Threshold a continuous prediction to a categorical layer

It is common in many applications of species distribution modelling that estimated continuous suitability surfaces are converted into discrete representations of where suitable habitat might or might not exist. This so called threshold'ing can be done in various ways which are further described in the details.

In case a SpatRaster is provided as input in this function for obj, it is furthermore necessary to provide a sf object for validation as there is no DistributionModel to read this information from.

Note: This of course also allows to estimate the threshold based on withheld data, for instance those created from an a-priori cross-validation procedure.

For BiodiversityScenario objects, adding this function to the processing pipeline stores a threshold attribute in the created scenario object.

Usage

threshold(
  obj,
  method = "mtp",
  value = NULL,
  point = NULL,
  field_occurrence = "observed",
  format = "binary",
  return_threshold = FALSE,
  ...
)

# S4 method for ANY
threshold(
  obj,
  method = "mtp",
  value = NULL,
  point = NULL,
  field_occurrence = "observed",
  format = "binary",
  return_threshold = FALSE,
  ...
)

# S4 method for SpatRaster
threshold(
  obj,
  method = "fixed",
  value = NULL,
  point = NULL,
  field_occurrence = "observed",
  format = "binary",
  return_threshold = FALSE
)

# S4 method for BiodiversityScenario
threshold(
  obj,
  method = "mtp",
  value = NULL,
  point = NULL,
  field_occurrence = "observed",
  format = "binary",
  return_threshold = FALSE,
  ...
)

Arguments

obj: A BiodiversityScenario object to which an existing threshold is to be added.
method: A specifc method for thresholding. See details for available options.
value: A numeric value specifying the specific threshold for scenarios (Default: NULL Grab from object).
point: A sf object containing observational data used for model training.
field_occurrence: A character location of biodiversity point records.
format: character indication of whether "binary", "normalize" or "percentile" formatted thresholds are to be created (Default: "binary"). Also see Muscatello et al. (2021).
return_threshold: Should threshold value be returned instead (Default: FALSE)
...: Any other parameter. Used to fetch value if set somehow.

Value

A SpatRaster if a SpatRaster object as input. Otherwise the threshold is added to the respective DistributionModel or BiodiversityScenario object.

Details

The following options are currently implemented:

'fixed' = applies a single pre-determined threshold. Requires value to be set.
'mtp' = minimum training presence is used to find and set the lowest predicted suitability for any occurrence point.
'percentile' = For a percentile threshold. A value as parameter has to be set here.
'min.cv' = Threshold the raster so to minimize the coefficient of variation (cv) of the posterior. Uses the lowest tercile of the cv in space. Only feasible with Bayesian engines.
'TSS' = Determines the optimal TSS (True Skill Statistic). Requires the "modEvA" package to be installed.
'kappa' = Determines the optimal kappa value (Kappa). Requires the "modEvA" package to be installed.
'F1score' = Determines the optimal F1score (also known as Sorensen similarity). Requires the "modEvA" package to be installed.
'F1score' = Determines the optimal sensitivity of presence records. Requires the "modEvA" package to be installed.
'Sensitivity' = Determines the optimal sensitivity of presence records. Requires the "modEvA" package to be installed.
'Specificity' = Determines the optimal sensitivity of presence records. Requires the "modEvA" package to be installed.
'AUC' = Determines the optimal AUC of presence records. Requires the "modEvA" package to be installed.

References

Lawson, C.R., Hodgson, J.A., Wilson, R.J., Richards, S.A., 2014. Prevalence, thresholds and the performance of presence-absence models. Methods Ecol. Evol. 5, 54–64. https://doi.org/10.1111/2041-210X.12123
Liu, C., White, M., Newell, G., 2013. Selecting thresholds for the prediction of species occurrence with presence-only data. J. Biogeogr. 40, 778–789. https://doi.org/10.1111/jbi.12058
Muscatello, A., Elith, J., Kujala, H., 2021. How decisions about fitting species distribution models affect conservation outcomes. Conserv. Biol. 35, 1309–1320. https://doi.org/10.1111/cobi.13669

Examples

if (FALSE) {
 # Where mod is an estimated DistributionModel
 tr <- threshold(mod)
 tr$plot_threshold()
}