Generate output (preds) in common geospatial formats

justlab / Earth_obs_cleaning

ML Cleaning for Earth Obs Using Collocated Datasets

1 stars 0 forks source link

Generate output (preds) in common geospatial formats #18

Open allanjust opened 2 years ago

allanjust commented 2 years ago

Potentially use gdal to generate multilayer cloud-optimized geotiff by year (UTC). One layer per overpass over the prediction region.

overpass time was tile dependent (with ~5 min variation across adjacent tiles) -- perhaps will need to pick a reference tile?
unclear what other metadata to pass through
sometimes helpful to have original (raw) data for contrast with predicted values (pred); is this just useful for our evaluation papers or should we export both values for others or as separate (optional) files?

allanjust commented 2 years ago

This looks helpful: https://geoexamples.com/other/2019-02-08-cog-tutorial/

allanjust commented 2 years ago

perhaps we could consider parquet as an efficient target store for the prediction output (as big tabular data) still within the targets workflow (and more language agnostic than fst); targets appears to leverage arrow::read_parquet() and arrow::write_parquet()