Land-cover fractions on a grid from the NRCan 2020 land cover

Downloads the Natural Resources Canada 2020 Land Cover of Canada (30 m, Cloud-Optimized GeoTIFF, Open Government Licence - Canada) once, and averages it onto grid as one fraction layer per land-cover class of wet_chapman_table3(): the share of each cell covered by the NRCan codes that map to that class.

Usage

wet_landcover_nrcan(
  grid,
  table = wet_chapman_table3(),
  dir = "data/landcover",
  overwrite = FALSE,
  timeout = 7200,
  fact = 30
)

Arguments

grid: Target grid: a SpatRaster or a path to one (EPSG:4326 here).
table: Class table with class and nrcan_2020_codes (;-separated).
dir: Cache directory for the source file and the fractions.
overwrite: Logical. Rebuild the fractions even when cached.
timeout: Seconds allowed for the download (about 2.1 GB).
fact: Subdivision of each grid cell for the nearest-neighbour step; 30 makes a 30 arc-second grid 1 arc-second, about the source's 30 m.

Value

Path to a GeoTIFF with one fraction layer (0-1) per class.

Details

It is done in two steps. First the codes are resampled (nearest) onto a grid fact times finer than grid and aligned with it, in grid's CRS. Then a VRT carries one band per class, each a lookup table that turns the class's codes into 1 and every other code into 0, and terra::aggregate() takes their block means onto grid. A single average warp from the source CRS is not used: GDAL takes each target cell's footprint as an axis-aligned rectangle in the source CRS, and NRCan's Lambert conformal grid is rotated about 20 degrees against a lon/lat cell in western BC, which put single cells off by up to 0.3 (measured 2026-09-27; the two-step fractions match exact-overlap counts to a mean 0.002). Ground outside Canada (the product's no-data 0) and codes no class claims count as uncovered, so the fractions of a cell sum to less than 1 there and wet_aet_landcover() leaves that part of the cell unadjusted.

NRCan publishes no checksum for the file (its ETag is a multipart hash), so the download is written to .part and renamed only on HTTP 200, its md5 belongs in the caller's input manifest, and the fractions are cached under a key that includes the source file's md5 (hashing 2.1 GB takes a few seconds on each call). The option wet.nrcan_landcover_url only sets where a missing source file is fetched from; delete the cached file to fetch another.

Examples

if (FALSE) { # interactive()
# about 2.1 GB on first use
g <- terra::rast(xmin = -120, xmax = -119, ymin = 49, ymax = 50, res = 1 / 120)
fr <- terra::rast(wet_landcover_nrcan(g))
terra::global(fr, "mean")
}