Update ocean grid inputs to mosaic format

[dougiesquire]

Our ESM1.6 configurations currently still use the old FMS format to specify the ocean grid, inherited from ESM1.5. We should update these to use the newer mosaic format. See ACCESS-NRI/access-esm1.5-configs#23 for more details.

[access-hive-bot]

This issue has been mentioned on ACCESS Hive Community Forum. There might be relevant details there:

https://forum.access-hive.org.au/t/csiro-access-nri-standup-minutes/3789/26

[access-hive-bot]

This issue has been mentioned on ACCESS Hive Community Forum. There might be relevant details there:

https://forum.access-hive.org.au/t/coordinating-development-threads-for-next-phase-of-control-run/4523/1

[MartinDix]

Remapping weights in /g/data/vk83/configurations/inputs/access-esm1p5/modern/share/coupler/remapping_weights/global.oi_1deg.a_N96/2020.05.19 were originally created by OASIS. They can be recreated in a model run where the remapping weights are not initially present. This uses the grids.nc and masks.nc files (provenance of these is unknown). The areas.nc file is not required.

[dougiesquire]

They can be recreated in a model run where the remapping weights are not initially present. This uses the grids.nc and masks.nc files (provenance of these is unknown). The areas.nc file is not required.

Thanks @MartinDix. I tried this, but the remapping weights files generated on the fly were different than what we have in the config. Historical reproducibility tests also failed.

[MartinDix]

Original files are from 2012 so would have been created with a different version of OASIS and built with a much older compiler so not surprising there are differences. rmp_um1t_to_cice_CONSERV_DESTAREA.nc has roundoff level differences in the weights (~2e-8) but all the FRACNNEI files have large differences in the number of weights. E.g. for rmp_cice_to_um1t_CONSERV_FRACNNEI.nc 247204 .c.f. 165341.

ACCESS-CM2 files (in /g/data/access/projects/access/access-cm2/input_O1/cpl_n96) were created in 2014 and have a similar number of weights to newly created ESM1-6 files.

[dougiesquire]

This notebook documents updating the ocean grid and associated files: https://gist.github.com/dougiesquire/44c245749705fee7883e164ee38da083

[MartinDix]

Compared the original /g/data/vk83/configurations/inputs/access-esm1p5/modern/share/coupler/remapping_weights/global.oi_1deg.a_N96/2020.05.19/rmp_cice_to_um1t_CONSERV_FRACNNEI.nc and ones recreated by running the model (with the original ocean grid). In the original weights file there are no zero values in remap_matrix but in the new one there are 81863 so the number of non-zero weights is the same as the original file.

Using the old and new weights to remap the SST from a coupler restart i2a.nc to the atmospheric grid shows roundoff level differences in the tripole region.

[dougiesquire]

Thanks for checking this @MartinDix

[MartinDix]

Using the weights from the new MOM grid gives a max SST difference of 4e-5 at a point in the North Pole row. The UM zonally averages this but there's still a zonal mean difference of 3e-7. Nothing to worry about but it is a bigger perturbation than just recreating the weights.

The Helmholtz solver in the UM amplifies any differences very quickly. Here RMS differences of daily surface temperature wrt the standard run.

[dougiesquire]

In a recent meeting, we (@MartinDix, @AndyHoggANU, @blimlim, @chrisb13) went through the results from the comparison runs presented in this notebook. We agreed that we're happy to progress with the new grid, but we'd like to understand the cause of the ocean volume change:

I've now looked into this a little more. The reduction in volume with the updated grid is due to a reduction in the total number of wet cells relative to the current grid. Strangely, the cells that change from wet to dry are only at depth indexes 22 - 26 (depths 241.3m - 536.7m). The following plots show these cell locations in yellow:

And here's a zoomed-in example at 343.3 m (colors are t-cell thickness):

I'm going to push ahead with the grid update unless anyone opposes here. (Note the figures in this post are all available in this notebook: https://gist.github.com/dougiesquire/44c245749705fee7883e164ee38da083)

[AndyHoggANU]

Hmm. This is interesting. It might be that the shift in grid cell position is somehow changing the depth, and that perhaps we need to remake the bathymetry to get exact reproduction. But perhaps it’s getting down into the weeds and we should accept where we are and move on.