Project a fit onto rasters

Predicts one target per cell of a raster, each carrying the record of its own cell, and returns the predictions as a raster with one layer per response. This is BIOMOD_Projection() and BIOMOD_EnsembleForecasting(): a map is the fit applied to one target per cell, at the grain the fit reads.

Usage

project(
  fit,
  series = NULL,
  static = NULL,
  candidate = "ensemble",
  type = c("response", "binary", "spread"),
  chunk = 5000L,
  ...
)

Arguments

fit

A timesift() result.

series

A terra::SpatRaster whose layers are instants, for a fit made on a record; for a record of several variables, a list of them named as the fit's x columns. Every raster shares one geometry and one set of instants.

static

A SpatRaster whose layers are named as the fit's static columns.

candidate, type, ...

As for predict.timesift(): the candidate (the ensemble by default), "response", "binary" or "spread", and the arguments they take.

chunk

Number of cells predicted at once.

Value

A SpatRaster of the geometry of the inputs, one layer per response. Under type = "spread" one layer per response and statistic, named response.statistic.

Details

The record of a cell is its values through the layers of series, which are placed in time by terra::time(), and the static predictors of a cell are its values in the layers of static, named as the columns of targets were when the fit was made. A fit made with coords reads the centre of each cell. Cells are predicted in chunks, so a raster larger than memory is read a chunk of cells at a time. A cell is predicted where every input holds a value at that cell; a cell with a missing reading anywhere is NA in every layer.

A map for a later period is the same call with the later record, and the binary maps of the two are what range_change() compares.

Examples

if (requireNamespace("terra", quietly = TRUE)) {
  set.seed(1)
  r <- terra::rast(nrows = 6, ncols = 5, xmin = 0, xmax = 5, ymin = 0, ymax = 6)
  elev <- terra::setValues(r, rnorm(30))
  slope <- terra::setValues(r, rnorm(30))
  names(elev) <- "elev"
  names(slope) <- "slope"
  static <- c(elev, slope)
  cells <- data.frame(cell = seq_len(30), elev = elev[][, 1], slope = slope[][, 1])
  cells$sp1 <- rbinom(30, 1, plogis(cells$elev))
  cells$sp2 <- rbinom(30, 1, plogis(-cells$slope))
  fit <- timesift(cells, y = c(sp1, sp2), id = cell, static = c(elev, slope),
                  learners = elasticnet(), ensemble = FALSE, n_inner = NULL,
                  resampling = cv(v = 3), verbose = FALSE)
  project(fit, static = static, candidate = "elasticnet / static")
}