-
Notifications
You must be signed in to change notification settings - Fork 43
Description
Describe the bug
In #2672 , unwanted coordinates were removed from the source cube in order to perform the regridding.
However, if a recipe contains a dataset that has a horizontal grid close enough to the target_grid
, and therefore the dataset is ultimately not regridded, the cube loses necessary coordinates and steps like multi_model_statistics
cannot be computed afterwards because the affected cube is missing coordinates:
ValueError: Multi-model statistics failed to merge input cubes into a single array:
0: surface_temperature / (K) (time: 365; -- : 412; -- : 424)
1: surface_temperature / (K) (time: 365; grid_latitude: 412; grid_longitude: 424)
Coordinates in cube.dim_coords differ: grid_latitude, grid_longitude.
Coordinate-to-dimension mapping differs for cube.dim_coords.
A solution would be to perform this removal after the check for the horizontal grid is performed.
Example recipe with two datasets with the same grid that then fails at the multi_model_statistics step:
documentation:
title: |
test
authors:
- loosveldt-tomas_saskia
description: |
test
preprocessors:
ts_pp:
regrid:
target_grid: RCA4
scheme: linear
multi_model_statistics:
span: overlap
statistics: [mean]
datasets:
- {project: CORDEX, domain: EUR-11, dataset: RCA4, driver: CNRM-CERFACS-CNRM-CM5, ensemble: r1i1p1, institute: SMHI, rcm_version: v1}
- {project: CORDEX, domain: EUR-11, dataset: REMO2015, driver: IPSL-IPSL-CM5A-LR, ensemble: r1i1p1, institute: GERICS, rcm_version: v1}
diagnostics:
bias:
variables:
ts_cordex:
short_name: ts
mip: day
exp: historical
timerange: 2003/2003
preprocessor: ts_pp
scripts: null
The regridding step could be skipped in this case, but in cases like recipe ESMValGroup/ESMValTool#4199 where there are datasets in the same grid and datasets in a different grid, this issue prevents the recipe from running properly.
Metadata
Metadata
Assignees
Labels
Type
Projects
Status