3 new GEOE3 layers added, fires, drought, cyclones

[Gabe-Levin]

What I changed

• Created Processing scripts for 3 new GeoE3 layers
• Added processed parquet files to s3
• Added required metadata for each layer

Processing Notes

Summary

The fire data is coarser than our H3 level hexagons, so our existing approach for zonal stats created lots of gaps. This is because the center of the underlying raster pixels was used to assign values to the hexagons.
(See UnweightedZonalStats.png)

Changes

• To fix this, I used ExactExtract. This lets us assign a value to overlapping polygons based on the percentage of the pixel that the vector covers. Pretty helpful. I think we talked about doing this for other layers in the past. ExactExtract may work there as well.
• Included a quick sanity-check comparison for Kenya (see kenya_db_vs_source_comparison.jpg).
• The cyclones data is empty in Kenya, so I created another test for Japan (see cyclones_japan.png).

Outputs

Processed layers have been added to S3:

• cy_frequency_mean.parquet
• fires_density_sum_weighted.parquet
• drought_spei_1_5_rp100_mean.parquet
• landslide_susceptibility_mean_2023.parquet

Images

UnweightedZonalStats.png

kenya_db_vs_source_comparison.jpg

cyclones_japan.png

[ghost]

PR Deployment Details:
🚀 PR deployed to https://sukc96x8yd.execute-api.us-east-1.amazonaws.com/

[bpstewar]

For the fire data, I am a bit confused by the method (sum vs mean). The source data measures "Expected fire events per year 1998-2010" - are you taking the sum over those years, or the average of those years? I will accept the PR for now, but we should double check this and all the metadata at some point as we do away with the xlsx file