powerlaw and weather model correction for Sentinel 1 (small area observation)

[Laila]

Dear David and all,

I am trying to reduce tropospheric effect using TRAIN for 33 ascending scenes Sentinel-1 SAR to investigate slow movement landslide in Indonesia. The area is relative small (the investigated movement is right located on the middle scene).  The areas for a whole SAR scene and prone landslide area are, 6 x 8 km and 1.5 km x 0.7 km, respectively. It is located on mountanous area which i assume topographic term will have high influence for phase artefacts.

Therefore, after considering no unwrapped error to the final interferograms (rsb less than pi), i try to substract tropospheric effect with
- ERA-1 model
- MERRA2 model
- linear model
- power law model

Using your script (aph_powerlaw step 5) to calculate RMSE between different tropospheric correction methods, i found high error gap between power law and weather model. Based on your paper, the phase-based method will have better performance to reduce tropospheric delays if the regions are mainly corellated with topogrpahy. Otherwise it's better to use weather model if the regions may have turbulance in troposphere and high dynamic local weather. Therefore, due to resolution scale of weather model data, i think for my case study it's better to use power-law method (but have to make sure the spatial bands parameter is correctly set up to the size area of SAR scene)

My questions are:
1. I am confuse to set the right parameter of spatial bands, i already read some suggestions from you and people in this forum (to change patches parameter to '0'), especially for landslide study (~ short wavelenghts). I was playing for the spatial bands filter parameter and try to find the range value of tropospheric delay lower than phase range of unwrapped intfs (usb-d). So, is it right to set the spatial bands to [1000 1100] with the unwrap grid size for STAMP processing is 100 m?
I attempted to use powerlaw ridge constraint too, but regarding to small area, the results seems not promising.
2. To calculate the mean velocity and time series, i have tried many options, 'v-da' & 'v-dao', for powerlaw correction both SM and SB, but the time series still not really smooth and some atmospheric artefacts still left.
then i also tried to use ERA-1 model combined with atmosphere and orbit error (AOE) phase due to slave 's', and the TS looks moving more smoothly. Could I consider that 's' model correction from STAMPS is the source of mainly ionospheric delays (although the deformation signal could be leaking too on 's' step)?
I ask it since i haven't found explisitly to correct the ionospheric term in TRAIN 2 beta software.

Hopefully, there is a hint for my case study. Thank you for your help.

Best regards,
Laila

Note:
if there is an error to accessing MERRA2 files; aps_weather_model('merra2',1,1), try to download manually due to access to ftp site since June 2016 requires registered user.

[David Bekaert]

Hi Laila,

Your study area is small and that mainly the reason why you want to set the patch_number to 0, i.e. one patch. As it is small it does not really allow to estimate a spatial variable powerlaw and this is likely the reason why the ridge approach does not work well.

The filter bands (1000-1100) is typically determined by contamination of the deformation, in this case you landslides (up to 1.5 km x 0.7 km as you indicated). Based on this you probably want to have a band either smaller or larger than the spatial scale of your landslide. The stamps merge-resample size determines the shortest wavelength you can apply. So I would recommend not to go smaller than 2/3* merge_resample size. The bandfitering happens in the frequency domain, but requires the full image to be interpolated. For the shorter wavelengths you might have some issues there as the data is not uniformly distributed over your study area. so you could try to also use a bigger band filter, i.e. 1200-4000km or so. The maximum bandwidth is given by the extend of your study area, so would not go above 8km.

The discrepancy between the weather model and the powerlaw can be due to few reasons. Incorrect estimate of the powerlaw as you indicate is one of them as the powerlaw would only solve for a topography-correlated component. Also care should be given to the estimated weather model correction. The weather model typically comes at coarse resolution 75 km (ERA-I) to 50 km (MERRA-2). Therefore one would not really expect to see any turbulence in the model either at the scale of your study area, so would only see a smooth version with topography superimposed.

You will want to see that the uncertainties decrease (vs option).

Some of the interferograms have a strong signal.

One option would be to leave them out of the inversion and see if this makes a difference too.

The ionosphere is typically a longer-wavelength signal.
It has appeared in S1 for longer swath processing, or when having issues in aligning a stack in the burst overlap regions.

I would not expect it to have an impact over your study area.

I am not sure if the filtering 's' step in stamps is actually removing the APS estimates first and then doing the filtering.

Best to verify that. If this is not the case then atmosphere might be removed twice.

Note that 's' will smooth and clean the time-series as it it filters in time.

Thanks for the tip on MERRA2: A fix is included in the git version of TRAIN.

Cheers,

D.

--
You received this message because you are subscribed to the Google Groups "TRAIN - Toolbox for Reducing Atmospheric InSAR Noise" group.
To unsubscribe from this group and stop receiving emails from it, send an email to TRAIN_support+unsubscribe@googlegroups.com.
To post to this group, send email to TRAIN_...@googlegroups.com.
Visit this group at https://groups.google.com/group/TRAIN_support.
To view this discussion on the web visit https://groups.google.com/d/msgid/TRAIN_support/345f082d-eaa8-4b1a-a35f-05378d93b491%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Laila]

Hi David,

Thanks a lot for the explanation. It helps me to starting understand atmospheric correction and the filterband suitable to my case study.

The merge_resample_size is set to 50 m. I've tried to use bigger band filter as you wrote [1200 4000] and so on with the one patch (powerlaw_n_patches or patch_number = 0).
It seems the results still overfilter (the scale is much higher than scale of 'usb') so i will figure out finding the other bigger or smaller bandfilter which fit to my real landslide size area perhaps.

Another question, is it possible to combine both small and big band filter (more than 2 rows matrix spatial bands, e.g. [950 1000;1000 1100; 1200 4000;2000 5000]) using 0 patch_number and pick the best filter band into powerlaw_kept?
because when i attempted to run, i should have to set the patch_number at least to 1 and the result shows overfilter as well.

If it doesn't still work well. Another option as you suggested, i will consider try to remove some interferograms having strong signal out of the inversion.

----WRF model------

I also tried to use WRF model with the suggesting the resolution are higher (nested to ~ 7 km).
I've found some errors during processing.
What i realized, WPS and WRF Model Version 3.8.1 that i used, have some a bit different arrangement output for parameters: temperature, relative humidity, geopotential height, and pressure.

So i edited the aps_wrf_SAR.m script to make it run for my dataset.

%load variables
Temp = netcdf.getVar(ncid,10);       % temp in K
Hum = netcdf.getVar(ncid,11);        % relative humidity in percent
H = netcdf.getVar(ncid,12);          % Geopotential Height in m
Plevs = netcdf.getVar(ncid,7);      % 37 pressure levels in Pa

i don't know that is only happened to me or others too.

Best regards,
Laila

Cheers,

D.

To unsubscribe from this group and stop receiving emails from it, send an email to TRAIN_suppor...@googlegroups.com.

[reyhanaz...@mail.ugm.ac.id]

Dear laila,

How do you run Train software and both of the algorithm (weather, power-law) ? Can you give me a hint how to run it ?

Thankyou

Best Regrads,

Reyhan

[Nahidul Samrat]

Hi Reyhan,

Please go through the TRAIN manual for hints and details. Visit David website for download the manual. 

Thanks 

S