Skip to contents

Lazily resample/reproject onto a target grid.

align_to.Rd

Inserts an explicit warp step, executed as a GDAL VRT warp. Alignment stays explicit: binary ops never auto-resample.

Usage

align_to(x, to, resampling = "near")

Arguments

x

A LazyRaster.

to

Target grid: a GridSpec or another LazyRaster.

resampling

GDAL resampling method. The default, "near", copies source values unchanged, which categorical data (land cover, classes) and QA bitmasks need: interpolating them invents classes and bit patterns that do not exist. It matches the default of lazy_source() and lazy_dataset(). For continuous data (reflectance, elevation, temperature) choose "bilinear" (or "cubic") when resampling to a similar or finer resolution, and "average" when aggregating to a coarser one.

Value

A LazyRaster on the target grid.

Details

Paste fast path: when x is already exactly on the target grid (same CRS, transform, extent and dims; grid_equal()), align_to() is a no-op returning x: reads stay plain windowed reads, with no warp barrier splitting the plan. This is the single-CRS-zone workflow: pin the analysis grid to the sources' native grid and nothing warps. Only EXACT equality pastes (unlike odc-stac's tolerance-based ttol): a sub-pixel-shifted paste would silently move every pixel by up to half a cell, so near-misses warp.

Examples

f <- system.file("extdata", "garry-example.tif", package = "garry")
red <- lazy_source(f, band = 1L)
coarse <- grid_spec(crs = grid_crs(red), extent = grid_bbox(red), dims = c(30L, 20L))
dim(collect(align_to(red, coarse, resampling = "average")))
#> [1] 20 30