search

IAMconsortium / concordia

Apache License 2.0
0 stars 3 forks source link

Land-sea mask inconsistent with CMIP6 in LUC files #56

Open gidden opened 3 months ago

gidden commented 3 months ago

Originally opened by @dariak-bsc in #31

I ran the checker with the mask taken not from the reference file but from the corresponding variable in the file which is being checked, e.g. if I am checking the variable c3ann in the file multiple-states_input4MIPs_landState_RESCUE_PIK-MAgPIE-4-7-C-SSP2EU-NPi-2024.4.25_gn_2015-2100.nc then I am taking the missing values in c3ann at the first timestep as the mask to check the following timesteps. And in this case, I am not getting any missing values at the following timesteps. When I take the mask from the reference file, then I get missing values.

Is it fair to say (@dariak-bsc / @etiennesky / @pascal-sauer) that the issue here is what grid cells are considered land vs. ocean?

Here is the map showing the delta between the reference data and the forcings land use mask: image

gidden commented 3 months ago

@dariak-bsc just for clarity, can you tell us what files precisely are being compared in the above image?

etiennesky commented 3 months ago

yes it seems like the issue is at the border of oceans and inner lakes, so perhaps it's the remapping method that needs minor tweaking (e.g. using nearest neighbor or a binary land-sea mask)

pascal-sauer commented 3 months ago

I'm working on a nearest neighbor approach

dariak-bsc commented 3 months ago

@dariak-bsc just for clarity, can you tell us what files precisely are being compared in the above image?

Hi @gidden , for this image, I took the files on the ftp:

I extracted the masks as the mask for c3ann at the first timestep for both files, and then calculated the difference between the mask for the file (1) and the mask for the file (2).

I did the same for the historical file provided by @etiennesky but it was equal to the mask for the file (2).