Stitch time series files

[mfangaritav]

I have been thinking of a solution to reduce the processing time for large time series. I wrote a function here to stitch two time series h5 files into one when there is a spatial and temporal overlap. The function has four main steps:

1. Resample the secondary h5 file so it has the same resolution of the reference.
2. Change the reference pixel in the secondary h5 file so it matches the reference pixel in the reference file.
3. Find the median overlapping date and sets the corresponding time slice as the reference for both datasets.
4. Create a new h5 file that has the time slices from both the reference and secondary h5 files.

I would like to hear your opinions about this approach, whether it sounds reasonable, or how to improve it. I tried this fora time series at Okmok caldera and the results are similar to running a full inversion.

The panel (a) shows the RMSE of cumulative displacement of the stitched solution compared to the full inverted dataset. Panel (b) shows the time series comparison for the maximum displacement pixel

[Alex-Lewandowski]

This will be very helpful for users hitting memory constraints when processing deep time series with large spatial coverage. As you mentioned, @mfangaritav, it could also reduce processing time for systems with access to multiple nodes or compute instances by splitting large time series for parallel processing.

[mfangaritav]

I agree @Alex-Lewandowski and it would work great with your recent changes on asf_search