This is a guide to understanding the processing steps that are applied to both filtered and unfiltered data layers in GDS results, viewed in Iris.
Deformation time series results derived from InSAR are relative in two ways; the results can only show how the surface changed in time and in space relative to other points in time and in space. This means that the results must be referenced to an epoch (time), typically the first one, and deformation at that epoch is set to be zero.
Additionally, the results must be referenced in space. Often results are referenced to a point, which we will call P. All deformation results are then reported relative to the deformation at P. InSAR derived motion can be thought of as similar to periodic survey results where all elevation data is reported relative to a particular point.
This leads to the question of where to choose P. This problem is described in some detail here. At a high level, P ideally would be at a location that is not deforming (e.g. is completely stationary), is close to the area of interest, and is located at a point where the SAR signal has minimal phase-noise (i.e. the deformation at that point is reliable). It is often hard to satisfy all these constraints, and this can lead to artifacts. If P is selected to be a point that is sinking, the InSAR results will report everywhere far from P as uplifting (and, by construction, report P as stationary.) If P is far from the area of interest, atmospheric effects will degrade the quality of the measurement at the area of interest. If the SAR imagery has lots of phase noise at P, that phase noise will impact all points in the reported deformation results. Further, a successive analysis may not include point P due to loss of coherence, making comparisons difficult.
To address this, GDS uses a spatial filtering algorithm to remove the long-wavelength deformation trend from the data, and to isolate areas of localized deformation. Essentially, the deformation at any point is reported relative to the surrounding deformation. The process of removing a large spatial trend preserves the qualitative deformation results, without requiring a reference point. The image below shows an example of this over an open pit mine, where we can see that the large subsidence of the pit wall is captured after removing the long range spatial trend. Put another way, the subsidence at the pit wall is distinct from the large scale subsidence.
We are able to recreate this signal in our unfiltered layer by moving the reference (shown as an orange marker below) to one of the stable points away from the pit wall. In this layer, spatial filtering has not been applied, the deformation is reported relative to the orange marker, and if we moved the green test point to the orange marker we would report identically zero.
Spatial filtering solves a number of problems for us, but it does not work perfectly in all scenarios; for example, it may remove some of the actual signal we would like to measure, or a very large signal may also still bias surrounding data. For this reason, within our Iris interface, we also provide the unfiltered data, referred to as “LOS (Unfiltered)”, “Up (+) Down (-) Unfiltered”, and “East (+) West (-) (Unfiltered)”, depending on which dataset you’re looking at. This will display the motion at the green point relative to the orange point. We recommend placing the reference point (orange) at a spot near the area of interest but in a spot that is not part of the mining complex. This layer can also be useful for comparing the motion across sharp boundaries to understand the differential motion across points.
