Prepare S1 STAC catalog for USFIMR flood events

[echeipesh]

This is data preparation step for training a flood inundation model using USFIMR as label data

• [x] Use s3://usfimr-data/catalog.json as a way to get flood event locations and times
  • Using item simplified geometry or bbox should be enough to make a request to Sentinel Hub
• [ ] Using Sentinel Hub batch API prepare a STAC catalog of Sentinel-1 GRD, orthorectified sigma0 scenes.
• [ ] Save stac catalog and raster scenes on our S3 bucket for later use
• [ ] verify that data product is compatible with GEE Sentinel-1 sigma0 product used by Sen1Floods11
  • https://developers.google.com/earth-engine/datasets/catalog/COPERNICUS_S1_GRD
  • https://developers.google.com/earth-engine/sentinel1

Notes:

• This MAY be good use case for STAC GeoPandas frame, consider using it before beginning
• Downloading S1 imagery may be something that needs to be distributed. If so, consider trying out nextflow to do so.

[echeipesh]

Given that the age of some USFIMR events and general Sentinel-1 coverage we're able to cover 5 of the events with +/- 5 hour buffer and 6 events with +/- 48 hour buffer.

That amounts to:

4 results for USFIMR ID=1
sq meters 6096032.0915204
2 results for USFIMR ID=2
sq meters 2487657.46065816
4 results for USFIMR ID=3
sq meters 6177288.656697515
2 results for USFIMR ID=15
sq meters 860384.8119924887
3 results for USFIMR ID=16
sq meters 5101076.359155779

Each event is = 2000x2000 pixels of Sentinel-1 data. For reference that is equivalent to 900 Sen1Floods11 chips, twice the size of the hand-labeled set and 20% of the "NoQC" weak labeled set.

That's not as large as one would have hoped for, but I think enough to get a comparison.