Incorrect dimension labelling for multi-cube NetCDF save

[bjlittle]

🐛 Bug Report

iris fails to correctly save multiple-cubes with varying dimensionality, all of which are based on the same 2D geospatial auxiliary coordinates nat_lat and nav_lon.

How To Reproduce

Steps to reproduce the behaviour:

The following example contains a mixture of 3D and 4D cubes, all of which have common 2D x and y coordinates (nav_lat and nav_lon), which can be successfully loaded by iris, but incorrectly serialises them to a NetCDF file e.g.,

import iris

fname = "nemo_dn353o_1m_19000101-19000201_grid-T.nc"
cubes = iris.load(fname)
iris.save(cubes, "out.nc")

The example file cube list contains the following cubes:

0: sea_surface_conservative_temperature / (degC) (time: 1; -- : 331; -- : 360)
1: icb heat flux to ocean due to melting latent heat / (J/m2/s) (time: 1; -- : 331; -- : 360)
2: Evap minus Precip over ice / (kg/m2/s) (time: 1; -- : 331; -- : 360)
3: Ice shelf heat content flux of injected water ( from isf to oce ) / (W/m2) (time: 1; -- : 331; -- : 360)
4: sea_surface_absolute_salinity / (g/kg) (time: 1; -- : 331; -- : 360)
5: icb melt rate of icebergs / (kg/m2/s) (time: 1; -- : 331; -- : 360)
6: sea bottom conservative temperature / (degC) (time: 1; -- : 331; -- : 360)
7: sea bottom absolute salinity / (g/kg) (time: 1; -- : 331; -- : 360)
8: Ice shelf latent heat flux ( from isf to oce ) / (W/m2) (time: 1; -- : 331; -- : 360)
9: Evap minus Precip over ocean / (kg/m2/s) (time: 1; -- : 331; -- : 360)
10: Downward salt flux / (g/m2/s)       (time: 1; -- : 331; -- : 360)
11: Ice shelf fresh water flux ( from isf to oce ) / (kg/m2/s) (time: 1; -- : 331; -- : 360)
12: Water flux due to freezing/melting / (kg/m2/s) (time: 1; -- : 331; -- : 360)
13: cell_thickness / (m)                (time: 1; depth: 75; -- : 331; -- : 360) ❌
14: downwelling_shortwave_flux_in_sea_water / (W/m2) (time: 1; depth: 75; -- : 331; -- : 360) ❌
15: ocean_mixed_layer_thickness / (m)   (time: 1; -- : 331; -- : 360)
16: precipitation_flux / (kg/m2/s)      (time: 1; -- : 331; -- : 360)
17: sea_ice_area_fraction / (1)         (time: 1; -- : 331; -- : 360)
18: sea_surface_height_above_geoid / (m) (time: 1; -- : 331; -- : 360)
19: sea_water_absolute_salinity / (g/kg) (time: 1; depth: 75; -- : 331; -- : 360) ❌
20: sea_water_conservative_temperature / (degC) (time: 1; depth: 75; -- : 331; -- : 360) ❌
21: snowfall_flux / (kg/m2/s)           (time: 1; -- : 331; -- : 360)
22: square_of_sea_surface_height_above_geoid / (m2) (time: 1; -- : 331; -- : 360)
23: surface_downward_heat_flux_in_sea_water / (W/m2) (time: 1; -- : 331; -- : 360)
24: water_flux_into_sea_water_from_rivers / (kg/m2/s) (time: 1; -- : 331; -- : 360)
25: water_flux_out_of_sea_ice_and_sea_water / (kg/m2/s) (time: 1; -- : 331; -- : 360)
26: water_flux_out_of_sea_water_due_to_newtonian_relaxation / (kg/m2/s) (time: 1; -- : 331; -- : 360)

The provided example input file has the following NetCDF dimensions section:

dimensions:
        y = 331 ;
        x = 360 ;
        time_counter = UNLIMITED ; // (1 currently)
        nvertex = 4 ;
        depth = 75 ;
        axis_nbounds = 2 ;

The iris generated output NetCDF file contains the following dimensions section:

dimensions:
        time_counter = 1 ;
        dim1 = 331 ;
        dim2 = 360 ;
        bnds = 2 ;
        bnds_4 = 4 ;
        depth = 75 ;
        dim2_0 = 331 ;
        dim3 = 360 ;

Given the above we have the following NetCDF dimensions mapping, as iris has saved a 3D cube first)

• y is mapped to dim1 ✅
• x is mapped to dim2 ✅
• time_counter is mapped to time_counter ✅
• nvertex is mapped to bnds_4 ✅
• depth is mapped to depth ✅

The following mapping behaviour is the issue:

• dim2_0 is akin to dim1 (i.e., y) but for a 4D cube ❌
• dim3 is akin to dim2 (i.e., x) but for a 4D cube ❌

Unfortunately the nav_lat and nav_lon coordinates are first saved referencing (dim1, dim2) and not (dim2_0, dim3).

This behaviour would be different but also incorrect if a 4D cube was saved first from the cube list. Hence the 4D cubes are incorrectly serialised as the nav_lat and nav_lon do not match the dimension names associated with th 4D NetCDF variables.

Expected behaviour

No duplication of spatial dimensions.

Environment

• Applies up to the latest development version of iris

[bjlittle]

Example data for this issue which has been archived under ${DATADIR}/iris-issue-6486-data for the MO team account 👍