This PR updates the coarsening method for snoalb (maximum snow albedo), shdmin (minimum green vegetation fraction), and shdmax (maximum green vegetation fraction) from a block minimum or maximum over the dominant surface type to a simple area-weighted average over the dominant surface type.
This is because an area-weighted average over the dominant surface type is a more physically reasonable approach (a block min or block max emphasizes the extreme of a single fine grid cell, while an average is more representative of all fine grid cells in a coarse grid cell). We have known this for a while, but just have not gotten around to updating our code. The main variable where this has an important impact is snoalb, but for consistency we remove this block min / max coarse-graining strategy from the shdmin / shdmax variables as well.
This notebook from early 2022 illustrates that biases are improved in baseline simulations started from initial conditions where these fields were coarsened in this way, when compared with the previous approach, particularly for upward shortwave radiative flux at the surface (we get a 37% improvement in the RMSE of the one-year time mean over land).
Significant internal changes:
Coarsen snoalb, shdmin, and shdmax in sfc_data restart files with an area-weighted average over the dominant surface type instead of a block minimum or maximum over the dominant surface type.
This PR updates the coarsening method for
snoalb(maximum snow albedo),shdmin(minimum green vegetation fraction), andshdmax(maximum green vegetation fraction) from a block minimum or maximum over the dominant surface type to a simple area-weighted average over the dominant surface type.This is because an area-weighted average over the dominant surface type is a more physically reasonable approach (a block min or block max emphasizes the extreme of a single fine grid cell, while an average is more representative of all fine grid cells in a coarse grid cell). We have known this for a while, but just have not gotten around to updating our code. The main variable where this has an important impact is
snoalb, but for consistency we remove this block min / max coarse-graining strategy from theshdmin/shdmaxvariables as well.This notebook from early 2022 illustrates that biases are improved in baseline simulations started from initial conditions where these fields were coarsened in this way, when compared with the previous approach, particularly for upward shortwave radiative flux at the surface (we get a 37% improvement in the RMSE of the one-year time mean over land).
Significant internal changes:
snoalb,shdmin, andshdmaxinsfc_datarestart files with an area-weighted average over the dominant surface type instead of a block minimum or maximum over the dominant surface type.xref: ai2cm/fv3gfs-fortran#372
Coverage reports (updated automatically):