block_reduce loses mask data

[marthinwurer]

As per https://stackoverflow.com/a/56331273/3000741, using the scipy block reduce loses all mask information. This can cause lots of calculation errors. The one that I'm running into at the moment is with the CHELSEA dataset. When I try to block reduce, the masked array is filled with the no data value and the mean calculation is completely thrown off in blocks with masked pixels.

[ozak]

Sorry @marthinwurer I have been busy so have not had time to look into this. If you can put some code to show the issue with the data you are using (or a subset of it) I may be able to look at it once I have some time. In the tests I had performed using the np.ma.sum or similar functions seemed to work correctly, but may have been just the use case I had. It would be good to ensure this is properly treated.