Skip to contents

Create a quick SDM using elastic net regression

elasticSDM.Rd

This function quickly creates a SDM using elastic net regression, and it will properly format all data for downstream use with the safeHavens workflow. Note that elastic net models are used for a couple very important reasons: they rescale all input independent variables before modelling, allowing us to combine the raw data with the beta coefficients to use for clustering algorithms downstream. They also allowing for 'shrinking' of terms from models by shrinking terms from models we are able to get levels of ecological inference prohibited by older model selection frameworks.

Usage

elasticSDM(x, predictors, planar_projection, domain, quantile_v)

Arguments

x

A (simple feature) sf data set of occurrence data for the species.

predictors

a terra 'rasterstack' of variables to serve as indepedent predictors.

planar_projection

Numeric, or character vector. An EPSG code, or a proj4 string, for a planar coordinate projection, in meters, for use with the function. For species with very narrow ranges a UTM zone may be best (e.g. 32611 for WGS84 zone 11 north, or 29611 for NAD83 zone 11 north). Otherwise a continental scale projection like 5070 See https://projectionwizard.org/ for more information on CRS. The value is simply passed to sf::st_transform if you need to experiment.

domain

Numeric, how many times larger to make the entire domain of analysis than a simple bounding box around the occurrence data in x.

quantile_v

Numeric, this variable is used in thinning the input data, e.g. quantile = 0.05 will remove records within the lowest 5% of distance to each other iteratively, until all remaining records are further apart than this distance from each other. If you want essentially no thinning to happen just supply 0.01. Defaults to 0.025.

Value

A list of 12 objects, each of these subsequently used in the downstream SDM Post processing sequence, or which we think are best written to disk. The actual model prediction on a raster surface are present in the first list 'RasterPredictions', the indepedent variables used in the final model are present in 'Predictors, and just the global PCNM/MEM raster surfaces are in 'PCNM'. The fit model is in 'Model', while the cross validation folds are stored in 'CVStructure', results from a single test/train partition in 'ConfusionMatrix', and the two data split in 'TrainData' and 'TestData' finally the 'PredictMatrix' which was used for classifying the test data for the confusion matrix.

Examples

if (FALSE) { # \dontrun{

 x <- read.csv(file.path(system.file(package="dismo"), 'ex', 'bradypus.csv'))
 x <- x[,c('lon', 'lat')]
 x <- sf::st_as_sf(x, coords = c('lon', 'lat'), crs = 4326)

 files <- list.files(
   path = file.path(system.file(package="dismo"), 'ex'), 
   pattern = 'grd',  full.names=TRUE )
 predictors <- terra::rast(files)

sdModel <- elasticSDM(
   x = x, predictors = predictors, quantile_v = 0.025,
   planar_projection =
     '+proj=laea +lon_0=-421.171875 +lat_0=-16.8672134 +datum=WGS84 +units=m +no_defs')
     
 terra::plot(sdModel$RasterPredictions)
} # }