Skip to contents

Cell level regression

regress.Rd

Run a regression model for each cell of a SpatRaster. The independent variable can be defined either by:

  • a numeric vector of length nlyr(y) – one covariate that varies across the layers but is constant across cells (e.g. time);

  • a SpatRaster of the same number of layers as y – one covariate that varies both across layers and across cells;

  • a data.frame with one row per layer of y and one column per predictor – multiple covariates that vary across layers but are constant across cells (e.g. an experimental design like NPK fertilizer levels).

Usage

# S4 method for class 'SpatRaster,numeric'
regress(y, x, formula=y~x, na.rm=FALSE, cores=1, filename="", overwrite=FALSE, ...)

# S4 method for class 'SpatRaster,SpatRaster'
regress(y, x, formula=y~x, na.rm=FALSE, cores=1, filename="", overwrite=FALSE, ...)

# S4 method for class 'SpatRaster,data.frame'
regress(y, x, formula=NULL, na.rm=FALSE, cores=1, filename="", overwrite=FALSE, ...)

Arguments

y

SpatRaster. The dependent variable. Each layer is one observation per cell.

x

The independent variable(s):

  • SpatRaster of the same number of layers as y;

  • numeric vector of length nlyr(y);

  • data.frame with nlyr(y) rows and one column per predictor (no NAs allowed).

formula

regression formula. For the numeric and SpatRaster methods the default is y ~ x; you can add additional terms such as I(x^2). For the data.frame method the default (when formula=NULL) is y ~ ., expanding to all columns of x; you can also pass a custom formula whose right-hand side references the column names of x (the left-hand side is ignored).

na.rm

logical. If TRUE, layers with NA response in a given cell are dropped before fitting. A cell whose remaining row count drops below the number of regression coefficients yields all NAs.

cores

positive integer. If cores > 1, a 'parallel' package cluster with that many cores is created and used. You can also supply a cluster object.

filename

character. Output filename

overwrite

logical. If TRUE, filename is overwritten

...

list with named options for writing files as in writeRaster

Value

SpatRaster. One layer per regression coefficient, named after the corresponding column of model.matrix(formula, x) (e.g. (Intercept), the names of the predictors, and any interaction or polynomial terms).

Examples

s <- rast(system.file("ex/logo.tif", package="terra"))
x <- regress(s, 1:nlyr(s))

# data.frame method: a small NPK fertilizer experiment.
# Suppose `y` has one layer per (N, P, K) treatment.
if (FALSE) { # \dontrun{
NPK <- data.frame(
    N = c(0,  50, 100,   0,  50, 100,   0,  50, 100),
    P = c(0,   0,   0,  25,  25,  25,  50,  50,  50),
    K = c(0,  10,  20,   0,  10,  20,   0,  10,  20)
)
# Default formula y ~ N + P + K:
fit <- regress(y, NPK)

# Custom formula with interactions and a quadratic term:
fit2 <- regress(y, NPK, formula = yield ~ N + P + K + I(N^2) + N:P)
} # }