Is it correct for regrid_to_lower_res() to assume second axis is always latitude?

[tomvothecoder]

I'm refactoring this function in PR #677. It compares the number of latitude coordinates to determine which variable is lower
resolution.

e3sm_diags/e3sm_diags/driver/utils/general.py

Lines 240 to 248 in d0aa16b

	def regrid_to_lower_res(mv1, mv2, regrid_tool, regrid_method):
	"""Regrid transient variable toward lower resolution of two variables."""
	axes1 = mv1.getAxisList()
	axes2 = mv2.getAxisList()
	# use nlat to decide data resolution, higher number means higher data
	# resolution. For the difference plot, regrid toward lower resolution
	if len(axes1[1]) <= len(axes2[1]):

I noticed that this function assumes the second element in .getAxisList() is latitude. However, I'm using a test dataset that has a Z axis, lev which is the second axis (example below). Is regrid_to_lower_res() incorrectly comparing the length of lev rather than lat when the dataset has a Z axis?

import cdms2

filepath =  "qa/658-lat-lon-set/cl_Amon_E3SM-2-0_historical_r1i1p1f1_gr_200001-201412.nc"
ds = cdms2.open(filepath)

cl = ds["cl"]

axes1 = cl.getAxisList()

axes1[1]

#    id: lev
#    Designated a level axis.
#    units:  hybrid_sigma_pressure
#    Length: 30
#    First:  3.6434658
#    Last:   992.5561
#    Other axis attributes:
#       axis: Z
#       long_name: hybrid level at layer midpoints (1000*(A+B))
#       positive: down
#       A_var: hyam
#       B_var: hybm
#       P0_var: P0
#       PS_var: PS
#       realtopology: linear
#    Python id:  0x7f58f03254b0

Additional Information

The call stack:

e3sm_diags/e3sm_diags/driver/lat_lon_driver.py

Lines 216 to 218 in d0aa16b

	create_and_save_data_and_metrics(
	parameter, mv1_domain, mv2_domain
	)

e3sm_diags/e3sm_diags/driver/lat_lon_driver.py

Lines 21 to 32 in d0aa16b

	def create_and_save_data_and_metrics(parameter, mv1_domain, mv2_domain):
	if not parameter.model_only:
	# Regrid towards the lower resolution of the two
	# variables for calculating the difference.
	mv1_reg, mv2_reg = utils.general.regrid_to_lower_res(
	mv1_domain,
	mv2_domain,
	parameter.regrid_tool,
	parameter.regrid_method,
	)
	diff = mv1_reg - mv2_reg

[tomvothecoder]

Do you have any insight @chengzhuzhang?

[chengzhuzhang]

regrid_to_lower_res() is always used with assumption that two 2D variables are being compared. In the case of 3D variable is supplied, to extract one vertical slice is an operation as prerequisite. When refactoring, we should use more specific axis, rather than asume second axis is always latitude.
i.e. it seems like using cmorized dataset for testing revealed a couple weak points, which is good to have. Existing tests mostly use native-ish E3SM output..

[tomvothecoder]

Gotcha, I now understand that this function assumes the variables are 2D.

In the case of 3D variable is supplied, to extract one vertical slice is an operation as prerequisite.

I now see the code for subsetting here:

e3sm_diags/e3sm_diags/driver/lat_lon_driver.py

Lines 165 to 183 in d0aa16b

	if mv1.getLevel() and mv2.getLevel():
	plev = parameter.plevs
	logger.info("Selected pressure level: {}".format(plev))
	mv1_p = utils.general.convert_to_pressure_levels(
	mv1, plev, test_data, var, season
	)
	mv2_p = utils.general.convert_to_pressure_levels(
	mv2, plev, ref_data, var, season
	)
	# Select plev.
	for ilev in range(len(plev)):
	mv1 = mv1_p[
	ilev,
	]
	mv2 = mv2_p[
	ilev,
	]

When refactoring, we should use more specific axis, rather than asume second axis is always latitude.

I used the handy lat = xcdat.get_dim_coords(dataarray, axis="Y") for this!

[chengzhuzhang]

I used the handy lat = xcdat.get_dim_coords(dataarray, axis="Y") for this!

This is much more elegant!