[8pt] Stage-based CatFIM creating weird boxes in gpkgs

[EmilyDeardorff]

Stage-based CatFIM has recently started creating geopackages with weird box-y shapes alongside the inundation polygons.

Current behavior

The stage-based CatFIM output geopackage has the expected inundation polygons with large rectangular box shapes included.

Expected behavior

We would expect to see the smaller inundation polygon without the large rectangular box.

Screenshots

Weird boxes:

[EmilyDeardorff]

Notes

• Only happening in stage-based CatFIM
• The issue is happening during or before the creation of the merged inundation tif

Debugging Progress (as of 10/7/24)

Folder setup

• Is it caused by not clearing the folder before re-running CatFIM?
  -> not the cause, tested with clean output folders and error still occurred
• Is it caused by metadata pre-pull?
  -> not the cause, tested stage-based CatFIM with and without the metadata pre-pull and the error still occurred

Docker image

• Is it caused by new docker image?
  -> probably not the cause, tested stage-based CatFIM on an older docker image (fim_4:dev_20240823_6321468) and the error still occurred

Fiona and Pyogrio

• Is it caused by the addition of engine='fiona' in file saves?
  -> not the cause, tested stage-based CatFIM with this commented out and the error still occurred
• Is it caused by the addition of gpd.options.io_engine = "pyogrio", added in this commit?
  -> not the cause, commented this line out of all three CatFIM scripts and the error still occurred

NoData Values

• Is it caused by how we are handling nodata values?
  -> Still testing

this nodata stuff seems to be new…? from this commit

here, nodata is set to -1

whereas here(for stage-based), in dev, nodata is set to 10?

test with stage-based nodata val set to 10

it looks like the boxes are still formed with that code variation →

[RobHanna-NOAA]

While unlikely... after Emily tries more of her code level tests, we could try taking a current docker container and manually rolling back some of the packages to see if that helps.

Also.. Emily is likely already on this, but in produce_tif_per_huc_per_mag_for_stage, we are really using a rather non-conventional to do it. Maybe we can look at some similar changes for making in things like agreedem.py. with rasterio.open(dem) as elev, rasterio.open(rivers_raster) as rivers: Might be worth just dumping this idea in and trying it. Obviously we don't need most of that code in agreedem.py but it does do dem masking a bit different and might give some clues (or even a fix).

[RobHanna-NOAA]

Another idea to chase later is maybe how it mosiacs?