Using xr.apply_ufunc in a masked DataArray in an efficient way

[ricardobarroslourenco]

This one is related to this other thread (#7591).

I'm applying multiple regressions over lat,lon pairs using apply_ufunc, but I have noticed (by simply printing within my function) that the apply does not take in consideration the masking I have done in the data (so several values with the where clause are NaN's, such any water body for example).

How should I deal with this situation? Is there a way to constrain the apply_ufunc only on valid data?

[dcherian]

No, that's not possible. You could short-circuit the all-NaN case in your custom function though.

So

if np.isnan(x).all():
    return np.nan

Or something lilke that

[ricardobarroslourenco]

Thanks for the answer. It is good to know.