Species Distribution Models
Definition
Species Distribution Models (SDMs) estimate the probability that a species occurs given environmental conditions and, when available, true absence or effort-corrected presence data. Methods range from regression and machine learning (GLM, GAM, MaxEnt, random forests, boosted trees) to joint and hierarchical models that account for detection bias and spatial dependence. Predictors include climate, terrain, soils, land cover, and biotic interactions. Careful partitioning avoids spatial leakage between train and test data.
Application
Conservation uses SDMs to prioritize reserves, forecast climate-driven range shifts, and plan corridors. Biosecurity screens invasive risk; wind and solar developers assess wildlife sensitivity; fisheries model suitable habitat. Citizen science platforms supply presence data that, when filtered for effort and bias, greatly expand coverage.
FAQ
How do you correct for sampling bias in presence-only data?
Use target-group background or bias files that mirror observer effort, include accessibility covariates, and thin clustered records to a consistent spatial spacing.
What is the advantage of hierarchical occupancy models?
They separate ecological occupancy from imperfect detection, using repeated surveys or auxiliary data to estimate detection probability, improving inference.
How should future climate projections be used responsibly?
Project within the envelope of training conditions when possible, report extrapolation areas, and use ensembles of climate models and scenarios to convey uncertainty.
Why are mechanistic and correlative SDMs sometimes combined?
Mechanistic models encode physiology and dispersal; correlative models capture realized niches. Hybrid approaches leverage both to improve transferability across regions or future conditions.