Do not limit analysis to pixels where all data is valid.
Note it is important that the NPP stats are not restricted to pixels where all data is valid, handling of fill-values should be done on a variable by variable basis.
This is tricky to say the least, because the way PyCAMA is set up. We can rather easily remove variables from the analysis that are all fill value for a granule, but handling variables with some fill values without removing those is not foreseen.
Because the way other products are handled, I don't really want to let fill values past the filter I have now. This would increase the burden of testing significantly.
Can you be more specific on the occurrence of these fill values, paterns, reasons etc.
The VIIRS radiances need treating totally independently as they come from separate files (different to cloud mask and different for each VIIRS band). Its clearly possible that one or other file could be missing - our code will still report cloud-mask if radiance in channel X is missing and radiance in channel X if channel Y is missing etc; would not want to stop monitoring things because one or other VIIRS file is missing. Is that enough?