generate_cv_folds: Generate cross-validation folds

Description

Creates cross-validation fold assignments for presence-absence or presence-only data, supporting three types of strategies: k-fold, spatial blocks (through blockCV R package), and temporal blocks.

Usage

generate_cv_folds(
  data,
  method = "k-fold",
  block_method = "predictors_autocorrelation",
  block_size = NULL,
  k = 10,
  predictor_raster = NULL,
  model_residuals = NULL,
  coords = c("decimalLongitude", "decimalLatitude")
)

Value

A list with the following elements:

folds: A vector of fold assignments (one per row in `data`).
method: The CV method used.
block_method: The spatial block size method (if applicable).
block_size: The estimated or manual block size (in meters), if spatial blocks were used.

Arguments

data: A `data.frame` with at least presence-absence data (`pa`), coordinates, and optionally a `timestamp`.
method: The cross-validation strategy. One of: `"k-fold"`, `"spatial_blocks"`, `"temporal_blocks"`.
block_method: For spatial blocks, how to determine block size. One of: `"residuals_autocorrelation"`, `"predictors_autocorrelation"`, `"manual"`.
block_size: Numeric. Manual block size in meters (used if `block_method = "manual"`).
k: Integer. Number of folds to generate.
predictor_raster: A `terra::SpatRaster` used for estimating spatial autocorrelation (only needed if `block_method = "predictors_autocorrelation"`).
model_residuals: A `data.frame` with residuals and coordinates (only needed if `block_method = "residuals_autocorrelation"`).
coords: A character vector of length 2 indicating the longitude and latitude column names.