How to regrid the netcdf files?
Here we will discuss two most common methods that are used for regridding: Tempestremap and ncremap
Acknowledgment: Dr. Min Xu and Dr. Forrest M. Hoffman, Oak Ridge National Laboratory
Some points to keep in mind before you use the following techniques:
- The variable should be assumed such that in any location within a grid, the values are same. For instance, flux density, precipitation, and temperature follow the rule.
- Use density (g m-2 s-1) variables rather than mass flux (g s-1) as conservative remapping means the area-weighting summary or average. For GPP in gC/mon, the rule is broken.
- the variable type should be double of dimensions.
- run ncdump -h on the source file, if the masked data does not show as “- -“, mask the data properly, see: link.
- The masking and _FillValue should be link 1.e36. e.g. the missing value of ERA5 dataset is -32767s and tempest does not recognize this fill and missing value.
- The units of area variable are stredians. To get the area in km2 use a multiplying facor of 6371*6371 km^2.
- If the data is of high resolution, use tempestremap.
- For other commonly used nco commands, see link
Steps for redrigging for both techniques:
1.TempestRemap (by Paul Ulrich) is a conservative, consistent and monotone remapping package for arbitrary grid geometry with support for finite volumes and finite elements.
./bin/GenerateRLLMesh --lon 720 --lat 360 --file grid_360x720.g
./bin/GenerateRLLMesh --lon 360 --lat 180 --file grid_180x360.g
./bin/GenerateOverlapMesh --a grid_360x720.g --b grid_180x360.g --out overlap_grid_360to180.g
./bin/GenerateOfflineMap --in_mesh grid_360x720.g --out_mesh grid_180x360.g \
--ov_mesh overlap_grid_360to180.g --in_np 2 \
--out_map map_tempest_360x720_to_180x360.20210701.nc
./bin/ApplyOfflineMap --map map_tempest_360x720_to_180x360.20210701.nc --var gpp\
--in_data GPP_ANN_CRUNCEPv6_Monthly_1980_2013_Updated.nc\
--out_data GPP_ANN_CRUNCEPv6_Monthly_1980_2013_tempest_180x360.nc
2.ncremap has many options of regridding that includes conservative, biliniear, nearest and also sub grade scale (sgs). sgs is especially useful incase the climate variable is dependent on landfrac.
Options:
ncremap -a aave -s src_grd.nc -g dst_grd.nc -m map.nc in.nc out.nc
- default: bilinear
- -a aave : for conserve
- -P sgs: for sub grid scale if you have landfraction
- src_grd.nc : is source grid scrip file
- dst_grd.nc : is destination grid scrip file
- -m map.nc : the weight nc file, (Min Said, if you don’t alreaddy have it, it will be created)
- in.nc : is the input file
- out.nc : output file name
export NCO_PATH_OVERRIDE='No'
export HDF5_USE_FILE_LOCKING=FALSE
module load ncl
ncks --rgr grd_ttl='Equi-Angular grid 192x288'#latlon=192,288#lat_typ=uni#lon_typ=grn_wst \
--rgr scrip=192x288_SCRIP.20210701.nc \
CESM2_ssp585_r1i1p1f1_gpp_gCm-2.nc \
test_foo_cesm.nc
ncks --rgr grd_ttl='Equi-Angular grid 64x128'#latlon=64,128#lat_typ=uni#lon_typ=grn_wst \
--rgr scrip=64x128_SCRIP.20210701.nc \
CESM2_ssp585_r1i1p1f1_gpp_gCm-2.nc \
test_fooo_cesm.nc
ncremap -a aave -P sgs \
-s 192x288_SCRIP.20210701.nc \
-g 64x128_SCRIP.20210701.nc \
-m map_aave_sgs_nco_192x288_and_64x128.20210701.nc \
CESM2_ssp585_r1i1p1f1_gpp_gCm-2.nc \
CESM2_ssp585_r1i1p1f1_gpp_gCm-2_nco_sgs_aave_64x128.nc
If you have the access to Cori Nersc, visit the dir : /global/cfs/cdirs/m2467/bharat/Regridding_Techniques. It has observation and model data and scripts designed to regrid them to coarser resolution. The file Regridding_Example.ipynb is designed to test the output of your regrided file and compare with source file.
The comparison of regridded data vs original data:
- For a random time of 1981-12-31 for Flux ANN, Total GPP:
- Original (360x720) : 7.24 PgC
- Regrid Tempest (180x360) : 6.14 PgC (when mask was NaN)
- Regrid ncremap -conserve (180x360) : 6.74 PgC (when mask was NaN)
- Regrid ncremap -bilinear (180x360) : 7.02 PgC (when mask was NaN)
- Regrid (FV) ncremap -conserve (180x360) : 7.24 PgC (when mask was “- -“)
- For a random time of 1851-12-15 for CESM2, Total GPP:
- Original (192x288) : 7.51 PgC
- Regrid Tempest (64x128) : 7.47 PgC
- Regrid ncremap -conserve (64x128) : 7.47 PgC
- Regrid ncremap -bilinear (64x128) : 7.47 PgC
- Original w/landfrac (192x288) : 6.18 PgC
- Regrid ncremap -conserve -sgs (64x128) : 6.15 PgC