search

google-research / weatherbench2

A benchmark for the next generation of data-driven global weather models.
https://weatherbench2.readthedocs.io
Apache License 2.0
400 stars 40 forks source link

Mean radiation fluxes contain NaNs #124

Open Arcomano1234 opened 6 months ago

Arcomano1234 commented 6 months ago

It seems that the mean radiation fluxes (e.g., mean_top_net_short_wave_radiation_flux, mean_surface_net_long_wave_radiation_flux, etc) are completely NaNs (at least for the select years I have checked). 

import xarray as xr

#Using 1959-2023_01_10-wb13-6h-1440x721_with_derived_variables.zarr
ds = xr.open_zarr('gs://weatherbench2/datasets/era5/1959-2023_01_10-wb13-6h-1440x721_with_derived_variables.zarr')

#Printing out a random date
ds.sel(time='1990-01-01-06')['mean_surface_net_long_wave_radiation_flux'].values

#Returns
#array([[nan, nan, nan, ..., nan, nan, nan],
#       [nan, nan, nan, ..., nan, nan, nan],
#       [nan, nan, nan, ..., nan, nan, nan],
#       ...,
#       [nan, nan, nan, ..., nan, nan, nan],
#       [nan, nan, nan, ..., nan, nan, nan],
#       [nan, nan, nan, ..., nan, nan, nan]], dtype=float32)

The variables that I have checked that are NaNs are mean_top_net_short_wave_radiation_flux, mean_surface_net_long_wave_radiation_flux, mean_surface_net_short_wave_radiation_flux, mean_top_downward_short_wave_radiation_flux, and mean_top_net_long_wave_radiation_flux.

Arcomano1234 commented 5 months ago

For those who are interested in these variables, I was able to get them from https://github.com/google-research/arco-era5/ on the same grid as WB2 (I believe).