Finite missing-data placeholders

[denispauline]

[Comment from Kenny]

GETM/BFM uses -9998 and -9999 as missing-data placeholders in some outputs. The validation workflow treated these as finite values, allowing them to enter interpolation and produce very large negative but apparently valid results.

Would you prefer missing data sentinels to be configurable globally for each model, separately for each variable, or both?

[acapet]

Normally, netcdf standards oblige to include "_Fill_Value" as an variable attribute (or similar, but the name of the attributed is standardized).

This is normally interpretated automatically by xarray, when reading a .nc file in python, resulting in xarray data sets containing NaNs (and treated as such in downstream operations).

If that has not been the case in your tests, I'd suspect that this attributes is missing, and having your model producing it would be the better way to handle this (it enforces NC-compliance of your output files).

Another, less god, but more permissive ways, would be to allow the .yml file to give the "fill value", and process files according ly (I think it can be given as an argument when loading the data in xarray).