Up to Specific issues in real coupled models
Good morning I'm struggling with a strange problem in EC-Earth 4 (OpenIFS 48r1 + NEMO 4.2.2 and OASIS3-MCT 5.2). The error is (oasis_advance_map) ERROR: glbpos sumdst is zero but sumsrc is not which happens when GLBPOS is applied to a field which is non-zero on the source grid but zero on the destination grid. The remapping is LOCTRANS SCRIPR CONSERV AVERAGE BILINEAR D SCALAR LATITUDE 1 GLBPOS opt The atmosphere is TL255 (about 80 km) and the ocean is eORCA1 (1deg, no ice cavities). I ran with EXPOUT and extra debug information and found the following: 1) The error occurs on one day in October when there is snowfall in OpenIFS over only one ocean point of about 8e-11 kg/m2/s, but when the snowfall from this point is remapped to NEMO it becomes 0. 2) The grid point where snowfall occurs is around 180W and 77S, so near the Antarctic ice shelf. This lon,lat is not masked on either OpenIFS or NEMO grids, i.e. 0 in masks.nc. 3) The masks in the rmp file are identical to those in masks.nc but with opposite definition, i.e. src_grid_imask = 1 - TL255.mask. I can't find anywhere in the OASIS code that does this change. 4) The remap_matrix has several links for the grid point with snowfall, all of which have weight > 0, so the nearby points on the NEMO grid should receive snowfall > 0, but they do not. I confirmed this by printing all fields with EXPOUT. My questions are: a) Why does the snowfall on the NEMO grid become 0 when the snowfall in OpenIFS is not zero and all remapping weights are > 0? b) Does it make sense that the masks in masks.nc are 0 over ocean and 1 over land, but the opposite is true for src_grid_imask and dst_grid_imask in the rmp file? c) The snowfall is quite small, 8e-11. Could it be a precision problem? Any ideas on how to pinpoint and solve the problem are much appreciated! I'm happy to share plots and data if needed. Many thanks /Joakim
HI Joakim, Sorry for the delay. Did you progress on your problem? Here are my answers: a) Hum, not sure. b) yes, I observe the same in my masks.nc and rmp files c) I guess it could be. But also I am not sure what a bilinear interpolation should give if only one source point has a value > 0. Did you try with a nearest-neighbour interpolation? Sorry to be so late and not of much help here. Regards, Sophie
