APS estimation
This topic contains 13 replies, has 8 voices, and was last updated by hamidbanan 2 months, 1 week ago.

AuthorPosts

February 4, 2015 at 10:56 pm #690
Dear dan,
Thanks for the excellent software.
The new version for SARPROZ is much more convenient than the previous one.
however, i have some quesitons about the processing.
1. in APS options group under APS processing, there are several options for it, including I.P., I.R., Tri.I.R. and stratif.
For APS processing, is there any basic idea to select one of them ?
Can you provide more details on them and also the different between them?
or some references for me to understand these different methods?
2. also, according to Timo’s instruction, In the APS Estimate group, by pressing the Test button, we can get an estimation of the APS estimation quality.
One of them is the atmosphere coherence. How does it calculate and represent?
Also, what’s the meaning of the picuture in the attachment?Thanks a lot!
Regards,
QingliAttachments:
You must be logged in to view attached files. 
February 5, 2015 at 6:58 am #692
Hi Qingli.
APS processing module options:
– I.P. stays for Inverted Parameters
– I.R. Inverted Residuals
– Tri. I.R. Triangular Inverted Residuals
Generally, you should use the I.R. option. If the APS looks ok and the test shows a homogeneous high coherence in your analyzed area, then you can proceed further and be assured that everything is working well.
However, in some more complex cases, the final coherence may be not satisfactory. Such cases should be analyzed one by one to understand the reason of the failure.
A possible solution is using the I.P. APS:
let’s assume that you trust the parameters that have been plotted at the previous step, then using the I.P. APS forces that solution.
The I.P. APS always brings a high final coherence. However, it’s a forced solution. So, be aware of that and use it with caution.
Also, if your parameters have a ramp and you remove it with the “flatten” option in the frame “estimated parameters”, then you must use the I.P. APS (it forces the APS to match the flattened parameters).The “test” operation tests the estimated APS.
Which means, it calculates the temporal coherence after removing the estimated parameters and the estimated APS. If the test coherence is high, it means that parameters and APS are matching the phase series.
The picture you attached is a kind of rough estimate of the residuals image by image. It may give you hints about a particularly noisy image or a particular temporal trend of the noise (e.g. if you observe it in a city with many high buildings, in which you did not remove thermal effects, it will show a seasonal trend). 
March 9, 2015 at 10:38 am #760
Hello,
I would like to ask what are the other parameters in this line:
– stratif – should it be used for ionospheric compensation (in case of Lband data)?
– R and DSF – I understand that it is something like the “smoothness” of the estimated atmosphere; however, why are two parameters and in what units are they?
Thank you,
Ivana 
March 9, 2015 at 10:48 am #761
The stratification option is used to estimate a correlation between APS and elevation (DEM).
If you work in a mountainous area and if you have coherent point at different altitudes, you can use it and it should help compensating the vertical stratification of the APS.DSF is the downsampling factor of the final APS. R is the length of the correlation filter (pixels).
d

May 22, 2015 at 11:02 am #909
Hello,
Could you offer me some guidance as for the Sparse Points Selection criteria in the APS tool?
I’m not very sure what each option means,
Thank you,Santiago


May 23, 2015 at 12:33 am #910
Hi Santiago,
1. you choose a parameter on which you want to place a threshold (more advanced option: building your own set of points through “save selection” in “scatter plots”, and select it as a real file here)
2. you choose the threshold
3. you choose if you want to make the selection of points sparser or not. If not, place DS=0, DL=0. If you want to make it sparser, place e.g. DS=30 and DL=30. in this way it will keep 1 single point withing a DSradius circle. If both numbers are nonzeros, it will divide the area in rectangles DSxDL, keeping 1 point in each of them (faster but not as good as the circle algorithm)
4. if you want you can manually save/load such selections of points.
I believe you find some examples of APS processing in the notes from Timo. 
May 27, 2015 at 9:01 am #924
Hi, Pro. Daniele,
(1) I’d like to know something about APS estimation.I used 32 scenes TerraSARX and TandemX images in all.The zip file contains the parameters of APS estimation and the results.I found the connection coherence is good,but the result of APS including “PS coherence after graph inversion and APS removal” are bad and the image coherence is low.I’m confused about the differences.This is the first problem.
(2) And,could you please explain the nonlinear weigh function
(3) Whether the interferogram processing has influences on the APS estimation?
Attachments:
You must be logged in to view attached files. 
June 15, 2015 at 7:08 am #994
Hello Daniele,
After having a look at timo’s notes, and the manual, I was able to get very good results for displacement in my study area. I’m really very happy with the sw.
However I still need to clear out some points of the APS processing, since I need to write a report explaining the processing I’ve carried out:
– In the processing parameters section, what do the ‘UW’ ans ‘Ext DEM’ options exactly mean?
– How is the Connections Coherence calculated?
– And, as aked by cnu_lee, could you explain the function of Nonlinear Weighting? How does it affect the computation of the Coherence?
– How is the coherence calculated after APS removal ?Thank you,
Santiago 
June 15, 2015 at 7:56 am #995
a. UW means unwrapped.
If you check it, the sw will try to read UW interferograms. Then, the multitemporal processing (estimation of parameters) will work using Least Squares (instead of the maximization of the periodogram).b. Ext DEM means removing the external DEM from time series.
c. the connections (temporal) coherence is calculated as gamma=sum(exp(j*phi_res))/N as described in the whole literature
d. after you process your PSC graph, you have to invert (integrate) it to get parameters/phase residuals on individual points (instead of connections). The temporal coherence is used as a weight. Since the temporal coherence is usually biased by the number of images (and by the baselines configuration), the nonlinear weighting helps deciding which connection is reliable and which not. It does not affect the temporal coherence. It affects the graph inversion/integration.
e. After APS removal the coherence is calculated the same as before. However, before APS removal phi_res=phiphi_model. After APS removal, phi_res=phiphi_modelphi_aps

August 25, 2015 at 9:55 pm #1158
Dear Prof.Daniele
I would like to ask,
1. based on what we can conclude that SAR has good results and can use the APS options ( IP , IR , Tri.IR ) in the area of the case study ? Can you attach some paper / reference, in order to we can understand more each parameter. It is not enough to judge like your say that “the APS looks ok and the test shows a homogeneous high coherence in your analyzed area, then you can proceed further and be assured that everything is working well used IR”2. My case study area of lowland , how do you think if I ignore Ext.DEM and replace Matr.Coher.Win 5×5 (I want to do filtering)
3. In the weights option, could you tell me what is different about “none, coher , amps” . Whether this affects the results of processing APS?
Best Regards,
Ana

August 26, 2015 at 1:42 pm #1169
Hi Anarizka,
1) Here my personal suggestion about your first question. The best way to understand the goodness of your results (both preliminary and final) is based on the knowledge about the process you are investigating. You should address your specific case study (landslide? subsidence? localised structural issues?) being driven by a general model based on geological/geophysical or structural information. In such way you will be able to judge your results by means of a robust model.
About your specific question, please consider that the multiimage InSAR is a basically a challenge against atmosphere and other noisy signal (e.g. orbital inaccuracies). For such reason, you will reach your target when you are able to separate the “real” signal (i.e. height, displacement information, seasonal deformation, etc.) from the noise given by atmosphere phase effect and others. To do this, all steps are fundamental. In the APS panel you have several steps to be considered: a good selection of PS candidates (set of points); a robust network (the spatial graph), and the estimation of preliminary parameters along connections of points (usually height and displacement as linear trend – but also, sometimes, non linear displacement). Once you get your parameters, you need to spatially invert the graph to get parameters on individual points. At this stage, if the temporal coherence is high on most of connection, the estimation of APS by using IR will be good (high coherence) for the whole set of point by using IR, which is a standard for all PSlike approach (you will find lot of reference about this). If it is not, you should investigate if other source of noise exists with a spatial distribution (few points? points too sparse? orbital inaccuracies?). As Prof. Perissin said, in fact: “…in some more complex cases, the final coherence may be not satisfactory. Such cases should be analyzed one by one to understand the reason of the failure.”
In such cases, if you judge your preliminary results good (remember the “conceptual model” I mentioned in the first part of my answer) and if you trust them, you can try to use IP instead of IR, thus forcing the solution because “you” decide that it is correct. Unfortunately, there is not “one perfect way”. InSAR is based on data (you have to analyze them carefully) there is not a standard way to perform all analyses: every case is a singularity and only by experiencing several case studies you will be able to find the correct way.2) It depends on what you want to do. It is not a matter of heights (lowlands and so on). If you ignore Ext.DEM and you use the spatial filter, you are simply not performing a standard “PS” analysis, and you are switching to distributed scatterers analysis (QuasiPS, etc.) (please refer to http://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=5978217). Such analysis is based on a totally different approach and also other aspects should be taken into account (e.g. the image graph which should maximize the spatial coherence, as the MST, for example).
3) Coher exploits the spatial coherence written in previous steps to weight to parameter estimation, while amps use the reflectivity value (it stays for amplitudes). If you choose none, any weights will be used. Indirectly the APS can be influenced by this step since it influences the parameter estimation (which is the starting point to estimate the APSs).
best regards
Alfredo


January 5, 2019 at 6:01 am #3685
Dear Prof.Daniele
I have a question about APS estimation. Actually I have no extra data or no idea about the amount of subsidence or deformation size. So in processing parameters, What strategy should be followed about numeric value of the range in linear trend? The default value (1.1) seems to be wrong in my study area.
regards
Hamid 
January 5, 2019 at 10:45 am #3686
the default value of linear deformation trend is 100 / 100 mm/year.
when you estimate the def trend in the APS processing, you estimate it between close points (the network connections). So, 100 / 100 is already a pretty big range.
in any case, if you have no previous knowledge, the histograms and all plots you get after the processing will help you understanding if the range is too big or too small
best 
January 5, 2019 at 11:42 am #3687
Dear Dr periz
thank you for quick reply. In InSAR analysis of 2 coherent images (with 1 year interval), I realized that the maximum range of subsidence is 17 cm in bowl center (Just as a theory), so is it logical to put (0 , 170) or (170 , 170) for linear trend? Also I found there is gap between the results of sentinelInSAR analysis in SARPROZ software and snap (Values of SARPROZ were more estimated than snap). How should this gap be interpreted?
regards
Hamid 
AuthorPosts
You must be logged in to reply to this topic.