Combine different geometries for decomposition of LOS motion?
August 15, 2017 at 1:14 pm #2390
I have two datasets of Sentinel images (one ascending and one descending). I would like to know if it is possible through the software to decompose the LOS motion in vertical and horizontal components.
Or simply, I have to process the two datasets separetely and then apply some formula using the LOS values? In that case, which formula(s)?
August 15, 2017 at 2:43 pm #2391
yes, you can, through this module: http://sarproz.com/manual/multi_sens.html
Unfortunately the module is a beta version and not all functions are available.
Anyway, you can align your datasets, select ascending/descending corresponding points, then plot the projections.
You have firstly to process asce/desce datasets separately, then export time series objects and then import them in such module.
Vertical/horizontal movements can be plotted either on AD pairs or on resampled grids. In the future we’ll give more options for resampling. At the moment the time series export in not implemented yet but we’ll do it soon.
Let me know if you get it working or if you need a remote session to show you how to use it
August 16, 2017 at 4:12 am #2396
Thanks for the super quick answer.
Yes I get a little bit lost in the export part. So I process the sets separately (and here all clear). About the export..I am working with a small area, so with Small Area Processing. So what I have to do is “just” click export the time series in the geocoding window, and then the software will recognize automatically the objects to import in that module?
If available, yes it would be very helpful to see a tutorial or something similar.
August 17, 2017 at 5:37 am #2398
Also, what happens if the PS points are not coincident in the two datasets?
August 17, 2017 at 6:58 am #2399
A Time series object is a Sarproz internal data structure,
formed by a file with .tsobj extension and a folder.
Such structure can be optionally generated either from the APS module, or from the MISP one (APS and MISP are module for estimating time series models).
Alternatively, there is a module where you can generate a TS object without estimating models, simply by reading information already available (e.g. because previously estimated): http://sarproz.com/manual/ts_module.html
At the moment you cannot create such TS objects from the small area module.
Many years ago it was possible to jointly combine multiple datasets using the small area module, but that functionality is currently disabled because obsolete.
About your second question, the Multi-Sensor module allows finding corresponding points from the 2 datasets, by better aligning the 2 datasets and by placing thresholds on the distance between corresponding points. Once they are aligned, you can also choose to resample your data on a regular grid, in this way you get corresponding areas instead of corresponding points.
We have several improvements to do on this subject, we’ll work on them in the next months.
August 24, 2017 at 11:53 am #2406
Thanks very much for the explanation.
Although, I have not really understood what I should do in practice (sorry, for me is a quite new world…and software).
I process the two datasets using Site Processing and then..? I use the multi sensor analysis module? Or I need to do something else in the middle?
Thanks for the infinite patience
August 24, 2017 at 12:18 pm #2407
the minimum required is estimating the time series using the APS module, then save the corresponding TS object, for both datasets.
when you have the 2 TS objects, you can import them in the multi-sensor moduel.
depending on what you are doing, you may want to process also the MISP after the APS.
here you see some exercises for the APS and MISP modules https://www.sarproz.com/by-timo-balz/
however, probably you need a training on the subject.
August 24, 2017 at 1:14 pm #2413
Thank you for the further clarification.
I will try then
October 9, 2017 at 1:01 pm #2519
Hello sorry, again me making questions:
which are the formula used by this software to make the projections in the horizontal and vertical direction?
If it is possible to know of course
October 9, 2017 at 1:23 pm #2520
We encourage you to ask as many questions as possible.
I will wait for Dr. Perissin on this question, but I think it is only a very simple geometry that you could draw yourself.
I append here a 2D (horizontal and vertical) decomposition here:
Now, with only one pass, it is impossible to know how to decompose LOS to horizontal and vertical. However, with two passes (one ascending and one descending), you can deterministically decompose into horizontal and vertical direction.
October 25, 2017 at 10:18 am #2547
so I used the software (multi sensor module) to do the combination of ascending and descending to get horizontal and vertical velocities. And, since I was interested to see the horizontal deformation along time, I took the values in the time series and I used the formula found in the paper “Space geodetic monitoring of engineered structures: The ongoing destabilization of the Mosul dam,Iraq” written by Milillo and others in 2016 (see Figure for formula). So basically I took two spatially close points (one asc and on desc), I took their deformation values in the time series and I applied the formulas. Only problem is that the trend of this time series is in the opposite as the software does (i.e. the software show postive values, so toward East, while my result shows that the final trend is negative, which means West). I tried to apply the formula for the vertical component and that is in agreement with the software.
So did I interpret wrong the forumula? Is because the software apply some kind of interpolation so it can be that software and manual computation do not agree?
I hope it was kind of clear
Attachments:You must be logged in to view attached files.
October 25, 2017 at 11:05 am #2549
to find out where is the sign problem, I’d suggest you to find a case where the Asce and Desce velocities have opposite sign.
Where A is + and D is -, the point must move to the W. If the plotted result is positive, it means that the E-W component was assumed positive to the W. If you can confirm it, we will change sign and change the name to East component…
October 25, 2017 at 11:19 am #2550
Yes in my case, if A is + and D is – then the resultant is negative (toward West). Yes I have no doubts about the sign per se, but about the fact that the software and my computation do not agree (on a determined location, software says + and my computation says -). Is this what I cannot figure out. I check the formulas many times and always got the same. Is because the formula method is extremily simplicistic while the software employs alogortihms etc ? I am simply trying to find an explanation for this difference. Because for example, they agree in vertical direction, both -.
October 25, 2017 at 12:13 pm #2551
Can you please post an example?
October 25, 2017 at 12:38 pm #2552
Let’s see. In the figure attached , there is the East -West velocities from the module. Blue is positive so East. So far so good.
But I took one descending point which has final trend of -20mm/year (the “typical” value you get in the small area processing), and one ascending point value of -7mm/year. So both negative.
The points are located in a blue area in the image produced in the multi sensor module. I applied the formula to those two values to get the horizontal component. So, following that formula in the paper, dhor= 1/2 *((dD/sin(incD)-(dA/sin(incA)).
In my case incD=33 and incA=38. So dhor= 1/2*(-20/sin33)-(-7/sin38))= -13 ca. So negative and hence West. But in the image the area of their interest is blue, so East.
October 25, 2017 at 12:44 pm #2553
October 25, 2017 at 12:56 pm #2555
A first question would be: did you select points representative of the area where they are located or did you choose outliers?
Secondly, the software decomposes asce/desce velocities only on AD pairs. AD pairs are points located closely in A/D datasets. The sw selects them based on your inputs (AD processing) and based on the corrected shifts between A/D clouds (offset estimation): http://sarproz.com/manual/multi_sens.html
From the picture you attached one can clearly see that the resolution of that map is very low, results look having been spatially averaged with about 1km radius. So, I am not surprised that the numbers you picked up from 2 points are not matching those colors. Moreover, even if it is difficult to say anything without knowing the color scale, the spatial variations don’t look very clean. I suggest you ask someone for some closer help and you go through all the processing steps and intermediate results…
October 25, 2017 at 1:18 pm #2556
I took two points spatially close (ca 200m distance). Because the area per se have not so many points.
Unfortunately, I do not know anyone close that can help me. I am doing this project basically from zero.
But ok thanks I will try to improve the resolution then
October 26, 2017 at 5:49 am #2558
Hi Cecilia, in order to better understand the issue, could you please attach a picture of the Asc and Desc results (PS) you imported in the decomposition module?
You must be logged in to reply to this topic.