Summary

While converting WG output data from Zarr to NetCDF, most grid points appear as missing (NaN or _). This affects variables like t2m and results in mostly empty output files.

Observed Behavior

Most grid points are missing (NaN or _), with only a few valid values.

Steps to Reproduce

Run the conversion script with a run ID and ERA5 sample:

uv run ./packages/evaluate/src/weathergen/evaluate/export_inference.py --run-id <RUN_ID> --stream ERA5 --output-dir ./netcdf_out --format netcdf --type prediction --sample 1

Expected Behavior

All grid points should be populated correctly after conversion or properly interpolated.

Possible Cause / Notes

The source data is in a reduced Gaussian grid, and the conversion may not handle the grid properly. Regridding or interpolation may be required.