Skip to contents

Tidy species distribution models (presence-background / presence-absence)

sdm.Rd

Fits presence-only (presence-background) or presence-absence models with a tidy interface. Engines: "glm" (logistic or cloglog), "gam" (mgcv when installed), and "maxnet" (native Maxent-like linear + quadratic features with ridge-penalised logistic regression; no Java Maxent, no biomod2 or dismo). Predictors may be columns or a SpatRaster stack from geo_bioclim() / geo_extract().

Usage

sdm_fit(
  data,
  predictors,
  lon = lon,
  lat = lat,
  present = NULL,
  engine = c("glm", "gam", "maxnet"),
  link = c("logit", "cloglog"),
  n_background = 500,
  features = c("linear", "quadratic"),
  lambda = 1e-04,
  seed = NULL
)

sdm_predict(object, newdata = NULL, type = c("response", "link"))

sdm_evaluate(object, newdata = NULL, threshold = NULL)

plot_sdm(x, basemap = "esri", ...)

Arguments

data

Occurrence table with coordinates and optional absence rows (present column 1/0). When absences are missing, background points are sampled from predictors or from the bounding box.

predictors

Character vector of predictor columns in data, or a SpatRaster of predictors.

lon, lat

Coordinate columns.

present

Optional presence/absence column (1/0). When NULL, every row is treated as a presence and background points are added.

engine

"glm", "gam", or "maxnet".

"logit" or "cloglog".

n_background

Number of background points for presence-only data.

features

Feature types for "maxnet": any of "linear", "quadratic".

lambda

Ridge penalty for "maxnet" (larger = more shrinkage).

seed

Optional RNG seed for background sampling.

object

A tmr_sdm from sdm_fit().

newdata

Optional data frame of prediction locations. When NULL and a predictor raster was supplied at fit time, predicts on that raster.

type

"response" (probability) or "link".

threshold

Suitability threshold for TSS / sensitivity. When NULL, the threshold that maximises TSS on the training data is used.

x

A tmr_sdm, a prediction tibble from sdm_predict(), or a suitability SpatRaster.

basemap

Basemap for ggplot maps. See geo_basemap().

...

Passed through to the surface / plot_geo() helpers (for example legend_position, scale_bar, north_arrow, subtitle). Regular prediction grids are drawn as a continuous raster surface; sparse tables stay as points.

Value

An object of class tmr_sdm with elements engine, model, predictors, coefficients (tibble), training (tibble), link, and formula.

Examples

set.seed(1)
# Species present only at high bio1
grid <- expand.grid(lon = seq(145, 145.5, length.out = 12),
                    lat = seq(-5.9, -5.5, length.out = 10))
grid$bio1 <- 15 + 10 * (grid$lat - min(grid$lat)) /
  diff(range(grid$lat))
occ <- dplyr::filter(grid, bio1 > 22) |>
  dplyr::slice_sample(n = 25)
fit <- sdm_fit(occ, predictors = "bio1", engine = "glm")
sdm_evaluate(fit)
#> # A tibble: 1 × 8
#>     auc   tss threshold sensitivity specificity     n n_presence n_absence
#>   <dbl> <dbl>     <dbl>       <dbl>       <dbl> <int>      <int>     <int>
#> 1 0.523 0.268    0.0454           1       0.268   525         25       500