revisions of ObsFcstAna postprocessing package

[gmao-rreichle]

Minor fixes to python package for post-processing ObsFcstAna output.

cc: @amfox37

[gmao-rreichle]

@gmao-qliu, I started a new PR for the ObsFcstAna package because I forgot to ask you about a couple things before merging #87:

• "domain" was specified in exp_sup1, but there was no check on consistency with "domain" of exp_main. I think that as long as the two experiments are in the same tile space, cross-masking would work even when the domains are different. But I'm not sure if this general functionality would have worked, and it certainly wasn't documented. It's probably simplest to require that the "domain" is the same across all experiments, and to specify "domain" only for exp_main. I took a stab at implementing this in the first commit of the PR. Please double check.
• Re.

  GEOSldas_GridComp/GEOSldas_App/util/postproc/ObsFcstAna_stats/Plot_stats_maps.py

  Line 141 in a629ef2

  grid_data, uy, ux = remap_1d_to_2d(tile_data, lat_1d = tc['com_lat'], lon_1d = tc['com_lon'])

  Does this work for a cube-sphere tile space? For the cs tile space, the obs are effectively on an unstructured grid (wherever the com_lat/lon of the tiles happen to be), and the 2d array would probably require huge x and y dimensions (effectively N_tile-by-N_tile), which may crash the node. We haven't had much experience with plotting obs space maps when the model is in cs-space. I think we probably need the true "tile2grid()" functionality, which isn't what remap_1d_to_2d() provides. Perhaps I'm misreading the code. I suspect this needs further work.

[amfox37]

I've been using this package quite extensively over the last few days and it seems great for my use cases at least.
The only question I have is whether the Plot_stats_timeseries.py should combine all species into a single mean by default
i.e. OmFm = np.nansum(OmFm*Nobsm)/Nobsm
As this then means the spatial_stats file is a little inconsistent with the temporal_stats file which has output by species.

[gmao-qliu]

I've been using this package quite extensively over the last few days and it seems great for my use cases at least. The only question I have is whether the Plot_stats_timeseries.py should combine all species into a single mean by default i.e. OmFm = np.nansum(OmFm*Nobsm)/Nobsm As this then means the spatial_stats file is a little inconsistent with the temporal_stats file which has output by species.

I can see that one might want to save stats of individual species and only combine all species before plotting. I modified to enable it.

[gmao-qliu]

@gmao-rreichle @amfox37 please review the changes made.

[gmao-rreichle]

@gmao-qliu : The main scripts are still rather confusing. I added a few commits and comments but ran out of time to really wrap my mind around the whole thing. In a nutshell, I think we need to clearly separate the tasks accomplished by the various "main" scripts. If we have plot functions, they should focus on plotting and avoid doing things like averaging metrics across species or applying the Nmin threshold. The latter would be better off in the main() portion of the respective scripts. Not sure if all of the comments taken together make sense. We can discuss in person tomorrow

[gmao-qliu]

I've completed the latest round of changes. Please review and test. @gmao-rreichle @amfox37

[gmao-rreichle]

@gmao-qliu : I made a few more changes (https://github.com/GEOS-ESM/GEOSldas_GridComp/compare/f5fa08c..b4db565?w=1; white-space changes hidden) and added a few comments below.