GitHub

Interpolation and Regridding

Regridding is usually the bridge between raw datasets and a comparable climate-scenario workflow. Observations and model simulations often start on different grids, and most bias-correction methods are easiest to interpret once the fields share a common spatial support.

When to Regrid

You should usually regrid when:

  • observations and simulations are on different regular grids
  • the simulation uses a rotated pole or curvilinear mesh
  • you need a common analysis grid for multi-model comparison
  • you want the corrected scenario on the observational grid

Choosing a Method

ClimateTools exposes two main Regridder methods for general workflows:

  • "bilinear" or "linear": smoother, usually preferred for temperature-like continuous fields
  • "nearest_s2d" or "nearest": simpler nearest-neighbor transfer, useful when continuity is less important or for categorical behavior

As a rough rule:

  • use bilinear for temperature and pressure-like continuous variables
  • be more cautious with precipitation and thresholded variables, especially at coastlines or sharp gradients

Reusable Regridder Workflow

Build the regridder once and reuse it across multiple variables or periods.

regridder = Regridder(source_grid_cube, destination_grid_cube; method="bilinear")

out1 = regrid(source_cube_1, regridder)
out2 = regridder(source_cube_2)

This is the preferred workflow when you will apply the same geometry transformation many times.

Persisting a Regridder

save_regridder("regridder.bin", regridder)
regridder = load_regridder("regridder.bin")

Saved regridders validate the incoming source lon-lat grid before applying the weights. The saved format is intended for reuse within the same Julia and ClimateTools environment.

Missing Values During Regridding

skipna and na_thres control how missing values affect the output.

out = regrid(source_cube, regridder; skipna=true, na_thres=0.25)

These keywords matter when the source grid contains masks, coastlines, or large missing regions.

Dataset-Aware Workflow for Rotated or Curvilinear Grids

For datasets with a grid_mapping attribute, pass the dataset rather than a plain variable cube. This lets ClimateTools find the supporting 2D geographic coordinates and projection metadata.

using YAXArrays

ds = open_dataset("climex_file.nc")
dest = YAXArray(
    (Dim{:longitude}(dest_lon), Dim{:latitude}(dest_lat)),
    zeros(length(dest_lon), length(dest_lat)),
)

regridder = Regridder(ds, :pr, dest)
out = regrid(ds[:pr], regridder)

The dataset constructor automatically:

  1. detects grid_mapping in the variable metadata
  2. reads 2D lon and lat arrays when available
  3. derives geographic coordinates from the rotated-pole metadata when needed

This is the correct route for many regional climate-model products.

One-Shot Regridding

For one-off use, regrid_cube remains available.

out = regrid_cube(source_cube, destination_cube)

Or, with a dataset-aware source:

out = regrid_cube(ds, :pr, dest)

Curvilinear and Explicit Array Interfaces

For already-geographic curvilinear grids:

out = regrid_curvilinear_to_regular(loncurv, latcurv, data, londest, latdest; method="linear")

For a 2D YAXArray field on a curvilinear grid:

out = regrid_curvilinear_to_regular(source_cube_2d, destination_grid_cube; method="linear")

For rotated grids with explicit arrays:

out = regrid_rotated_curvilinear_to_regular(lonrot, latrot, data, londest, latdest;
    grid_north_longitude=0.0,
    grid_north_latitude=90.0)

Common Pitfalls

  • Passing a rotated-grid variable cube directly to Regridder(cube, dest) instead of using the dataset-aware constructor
  • Regridding after bias correction instead of before, which can make interpretation harder
  • Ignoring missing-value behavior over coastlines or masked domains
  • Comparing observational and model fields before confirming that the lon-lat conventions are compatible

Role in Scenario Construction

In a climate-scenario workflow, regridding usually comes before bias correction:

  1. choose the target grid
  2. regrid the historical and future simulations
  3. regrid the observation if needed
  4. run the correction method on aligned fields

See Building Climate Scenarios for an end-to-end example.