InSAR Validation Using Free GPS Data in Iceland
When it comes to remote sensing Synthetic Aperture Radar (SAR) satellites have a unique ability that traditional optical satellites can’t do: monitor ground deformation. This is because of their active radar sensor which allows the satellite to monitor how long it takes for the signal to return from the ground. They can calculate the distance of the signal thus they can calculate the shape of the terrain. Calculating the signal differences between two images is called radar interferometry, otherwise known as InSAR, and is used to monitor ground deformation using radar remote sensing.
You can retroactively monitor ground deformation data depending on when the satellite started operations. This is really useful for infrastructure and hazard monitoring in areas where full historical GPS/GNSS data may not exist.
In this post I’ll be doing a quick 1-year analysis InSAR in the area near Stafnesviti, Iceland from August 2019 — August 2020 using Sentinel-1 data. I’ll just be using default parameters for the algorithms and I’ll share some quick screenshots. I have ground validation data thanks to the Nevada Geodetic Laboratory (NGL). They provide thousands of free GPS/GNSS monitoring sites around the world that can act as ground validation data.
GPS Ground Truth
First let’s go over the the ground truth data of the NYLA station. The ground truth is located at the coordinates 63.974, -22.738 degrees (Latitude, Longitude). Based on Google Maps imagery the station is located in an open field with no nearby structures and the terrain appears to be flat.
GPS data goes back to 2017 and you can see the GPS data itself is still quite noisy but you can see an overall trend. The red trend line in the graph is based on a MIDAS trend estimator algorithm developed by NGL themselves.
The tsenv3 files were parsed to extract dates that were from August 2019 — August 2020. I will just be using a simple linear regression model to estimate velocity for the time period. The graph also shows a 95% confidence interval which is the translucent red area. The first row in the dataset will be used as a reference date when calculating vertical change.
Based on the data, there is a slight downward trend during August 2019–August 2020 at -0.034 mm per year with a standard error of ±0.001 mm. The results from InSAR would hopefully be around this number.
Now let’s started with some InSAR.
Distributed Scatterer Interferometry (MintPy)
Miami INsar Time-series software in PYthon (MintPy) is a relatively new timeseries algorithm based on the small baseline method for InSAR. It is based on the work by Yunjun, Z., Fattahi, H., and Amelung, F. (2019). In this quick analysis we are just using the default parameters. I prefer working with this algorithm because it is actively maintained and uses pure Python code that works in all operating systems.
The large positive ground deformation you see in the map is actually Fagradalsfjall volcano that was about to erupt. It eventually erupted in March 2021 which is beyond this analysis.
If we look at the deformation data at the exact coordinates of the NYLA GPS station this is what we get:
A stable negative displacement is observed with one data that seems to be an outlier. The MintPy algorithm yields a linear velocity of -0.79 mm per year with a standard error of ±3.99 mm.
Persistant Scatterer Interferometry (StaMPS)
The Stanford Method for Persistant Scatterer (StamPS) is an old but popular timeseries algorithm in the InSAR community. It was written by Hooper et al. (2006). The drawback with this algorithm is that the barrier of entry is high since it requires the user to be familiar with Linux, terminal commands, a paid MATLAB license, and some programming skills because the algorithm is no longer maintained by its original author. Bug fixes that have been identified by the community have to be fixed by the user.
Results of the StaMPS algorithm (no atmospheric correction applied) can be seen below:
The StaMPS algorithm is not as robust to low coherence areas so areas like open fields do not yield as much persistant scatterers. Thus there was no persistant scatterer at the exact point of the GPS station. A nearby point is selected. StaMPS yielded a velocity of -0.002 mm per year with a standard error of ±1.967
Comparing the Results
GPS: -0.034 mm per year with a standard error of ±0.001 mm
MintPy: -0.79 mm per year ±3.99 mm
StaMPS: -0.002 mm per year ±1.967 mm
When looking at the results together we see that MintPy and StaMPS both show downward trends. Despite the large jumps in the graph StaMPS seems to be more accurate in this case. Although the PS point is not exactly on NYLA station it’s only tens of meters away in an open field.
MintPy is still close at -0.79 mm but also shows the largest deviation in its data (though that could be because of the one outlier).
