createEPMgrid {epm}

Create epmGrid object

Description

Creates an epmGrid object from a range of species-specific inputs.

Usage

createEPMgrid(
  spDat,
  resolution = 50000,
  method = "centroid",
  cellType = "hexagon",
  percentThreshold = 0.25,
  retainSmallRanges = TRUE,
  extent = "auto",
  percentWithin = 0,
  dropEmptyCells = TRUE,
  checkValidity = FALSE,
  crs = NULL,
  nThreads = 1,
  template = NULL,
  verbose = FALSE,
  use.data.table = "auto"
)

Arguments

spDat

a number of possible input formats are possible. See details below.

resolution

vertical and horizontal spacing of grid cells, in units of the polygons' or points' projection.

method

approach used for gridding. Either centroid or percentOverlap. See details below.

cellType

either hexagon or square. See details below.

percentThreshold

the percent that a species range must cover a grid cell to be considered present. Specified as a proportion.

retainSmallRanges

boolean; should small ranged species be dropped or preserved. See details.

extent

if 'auto', then the maximal extent of the polygons will be used. If not 'auto', can be a SpatialPolygon, sf object, or raster, in which case the resulting epmGrid will be cropped and masked with respect to the polygon; or a spatial coordinates object, from which an extent object will be generated; or a numeric vector of length 4 with minLong, maxLong, minLat, maxLat. If 'global', a global extent will be specified. See interactiveExtent to draw your own extent.

percentWithin

The percentage of a species range that must be within the defined extent in order for that species to be included. This filter can be used to exclude species whose range barely enters the area of interest. The default value of 0 will disable this filter. If extent == 'auto', then this filter will also have no effect, as the extent is defined by the species' ranges.

dropEmptyCells

only relevant for hexagonal grids, should empty cells be excluded from the resulting grid. Default is TRUE. Reasons to set this to FALSE may be if you want to retain a grid of a certain extent, regardless of which cells contain species.

checkValidity

if TRUE, then check polygon validity and repair if needed, using sf::st_make_valid.

crs

if supplying occurrence records in a non-spatial format, then you must specify the crs. For unprojected long/lat data, you can simply provide crs = 4326.

nThreads

if > 1, then employ parallel computing. This won't necessarily improve runtime.

template

a grid (SpatRaster, RasterLayer or sf) that will be directly used as the reference grid, bypassing any inference from the input data.

verbose

if TRUE, list out all species that are dropped/excluded, rather than counts.

use.data.table

if 'auto', this is determined by the size of the dataset. Primarily intended for debugging.

Details

Types of accepted inputs for argument spDat:

1. a list of polygon objects (sf or sp), named with taxon names.

2. a list of SpatRaster or RasterLayer grids, named with taxon names.

3. a multi-layer RasterStack or multi-layer SpatRaster.

4. a set of occurrence records, multiple accepted formats, see below.

5. a site-by-taxon presence/absence matrix.

If input data consist of occurrence records rather than polygons, then a couple of formats are possible:

1. You can provide a list of species-specific spatial point objects.

2. You can provide a single spatial object, where points have a taxon attribute.

3. You can provide a list of non-spatial species-specific dataframes.

4. You can provide a single non-spatial dataframe.

For options (1) and (3), the taxon names must be provided as the list names. For options (2) and (4), the columns must be 'taxon', 'x' and 'y' (or 'long', 'lat'). For options (3) and (4), as these are non-spatial, you must provide a crs object to the crs argument, so that the function knows what projection to use.

It is also possible to supply a matrix with sites as rows and taxa as columns. The contents of this matrix must be either 0 or 1. If this is the case, then a raster grid must be supplied under the template argument. This will be the grid system used for converting this presence/absence matrix to an epmGrid object. It is expected that the index order of the grid is the same as the row order of the matrix.

If input is a set of species-specific grids, then it is expected that all grids belong to the same overall grid system, i.e. that the cells align and that all grids have the same resolution. Grids do not need to have the same extent.

Any SpatialPolygon or SpatialPoints objects are converted to objects of class sf.

If cellType = 'hexagon', then the grid is made of polygons via the sf package. If cellType = 'square', then the grid is a raster generated via the terra package. Hexagonal cells have several advantages, including being able to be of different sizes (if the grid is in unprojected long/lat), and may be able to more naturally follow coastlines and non-linear features. However, the raster-based square cells will be much less memory intensive for high resolution datasets. Choice of grid type matters more for spatial resolution (total number of cells), than for number of species.

In the polygon-to-grid conversion process, two approaches are implemented. For method = 'centroid', a range polygon registers in a cell if the polygon overlaps with the cell centroid. For method = 'percentOverlap', a range polygon registers in a cell if it covers that cell by at least percentThreshold fraction of the cell.

If retainSmallRanges = FALSE, then species whose ranges are so small that no cell registers as present will be dropped. If retainSmallRanges = TRUE, then the cell that contains the majority of the the small polygon will be considered as present, even if it's a small percent of the cell.

If retainSmallRanges = TRUE, and an extent is provided, then species may still be dropped if they fall outside of that extent.

You may see the message Failed to compute min/max, no valid pixels found in sampling. (GDAL error 1) . This just means that a species did not register in any grid cells. If you specified retainSmallRanges = TRUE, then those species will be included in a subsequent step. Therefore, this message can be ignored.

For very large datasets, this function will make a determination as to whether or not there is sufficient memory. If there is not, an alternative approach that uses the data.table package will be employed. Please install this R package to take advantage of this feature.

Value

an object of class epmGrid.

Author(s)

Pascal Title

Examples

library(sf)
# example dataset: a list of 24 chipmunk distributions as polygons
head(tamiasPolyList)

# hexagonal grid

tamiasEPM <- createEPMgrid(tamiasPolyList, resolution = 50000,
	cellType = 'hexagon', method = 'centroid')
tamiasEPM

# square grid
tamiasEPM2 <- createEPMgrid(tamiasPolyList, resolution = 50000,
	cellType = 'square', method = 'centroid')
tamiasEPM2

# use of a grid from one analysis for another analysis

tamiasEPM <- createEPMgrid(tamiasPolyList, resolution = 50000,
	cellType = 'hexagon', method = 'centroid')

tamiasEPM <- createEPMgrid(tamiasPolyList, resolution = 50000,
	cellType = 'hexagon', method = 'centroid', template = tamiasEPM[[1]])
	
#######

# demonstration of site-by-species matrix as input.
tamiasEPM2 <- createEPMgrid(tamiasPolyList, resolution = 50000,
	cellType = 'square', method = 'centroid')

## first we will use the function generateOccurrenceMatrix() to get
## a presence/absence matrix
pamat <- generateOccurrenceMatrix(tamiasEPM2, sites = 'all')

# here, our grid template will be tamiasEPM2[[1]]
tamiasEPM2[[1]]
xx <- createEPMgrid(pamat, template = tamiasEPM2[[1]])


#######
# demonstration with grids as inputs
## We will first generate grids from the range polygons
## (you normally would not do this -- you would have grids from some other source)

# define the extent that contains all range polygons
fullExtent <- terra::ext(terra::vect(tamiasPolyList[[1]]))
for (i in 2:length(tamiasPolyList)) {
	fullExtent <- terra::union(fullExtent, terra::ext(terra::vect(tamiasPolyList[[i]])))
}

# create raster template
fullGrid <- terra::rast(fullExtent, res = 50000, crs = terra::crs(terra::vect(tamiasPolyList[[1]])))

# now we can convert polygons to a common grid system
spGrids <- list()
for (i in 1:length(tamiasPolyList)) {
	spGrids[[i]] <- terra::trim(terra::rasterize(terra::vect(tamiasPolyList[[i]]), fullGrid))
}
names(spGrids) <- names(tamiasPolyList)

createEPMgrid(spGrids)


#######
# With point occurrences
## demonstrating all possible input formats

# list of sf spatial objects
spOccList <- lapply(tamiasPolyList, function(x) st_sample(x, size = 10, type= 'random'))
tamiasEPM <- createEPMgrid(spOccList, resolution = 100000, cellType = 'hexagon')

# list of coordinate tables
spOccList2 <- lapply(spOccList, function(x) st_coordinates(x))
tamiasEPM <- createEPMgrid(spOccList2, resolution = 100000, cellType = 'square', 
	crs = st_crs(tamiasPolyList[[1]]))

# single table of coordinates
spOccList3 <- spOccList2
for (i in 1:length(spOccList3)) {
	spOccList3[[i]] <- cbind.data.frame(taxon = names(spOccList3)[i], spOccList3[[i]])
	colnames(spOccList3[[i]]) <- c('taxon', 'X', 'Y')
}
spOccList3 <- do.call(rbind, spOccList3)
rownames(spOccList3) <- NULL
spOccList3[, "taxon"] <- as.character(spOccList3[, "taxon"])
tamiasEPM <- createEPMgrid(spOccList3, resolution = 100000, cellType = 'square', 
	crs = st_crs(tamiasPolyList[[1]]))

# a single labeled spatial object
spOccList4 <- st_as_sf(spOccList3[, c("taxon", "X", "Y")], coords = c("X","Y"),
crs = st_crs(spOccList[[1]]))
tamiasEPM <- createEPMgrid(spOccList4, resolution = 100000, cellType = 'square')