fragmentation,SpatRaster-method: Landscape fragmentation class following Riitters et al. (2020)

Description

Riitters et al. (2020) propose a classification scheme for forest fragmentation (which can be applied to any habitat type). The scheme relies on calculating density (e.g., number of forested cells in a window around a focal cell) and connectivity (number of cases where neighboring cells are both forested). This function calculates these classes from a GRaster or SpatRaster in which the focal habitat type has cell values of 1, and non-focal habitat type has cell values of 0 or NA.

Note that by default, the SpatRaster and GRaster versions will create different results around the border of the raster. The SpatRaster version uses the terra::focal() function, which will not return an NA value when its window overlaps the raster border if the na.rm argument is TRUE. However, the GRaster version uses the GRASS module r.neighbors, which does return NA values in these cases.

The fragmentation classes are:

Value provided by none: None (i.e., no forest; default is NA).
1: Patch
2: Transitional
3: Perforated
4: Edge
5: Undetermined (not possible to obtain when w = 3)
6: Interior

Usage

# S4 method for SpatRaster
fragmentation(
  x,
  w = 3,
  undet = "undetermined",
  none = NA,
  na.rm = TRUE,
  cores = faster("cores"),
  verbose = TRUE
)
# S4 method for GRaster
fragmentation(x, w = 3, undet = "undetermined", none = NA, verbose = TRUE)

Value

A categorical SpatRaster or GRaster. The values assigned to each class can be seen with levels().

Arguments

x: A SpatRaster or GRaster.
w: An odd, positive integer: Size of the window across which fragmentation is calculated (in units of "rows" and "columns"). The default is 3, meaning the function uses a 3x3 moving window to calculate fragmentation. For large rasters, compute time is ~O(N) + O(N * w^2), where N is the number of cells in the raster. So, even a small increase in w can increase compute time by a lot.
undet: Character: How to assign the "undetermined" case. Valid values are "perforated" (default), "edge", and "undetermined". Partial matching is used. If Pf is the proportional density raster cell value and Pff the proportional connectivity raster cell value, the undetermined case occurs when Pf > 0.6 and Pf == Pff.
none: Integer or NA (default): Value to assign to a cell with no focal habitat. Riitters et al. use NA. This will be forced to an integer if it is not an actual integer.
na.rm: Logical: If TRUE (default) and x is a SpatRaster, then cells near the edge of the raster where the window overlaps the edge can still be assigned a fragmentation class. If FALSE, these cells will be assigned a value of none.
cores: Integer: Number of processor cores to use for when processing a SpatRaster.
verbose: Logical: If TRUE (default), display progress.

References

Riitters, K., J. Wickham, R. O'Neill, B. Jones, and E. Smith. 2000. Global-scale patterns of forest fragmentation. Conservation Ecology 4:3. URL: http://www.consecol.org/vol4/iss2/art3/. Also note the errata.

Examples

Run this code

if (grassStarted()) {

# Setup
library(terra)

# Example data:
madForest <- fastData("madForest2000") # raster

### Fragmentation classes from a SpatRaster

fragTerra <- fragmentation(madForest)
plot(fragTerra)
levels(fragTerra)
freq(fragTerra)

### Fragmentation classes from a GRaster

# Convert to GRaster:
forest <- fast(madForest)

# Fragmentation class:
frag <- fragmentation(forest)
plot(frag)
levels(frag)
freq(frag)

}