Interpreting ISCE2 + MintPy results / the way to calculate accumulate of displacement

[gjustin]

Hi, I am a beginner in this field and I have some basic questions as I am new to using ISCE and MintPy.

I performed Time-Series analysis using ISCE2 and MintPy. In ISCE2, I created a stack using `stackSentinel.py` and performed Time-Series analysis using `smallbaselineApp.py` in MintPy with the created stack.
1. Data: 28 images of Sentinel-1 SLC (20200120 ~ 20201221)
2. Study area (bbox): 11.0295 11.1651 106.5815 106.7192
3. Purpose: land subsidence (displacement)

The commands, functions, and variables I used are as follows:
(ISCE2)
- `stackSentinel.py -s SLC/ -d DEM/demLat_N10_N12_Lon_E105_E107.dem.wgs84 -b '11.0295 11.1651 106.5815 106.7192' -a AuxDir/ -o Orbits -c 2`
- `run_01_unpack_topo_reference` to `run_16_unwrap`

(MintPy)
- `smallbaselineApp.py study_area.txt`

I haven't encountered any specific errors during these steps, and the results seem to be appearing correctly. Here are the questions I have about the results:

1. When looking at the Time-Series result images (png), they appear larger than the bbox I set. Why is that?
(I found that this is because the processing is done on a burst level. Is that correct?)
1-1. If so, how can I view only the bbox I'm interested in?
(Can I simply use `subset.py` for this?)

2. In the Time-Series result images (png), there are black rectangles that seem to have no information in the middle. I found that these are called reference pixels. What role do they play?
(Although referred to as pixels, they seem to encompass a certain range rather than just one pixel. Why do they appear?)

3. All the results are given as velocities, but I'm interested in the accumulated displacement over a year. What information should I look at?
(In other words, if I asked "How much land subsidence occurred over the year?", what information should I reference?)
(If I need to derive additional results from the extracted data, could you provide the method or link?)

4. I'd like to interpret the results from MintPy's outputs individually. Is there any reference material available?

5. Due to the nature of SBAS, the resolution becomes excessively reduced. Is there any option available to mitigate resolution reduction even slightly?

I have attached  some images from MintPy for your reference.
Even if you don't provide direct answers, if you could share reference materials or links I can consult, I will try to resolve the issues on my own.

Thank you.
(As a reference, I executed all the commands with their default values.)

[Eric Fielding]

Hi,

The key reference for MintPy is this paper:

Yunjun, Z., H. Fattahi, and F. Amelung (2019). Small baseline InSAR time series analysis: Unwrapping error correction and noise reduction, Computers & Geosciences 133, no. 104331, doi:10.1016/j.cageo.2019.104331.

There are some tutorials on how to use MintPy in the tutorial repository:
https://github.com/insarlab/MintPy-tutorial

To answer your questions:
1) The minimum area that the ISCE2 stackSentinel workflow can process is one TOPS burst of Sentinel-1 data, which is about 20 km along-track and 80 km across-track. You can specify a subset of the data in your MintPy input file for the time-series processing.

2) The black square is showing the location of the reference point of the time-series analysis. This is only for the display, the reference is a single pixel. The reference point is set to zero for all the dates and all the interferograms, so all of the displacements are relative to that reference point. If you don't set a reference point in your input parameter file, then MintPy chooses a point where the spatial coherence is high. See Yunjun et al. (2019) paper for more details.

3) There are many different outputs from the time-series analysis when you run the whole smallbaselineApp. Look in a folder called "pic" to see images made from many of the outputs. The "timeseries.h5" file is the cumulative displacement from the reference date. The reference date is set to the first date of your stack, unless you specify a different reference date.

4) See the tutorials.

5) The multilooking done by ISCE2 in the stackSentinel processing provides results with about 80 m resolution. For land subsidence measurements, this resolution is usually fine enough. You can change the amount of multilooking in the ISCE2 stack processing, but it will take much longer to run and produce much larger files.

All the best,

                 ++Eric

[gjustin]

Thank you for your answer.

Additionally, I obtained results similar to the image through the following command. 

- view.py geo/geo_timeseries_ERA5_ramp_demErr.h5 --save -o geo_timeseries_subset_20200104.png -n 30 --noreference --mask geo/geo_maskTempCoh.h5 --ref-date 20200104 --sub-lat 11.0295 11.1651 --sub-lon 106.5815 106.7192

Is it correct that red (positive values) indicates erosion and blue (negative values) indicates deposition? In other words, is interpreting the results based on Line of Sight (LOS) correct?

2023년 8월 12일 토요일 오후 1시 7분 18초 UTC+9에 Eric Fielding님이 작성:

[Eric Fielding]

Your interpretation is not correct that positive values mean erosion. The positive values mean surface displacement towards the satellite, which means up or horizontally towards the satellite. If you don't expect much horizontal motion, then you can make the assumption that it is vertical displacement and red means upward.

It is important to understand that you cannot measure erosion or deposition with repeat-pass InSAR directly, because erosion and removal of material or deposition of material will change the surface and cause a loss of coherence for InSAR. You can only measure motion of a relatively unchanged surface. If there are geological or hydrological processes such as groundwater extraction occurring, then you can see surface subsidence or uplift.

The other caution is that you should be careful about interpreting apparent signals that are in or adjacent to areas of low temporal coherence. You applied the temporal coherence mask in your plot, so low coherence is white. The darker blue pixels in your plot are largely surrounded by low coherence areas, so they are likely to have errors.