Technically, should weight by area

As noted by @amoebaliz, the GLORYS SST comparison weights the statistics by area, but the OISST comparison doesn't: https://github.com/NOAA-GFDL/CEFI-regional-MOM6/blob/42d3647ae3e3c754cdd71c7958c0e1eaab6c9119/diagnostics/physics/sst_eval.py#L105

Since the OISST comparison is done on a regular lat/lon grid, it should be possible to simply use cosine(lat) as the weights.