Resample Raster

Draft: written, not yet checked by a person Operation · resample-raster

The same raster on a grid with a different cell size, in the same CRS.

Pitfall. Nearest neighbour on continuous data keeps one cell's value per new cell; average gives the mean of the cells it covers.

Pitfall. Interpolating categorical data (land cover codes) with bilinear or cubic makes codes that do not exist.

About this operation
Also called	Resample, change cell size, coarsen, aggregate cells, resize raster
Group	`raster/projection`
Checks	`raster.resampling, raster.nodata`
Pitfalls	2 listed above
Columbus tool	Not stated
Notes	Not stated
Status	active
Broader	Not stated
Narrower	Not stated
Related	`reproject-raster`

Parameters

Name	Type	Required	Default	Values	Note
`resolution`	number	yes	Not stated	Not stated	New cell size, in the raster's CRS units.
`resampling`	enum	no	`nearest`	`nearest, bilinear, cubic, average, mode`	nearest or mode for categories; average to coarsen continuous data.

Inputs and outputs

Direction	Name	Type	Geometry	Note
Input	`raster`	raster	Not stated	Not stated
Output	`output`	raster	Not stated	Not stated

In each engine

Engine	Tool	Parameters	Differences
`gdal`	`gdal raster reproject`	`resolution` → `resolution` : x and y: --resolution 20,20. `resampling` → `resampling` `raster` → `input` `output` → `output`	gdal raster resize does the same without reprojection support; without output-crs this keeps the raster's CRS.
`gdal`	`gdal raster resize`	`resolution` → `resolution` : x and y: --resolution 20,20. `resampling` → `resampling` `raster` → `input` `output` → `output`	None noted
`grass`	`r.resamp.stats`	`resolution` Set by the computational region (g.region res=...). `resampling` → `method` : average = average; mode = mode. not supported: nearest, bilinear, cubic `raster` → `input` `output` → `output`	For coarsening; r.resamp.interp interpolates and r.resample takes nearest neighbour.
`grass`	`r.resample`	`resolution` Set by the computational region (g.region res=...). `resampling` always nearest `raster` → `input` `output` → `output`	None noted
`postgis-raster`	`st_resample`	`resolution` scalex and scaley (scaley negative for a north-up raster). `resampling` → `algorithm` : nearest = NearestNeighbor; bilinear = Bilinear; cubic = Cubic. not supported: average, mode `raster` → `rast`	None noted
`qgis`	`gdal:warpreproject`	`resolution` → `TARGET_RESOLUTION` `resampling` → `RESAMPLING` : nearest = 0; bilinear = 1; cubic = 2; average = 5; mode = 6 `raster` → `INPUT` `output` → `OUTPUT`	The same tool reprojects; leaving TARGET_CRS empty keeps the raster's CRS.
`terra`	`terra::aggregate`	`resolution` fact: how many cells to combine in each direction (2 doubles the cell size). `resampling` → `fun` : average = mean; mode = modal. not supported: nearest, bilinear, cubic `raster` → `x`	Only coarsens, by whole factors; disagg makes cells smaller. On a SpatVector, aggregate dissolves instead.
`terra`	`terra::resample`	`resolution` y is a raster on the target grid; make one with rast(x) and res(r) <- value. `resampling` → `method` : nearest = near; bilinear = bilinear; cubic = cubic; average = average; mode = mode `raster` → `x`	None noted
`whiteboxtools`	`Resample`	`resolution` → `cell_size` `resampling` → `method` : nearest = nn; bilinear = bilinear; cubic = cc. not supported: average, mode `raster` → `inputs` `output` → `output`	Takes a list of inputs, which it also mosaics; base gives a raster whose grid to match.
`xarray-spatial`	`xrspatial.resample`	`resolution` → `target_resolution` `resampling` → `method` `raster` → `agg`	scale_factor resizes by a factor instead of to a cell size.