SpaColExt is an R package for inferring extinction and colonization dates from species sighting records. It provides Bayesian extinction inference through time, spatial helper functions, colonization inference by time reversal, and experimental effort-informed observation models.

The package currently implements methods based on Solow (1993), with experimental non-homogeneous observation-rate functions inspired by Kodikara et al. (2020).

The detailed documentation now lives in the Quarto manual: book/.

Render it from the package root with:

quarto render book

After rendering, open:

book/_book/index.html

The manual includes:

• model interpretation;
• basic extinction and colonization workflows;
• spatial posterior smoothing;
• informative-prior examples;
• simulation diagnostics for constant, increasing, and decreasing observation effort;
• comparison of threshold-based extinction dates with posterior median estimates of extinction time.

You can install the development version from GitHub:

# install.packages("remotes")
remotes::install_github("Clara-Casabona/SpaColExt")

Load the package with:

library(SpaColExt)

The main input is a numeric vector of sighting years. compute_posterior_solow1993() returns the posterior probability that the species is still extant at the end of the study interval.

sightings <- c(1901, 1902, 1903, 1905, 1908, 1910)
start_year <- 1900
end_year <- 1920

compute_posterior_solow1993(
  sightings = sightings,
  start_year = start_year,
  end_year = end_year,
  dprior_m = solowdprior_m,
  dprior_te = solowdprior_te,
  prior = 0.5
)
#> [1] 0.1388889

To estimate an extinction year directly from the posterior distribution of extinction time, use posterior_extinction_year_quantile(). The default quantile is the posterior median.

posterior_extinction_year_quantile(
  sightings = sightings,
  start_year = start_year,
  stop_year = end_year,
  method = "solow",
  probability = 0.5
)
#> [1] 1911.417

Reproducible simulation scripts are stored in:

analysis/simulation_study/

The newest diagnostics compare:

• threshold estimator: first year where posterior persistence drops below 0.5;
• posterior-time estimator: posterior median of extinction time te;
• constant, increasing, and decreasing observation-effort scenarios;
• matched-sightings scenarios that separate effort shape from the number of observations.

The current comparison figure is:

analysis/simulation_study/results/figures/extinction_quantile_estimator_comparison.pdf

• compute_posterior_solow1993(): posterior extant probability for one study interval.
• posterior_probability_extinction_varying_end_year(): posterior extant probabilities for a sequence of end years.
• posterior_extinction_year_quantile(): posterior quantile estimate of extinction year.
• posterior_probability_colonization_varying_year(): posterior colonization probabilities using time reversal.
• spatial_posterior_probability_extinction_varying_end_year(): applies the varying end-year workflow across spatial cells.
• smooth_spatial_posterior(): applies neighbour-informed smoothing to spatial posterior curves.
• compute_posterior_c2022_extinction(): posterior curve with optional informative priors.

Experimental non-homogeneous functions are available, but their interface and numerical behavior may still change.

Solow, A. R. (1993). Inferring extinction from sighting data. Ecology, 74(3), 962-964.

Solow, A. R. and Beet, A. R. (2014). On uncertain sightings and inference about extinction. Conservation Biology, 28(4), 1119-1123.

Kodikara, S., et al. (2020). Non-homogeneous sighting processes for extinction inference. Methods in Ecology and Evolution.