Species Distribution Modelling

Summary

Statistical models relate occurrence records to environmental variables to predict where species can live.

In depth

Algorithms such as MaxEnt and random forests relate presence records to climate and habitat variables. Predictions guide reserve planning and invasive species risk assessment, and project future ranges under climate change.

Presence-only data need background points; evaluation uses AUC and independent validation. Models show correlation, not causal niche limits. Projections to new climates assume niche conservatism.

Example: models forecast invasion risk of the Asian tiger mosquito.

SDMs predict suitability from environmental layers such as temperature, rainfall and land cover, using presence-only (MaxEnt), presence-absence (GLMs, boosted regression trees, random forests) or mechanistic approaches. Evaluation uses AUC, the true skill statistic and independent test data. Pitfalls include sampling bias in occurrence records, spatial autocorrelation, missing biotic interactions and poor transferability to new climates.

Uses include forecasting range shifts under climate scenarios, finding places for surveying rare species, predicting invasions (cane toads, lionfish), identifying disease risk (mosquito vectors) and planning protected areas with software such as Marxan. Hybrid models add demography and dispersal.

Key idea: models describe where conditions look suitable and not where a species will actually be found.