Open bziniti opened 7 years ago
This is a head scratcher because the underlying CHIRPS data know nothing about country boundaries and have no gaps. Something about the export is doing something differently starting about 2000 from what it is doing from 1980-2000.
Also true to the export files. I looked at a stack of 1999001 and 2000001. Here's a plot of the NA count.
It also must be because of this same issue that the disturbance module was giving strange results.
Here's a plot of (disturbance duration I think -- band 3):
The long durations occur along the edge. Maybe these NA were filled with something?
I changed the title of the issue because I think it is potentially applicable to other places than DRC and other data sources than CHIRPS. (We should verify that.)
But the question is why would a GIPS export for some years produce a slightly different boundary for the feature than for other?
I can confirm that this happens with a different export region (different shapefile in a different projection).
I noticed the output from multi temporal.py annual stats has missing values at border from 2000-2016.
See picture:
Maybe a projection change or interaction with gap fill??