seasonal PSM calibration takes a long time

[frodre]

I believe seasonal PSM calibration takes a long time because it is reloading the calibration dataset for each season defined. This can be fixed by holding the original calibration in memory, or if it's too big, resorting to using Dask to run the re-averaging operations. In my experience Dask has only worked with HDF5 files, but since we've upgraded to Python 3 this may be fixed for netCDF.

@rtardif Trying to configure just like P2B now that my seasonal PSMs are built (31hrs!)-- do I set all proxy_psm_type to bilinear in the config yml file, like this: -- dmanderson

[brews]

xarray chunking might be an easy solution for the netcdf files? (uses dask under the hood) -- just a thought.