zonal stats on a categorical raster

[dcherian]

I just had someone ask about zonal_stats with a categorical raster where the only "stat" that makes sense is a count (or perhaps area). Example: https://pythonhosted.org/rasterstats/manual.html?highlight=categorical#working-with-categorical-rasters

Is this supported or straightforward to implement?

[martinfleis]

aggregations can be specified using a string representing an aggregation method available as DataArray/Dataset or *GroupBy methods like DataArray.mean, DataArray.min or DataArray.max. Alternatively, you can pass a callable accepted by DataArray/Dataset.reduce. Alternatively, you can pass a tuple in a format (name, func) where name is used as a coordinate and func is either known string as above or a callable, or (name, func, {kwargs}), if you need to pass additional keyword arguments.

From https://xvec.readthedocs.io/en/stable/zonal_stats.html#aggregation-options

With categorical data you will probably need to turn the categories to a dimension on its own and return counts per unique category, which I am not certain you can achieve with a custom func. But it should not be too complicated to implement that, we have a similar logic for quantiles for example if you want to report multiple.

[dcherian]

For the rasterize version this could "just" become a multi-variable GroupBy:

xvec/xvec/zonal.py

Line 65 in 83cb199

groups = acc._obj.groupby(xr.DataArray(labels, dims=(y_coords, x_coords)))