Regrid

Nearest neighbor (fastest; good for masks/categorical data)

cdo remapnn,TARGET_GRID input.nc output.nc

Bilinear interpolation (smooth continuous fields; default for coarse → fine)

cdo remapbil,TARGET_GRID input.nc output.nc

Conservative (flux-preserving; default for fine → coarse)

cdo remapcon,TARGET_GRID input.nc output.nc

With precomputed weights (fast repeated regridding to same grid)

# Generate weights once
cdo gencon,TARGET_GRID input.nc weights.nc
# Apply weights
cdo remap,TARGET_GRID,weights.nc input.nc output.nc

Defining a Target Grid

Pass a grid description file or an existing NetCDF file as TARGET_GRID.

Grid description file (lonlat example)

gridtype  = lonlat
xsize     = 1440
ysize     = 721
xfirst    = -179.875
xinc      = 0.25
yfirst    = -90.0
yinc      = 0.25

Extract grid from an existing file

cdo griddes reference.nc > target_grid.txt
cdo remapbil,target_grid.txt input.nc output.nc

Useful CDO Flags

• -O — overwrite output
• -R — reset time counter
• -P N — parallel processing (N threads)
• -setmisstoc,VALUE — replace missing values with VALUE
• -selname,VAR — select variable before regridding
• -mergetime — merge multiple time-step files

Common Workflows

Single file

cdo -O remapnn,target_grid.txt input.nc output.nc

Batch regrid (shell loop with parallelism)

for f in input_*.nc; do
    cdo -O -P 8 remapbil,target_grid.txt "${f}" "${f%.nc}.regrid.nc"
done

Merge time steps then regrid

cdo -O mergetime input_*.nc merged.nc
cdo -O remapcon,target_grid.txt merged.nc output.nc

Fix missing coordinate attributes before regridding (ncatted)

ncatted -a units,lon,c,c,"degrees_east" \
        -a units,lat,c,c,"degrees_north" \
        input.nc input_fixed.nc
cdo remapnn,target_grid.txt input_fixed.nc output.nc

NCO / ncremap (ESMF-based alternative)

# Basic usage
ncremap -i input.nc -m weights.nc -o output.nc

# With MPAS projection and missing value
ncremap -i input.nc -m weights.nc -o output.nc -P mpas --mss_val=-9999

Requires the E3SM Unified environment or nco with ESMF support.

Python / Dask Distributed Regridding

For large batches, wrap CDO with Dask:

import subprocess
from dask.distributed import LocalCluster, Client

cluster = LocalCluster(n_workers=8)
client = Client(cluster)

def regrid_file(fn):
    out = fn.replace('.nc', '_regrid.nc')
    cmd = f"cdo -O remap,{GRID},{WEIGHTS} -setmisstoc,-99999 {fn} {out}"
    subprocess.run(cmd, shell=True, check=True)

futures = client.map(regrid_file, file_list)
client.gather(futures)

Alternatively use xesmf for pure-Python regridding:

import xesmf as xe
import xarray as xr

ds_in = xr.open_dataset("input.nc")
ds_out = xr.Dataset({"lat": (["lat"], lats), "lon": (["lon"], lons)})
regridder = xe.Regridder(ds_in, ds_out, "bilinear")
ds_regridded = regridder(ds_in)