Obtain Pseudoabsences

This function obtains pseudoabsences given a set of predictors.

Usage

pseudoabsences(occ,
               pred = NULL,
               method = "random",
               n_set = 10,
               n_pa = NULL,
               variables_selected = NULL,
               th = 0,
               size = 1,
               crs = 4326)

n_pseudoabsences(i)

pseudoabsence_method(i)

pseudoabsence_data(i)

Arguments

occ: A occurrences_sdm or input_sdm object.
pred: A sdm_area object. If NULL and occ is a input_sdm, pred will be retrieved from occ.
method: Method to create pseudoabsences. One of: "random", "bioclim", "mahal.dist" or "buffer_sdm". User can also provide a custom function (see details).
n_set: numeric. Number of datasets of pseudoabsence to create.
n_pa: numeric. Number of pseudoabsences to be generated in each dataset created. If NULL then the function prevents imbalance by using the same number of presence records (n_records(occ)). If you want to address different sizes to each species, you must provide a named vector (as in n_records(occ)).
variables_selected: A vector with variables names to be used while building pseudoabsences. Only used when method is not "random".
th: numeric Threshold to be applied in bioclim/mahal.dist projections. See details.
size: numeric The distance between the record and the margin of the buffer (i.e. buffer radius).
crs: numeric Indicates which EPSG it the size in.
i: A input_sdm object.

Value

A occurrences_sdm or input_sdm object with pseudoabsence data.

Details

pseudoabsences is used in the SDM workflow to obtain pseudoabsences, a step necessary for most of the algorithms to run. We implemented three methods so far: "random", which is self-explanatory, "bioclim", "mahal.dist" and "buffer_sdm". The two last are built with the idea that pseudoabsences should be environmentally different from presences. Thus, we implemented two presence-only methods to infer the distribution of the species. "bioclim" uses an envelope approach (bioclimatic envelope), while "mahal.dist" uses a distance approach (mahalanobis distance). th parameter enters here as a threshold to binarize those results. Pseudoabsences are retrieved outside the projected distribution of the species. If user provides a custom function, it must have the arguments env_sf and occ_sf, which will consist of two "sf"s. The first has the predictor values for the whole study area, while the second has the presence records for the species. The function must return a vector with cell_ids of the pseudoabsences.

n_pseudoabsences returns the number of pseudoabsences obtained per species.

pseudoabsence_method returns the method used to obtain pseudoabsences.

pseudoabsence_data returns a list of species names. Each species name will have a lists with pseudoabsences data from class sf.

Author

Luíz Fernando Esser (luizesser@gmail.com) https://luizfesser.wordpress.com

Examples

# Create sdm_area object:
sa <- sdm_area(parana, cell_size = 25000, crs = 6933)
#> ! Making grid over study area is an expensive task. Please, be patient!
#> ℹ Using GDAL to make the grid and resample the variables.

# Include predictors:
sa <- add_predictors(sa, bioc) |> select_predictors(c("bio1", "bio4", "bio12"))
#> ! Making grid over the study area is an expensive task. Please, be patient!
#> ℹ Using GDAL to make the grid and resample the variables.

# Include scenarios:
sa <- add_scenarios(sa)

# Create occurrences:
oc <- occurrences_sdm(occ[1:50,], crs = 6933)

# Create input_sdm:
i <- input_sdm(oc, sa)

# Pseudoabsence generation:
i0 <- pseudoabsences(i, method="random")

# Custom method example:
env_distance_pa <- function(env_sf, occ_sf, n_pa=n_pa) {

  # Transform the environmental data to a data frame and remove the cell_id and geometry columns
  df <- as.data.frame(env_sf)[, -which(names(env_sf) %in% c("cell_id", "geometry"))]

  # Define the center of the environmental space based on the mean of the predictors
  center <- colMeans(df)

  # Calculate the distance of each cell in the environmental space to the center
  d <- sqrt(rowSums((df - center)^2))

  # Convert distances to probabilities (the closer to the center, the higher the probability)
  p <- d / sum(d)

  # Sample pseudo-absences based on the calculated probabilities
  sample(env_sf$cell_id, size = n_pa, prob = p)

}

i1 <- pseudoabsences(i, method=env_distance_pa)