Soil level clm surface input data for clm5.2.0 have missing values in large domains #2744
Description
Brief summary of bug
Input surface data for clm on spectral element grid ne30 (inputdata/lnd/clm2/surfdata_esmf/ctsm5.2.0/surfdata_ne30np4.pg3_hist_1850_78pfts_c240216.nc) at least for clm5.2.0 have weird/missing values for parts of the grid.
Here is an example showing organic carbon summed over the soil levels and similar patterns of missing data seem to be present in all the soil level data I've tried plotting including PCT_CLAY, PCT_SAND:
This pattern shows up similarly in output datasets
General bug information
Plotting and regridding input data from surface input data for spectral element grid ne30 for clm5.2.0 (inputdata/lnd/clm2/surfdata_esmf/ctsm5.2.0/surfdata_ne30np4.pg3_hist_1850_78pfts_c240216.nc) data for variables on soil levels seem to be looking weird and missing data for what looks like some of the spectral element domains.
Above is an example plot of what organic carbon summed over soil levels looks like, but the pattern of missing data seems to be the same for all the soil level data that I have plotted so far. Looking at the LMR picture framing from https://journals.ametsoc.org/view/journals/mwre/134/12/mwr3360.1.xml I can almost see at least some of the domains that are missing, here is an attempt at colouring in the missing parts:
CTSM version you are using: [output of git describe]
clm5.2.0 ne30
Does this bug cause significantly incorrect results in the model's science? [Yes ]
Configurations affected: [Fill this in if known.]
Details of bug
See above
Important details of your setup / configuration so we can reproduce the bug
I regridded using the xesmf package following this excellent guide: https://ncar.github.io/esds/posts/2022/cam-se-regridding/
For reference here is my regridding code snippet
import numpy as np
import xarray as xr
import xesmf
def make_se_regridder(weight_file):
weights = xr.open_dataset(weight_file)
in_shape = weights.src_grid_dims.load().data
# Since xESMF expects 2D vars, we'll insert a dummy dimension of size-1
if len(in_shape) == 1:
in_shape = [1, in_shape.item()]
# output variable shape
out_shape = weights.dst_grid_dims.load().data.tolist()[::-1]
dummy_in = xr.Dataset(
{
"lat": ("lat", np.empty((in_shape[0],))),
"lon": ("lon", np.empty((in_shape[1],))),
}
)
dummy_out = xr.Dataset(
{
"lat": ("lat", weights.yc_b.data.reshape(out_shape)[:, 0]),
"lon": ("lon", weights.xc_b.data.reshape(out_shape)[0, :]),
}
)
regridder = xesmf.Regridder(
dummy_in,
dummy_out,
weights=weight_file,
method="bilinear",
reuse_weights=True,
periodic=True,
)
return regridder
def regrid_se_data(regridder, data_to_regrid):
print(data_to_regrid.dims)
if isinstance(data_to_regrid, xr.DataArray):
print(type(data_to_regrid))
updated = data_to_regrid.copy().transpose(..., "lndgrid").expand_dims("dummy", axis=-2)
else:
vars_with_ncol = [name for name in data_to_regrid.variables if "ncol" in data_to_regrid[name].dims]
updated = data_to_regrid.copy().update(
data_to_regrid[vars_with_ncol].transpose(..., "lndgrid").expand_dims("dummy", axis=-2)
)
regridded = regridder(updated.rename({"dummy": "lat", "lndgrid": "lon"}))
return regridded
I used this file for mapping inputdata/lnd/clm2/mappingdata/maps/ne30pg3/map_ne30pg3_to_0.5x0.5_nomask_aave_da_c180515.nc
The regridding looked ok and completely sane for other variables that didn't have soil level, and the pattern persisted regardless of whether I regridded a variable with the soil level dimension, or if I summed over and thereby eliminated the soil level dimension before regridding. It also looked the same if I regridded the same output variables from the first history files from a run.
Important output or errors that show the problem
See above