Skip to contents

Center and scale a GRaster, or the opposite

Source: R/scale.r
scale.Rd

scale() and scalepop() center and scale layers in a GRaster by subtracting from each raster its mean value (centering), then dividing by its standard deviation (scaling). This is useful for using the raster in a linear model, for example, because unscaled predictors can lead to numerical instability. The scale() function uses the sample standard deviation, and the scalepop() function uses the population standard deviation. For even moderately-sized rasters, the difference between these two is negligible, but the scalepop() function can be much faster than the scale() function.

The unscale() function does the opposite of scale() and scalepop(): it multiples each layer by a value (presumably, its standard deviation), and adds another value (presumably, its mean).

Usage

# S4 method for class 'GRaster'
scale(x, center = TRUE, scale = TRUE)

# S4 method for class 'GRaster'
scalepop(x, center = TRUE, scale = TRUE)

# S4 method for class 'GRaster'
unscale(x, center = NULL, scale = NULL)

Arguments

x

A GRaster.

center

Value depends on the function:

  • scale(): Logical: If TRUE (default), subtract from each raster layer its mean.

  • unscale(): Numeric vector or NULL (default): This can be a single value, which will be recycled if there is more than one layer in the raster, or one value per raster layer. If a value is NA, then no un-centering will be performed on the relevant raster layer. If NULL, then no un-centering is done.

scale

Value depends on the function:

  • scale(): Logical: If TRUE (default), divide each layer by its standard deviation.

  • unscale(): Numeric vector or NULL (default): This can be a single value, which will be recycled if there is more than one layer in the raster, or one value per raster layer. If a value is NA, then no unscaling will be done on the relevant raster layer. If NULL, then no un-scaling is done.

Value

All functions return a GRaster. The output of scale() and scalepop() will have two attributes, "center" and "scale", which have the means and standard deviations of the original rasters (if center and scale are TRUE, otherwise, they will be NA). These can be obtained using attributes(output_raster)$center and attributes(output_raster)$scale.

Examples

if (grassStarted()) {

# Setup
library(terra)

# Climate rasters:
madChelsa <- fastData("madChelsa")

# Convert to GRasters:
chelsa <- fast(madChelsa)

### Center and scale rasters
# Scale with using sample SD:
chScaled <- scale(chelsa)
chScaled

# Scale with using population SD:
chScaledPop <- scalepop(chelsa)
chScaledPop

# Means are very close to 0 and SDs to 1:
global(chScaled, c("mean", "sd", "min", "max"))
global(chScaledPop, c("mean", "sd", "min", "max"))

# Get original means and sd's:
centers <- attributes(chScaled)$center
scales <- attributes(chScaled)$scale
centers
scales

### Unscale rasters:
chUnscaled <- unscale(chScaled, center = centers, scale = scales)

# Means and SD are returned to original values:
global(chUnscaled, c("mean", "sd", "min", "max")) # unscaled
global(chelsa, c("mean", "sd", "min", "max")) # original

}