal's geometric calibration).
The calibration of ERS-2 is Parker's main work this summer. As he was for two weeks on his honeymoon I assisted in analyzing ERS-1 and ERS-2 images for this purpose. For the ERS-1 part the data that I analyzed for the Reflector Validation Task could be used, whereas for ERS-2 I analyzed a lot of images using disa again to obtain data for geometric and radiometric calibration.
While analyzing data I stumbled upon a difference in the geometric data: The y-coordinates that the SAR processor gave for the reflectors were more than 440 pixels off their theoretical location, which translates to a locating error of 5-6 kilometers! This error occurred in four images, so we reported the anomaly. This resulted in not delivering images that were taken after day 139 in 1996 to the users. As far as I know the responsible people did not find out the reason for this anomaly, and they probably did not put too much effort in it since ERS-1 was to be shut down on day 154 anyway.
Over the past years CalVal has built up quite a library of slides, negatives/prints, plus a compilation of these on Photo CD.
To organize all of these we devised a database using FileMaker Pro v2.1 on the Macintosh.
First all of the slides/negatives were sorted according to the year they were taken. The year is the first part of the Film ID #, followed by a sequential number for that given year; in other words the second film of 1996 (whether slide or negative) is labelled 1996-2. Each frame is given its own sequential number for a given Film ID, so you end up with something like 1996-2-18. These figures were printed on the slide or back of the prints. The slides and negatives were then sorted in a binder.
After the ground work the data had to be entered into FileMaker Pro. Therefore certain fields are defined:
-Frame Number
-Date when the picture was taken
-Description of that particular frame
-Group Description, for example "aerial photoset"
-Film ID -Film Type
- i.e. slide/negative
-Photo CD ID. Only if the same picture is found on a Photo CD
CalVal has all the images backed up on 8mm tape. To reduce redundancy on the tapes I checked about 70 tapes for images that had already been backed up elsewhere. As a result almost 30 tapes proved to be redundant backups, which we thus could use for storing new images.
Over the whole summer ASF deployed 5 new reflectors for geometric calibration at Delta Junction, of which I deployed three together with Ty and Parker. These three belonged to the category `Miscellaneous' in being pentagonal in shape yet mounted with a Gen I base. Each reflector took about 2 hours to set up.
While analyzing ERS-2 data it became obvious that the K value for the reflectors would be constantly above comparable ERS-1 measurements. Since ERS-2 follows exactly the same path as ERS-1 with a difference of one day, I looked for twin pairs over Delta Junction. Unfortunately as I write this only two pairs could be found, the first pair showing DJR 1, 2, 5, 6, 7, 9 and the second pair showing only DJR 8. So it is not representative, but still it gives quite an idea of an almost constant offset between ERS-1 and ERS-2 (See Figure 15).
ASF, in its role as the Americas ALOS Data Node (AADN), is hosting the 3rd ALOS Joint PI Symposium in Kona, Hawaii on 9-13 November. This meeting of Earth scientists using...
Bookmark in Delicious
and 
