Refactor Spatial Averaging: More Robust Handling of Longitude Spanning Prime Meridian

[pochedls]

What versions of software are you using?

• Package Version: CDAT v0.1.0

What are the steps to reproduce this issue?

fn = '/p/css03/esgf_publish/CMIP6/CMIP/CSIRO/ACCESS-ESM1-5/historical/r10i1p1f1/Amon/tas/gn/v20200605/tas_Amon_ACCESS-ESM1-5_historical_r10i1p1f1_gn_185001-201412.nc'
ds = xcdat.open_dataset(fn)
tsa = ds.xcdat.spatial_avg(lat_bounds=(-30, 30), lon_bounds=(0, 360)).tas.load()
print(tsa[0].values)

297.44541842495704

But the correct value is: 297.150202135895 (Difference of 0.31 K)

I think this can be traced to this line: domain_bounds should be dom_bounds in this case.

This will fix the issue in this case, but I realize there are specific situations that will give an (slightly) incorrect solution when the specified regional bounds are within a grid cell that spans across the prime meridian (e.g., -1 to 1).

Any other comments?

This code should be refactored to more robustly deal with averaging along the longitude axis (it needs to robustly handle domain bounds and regional selections that span the prime meridian). An approach I would like to try is to:

1. Convert longitude axes to 0 to 360 domain
2. If a grid cell boundary crosses the prime meridian, concatenate a version of that grid cell spanning 0 and 360 (e.g., [-1, 1] and [359, 361])
3. Limit the bounds of each to within 0 to 360 (e.g., [0, 1] and [359, 360]) so that the domain is the same, but now represented by two grid cells
4. Calculate weights based on this new dimension bounds array
5. Add the weight from each of the repeated grid cells together (so that the weight array is the length of the longitude dimension)