from_latlon_2d fill_value treatment or masking?

[michaelleerilee]

Earth Scientists have a habit of stuffing weird values into their data arrays, forcing tools to deal with these weird cases and semantics.

Is there a general way to handle this?

I'm currently looking at lon-lat arrays for GOES data that only has lon-lat where the Earth shows up on the image.

Maybe there's no general way to handle these cases, because we're intermediate between a general masked array an array with a regular pattern, and I'll have to pass cropped individual image rows to get the adaptive resolution.