This code was used in this 2019 paper: Intercomparison of machine learning methods for statistical downscaling: the case of daily and extreme precipitation
- python2.7
- xarray (http://xarray.pydata.org/en/stable/index.html)
- climate data operators (cdo) (https://code.zmaw.de/projects/cdo)
Merra 2 - A reanalysis dataset provided by NASA's Global Modeling and Assimilation Office. We extract preciptation from the land product to downscale. Reanalysis datasets are used to test a downscaling model's skill against an observed dataset. https://gmao.gsfc.nasa.gov/reanalysis/MERRA-2/
Prism - The prism 4km precipitation dataset is aggregated to 16km, which will be our observations. http://www.prism.oregonstate.edu/
- Interplate missing values
- Upscale Prism and remap to MERRA
- Merge all years into single files, 1 per data source
cd data
prism='prism_example.nc'
merra='merra_example.nc'
prism_upscaled='prism_upscaled.nc'
merra_filled='merra_filled.nc'
cdo griddes $merra > merra_grid
# This trick of setting and resetting the missing value seems to allow fillmiss to work
# if resetting is not done then merra_prism_example.py won't read the file correctly
cdo setmissval,nan $prism temp_miss.nc
cdo fillmiss temp_miss.nc tmp.nc
cdo setmissval,-9999 tmp.nc tmp_filled.nc
cdo remapbil,merra_grid -gridboxmean,3,3 tmp_filled.nc $prism_upscaled
cdo fillmiss $merra $merra_filled
rm tmp_filled.ncpython ../merra_prism_example.py $prism_upscaled $merra_filled ppt PRECTOTLAND merra_bc.nccdo griddes $prism > prism_grid
cdo remapbil,prism_grid merra_bc.nc merra_bc_interp.nc cdo remapbil,prism_grid $prism_upscaled prism_reinterpolated.nccdo ydayavg prism_reinterpolated.nc prism_interpolated_ydayavg.nc
cdo ydayavg $prism prism_ydayavg.nc
cdo div prism_ydayavg.nc prism_interpolated_ydayavg.nc scale_factors.ncpython ../spatial_scaling.py merra_bc_interp.nc scale_factors.nc merra_bcsd.ncThe dataset provided does not contain any bodies of water but when downscaling north america the ocean is filled with interpolated values. After spatial scaling we'll want to replace filled values with NaN. Here, we build a dataset with 1's over land and NaN over bodies of water.
cdo seltimestep,1 -div -addc,1 $prism -addc,1 $prism mask.nc
cdo mul mask.nc merra_bcsd.nc merra_bcsd_masked.nc
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent