Alternative methods for predicting species distribution: an illustration with Himalayan river birds

摘要: Summary 1. Current emphasis on species conservation requires the development of specific distribution models. Several modelling methods are available, but their performance has seldom been compared. We therefore used discriminant analysis, logistic regression and artificial neural networks with environmental data to predict presence or absence six river birds along 180 Himalayan streams. applied each method calibration sites independent test sites. With regression, we compared in predicting presence–absence using map-derived predictors (river slope altitude) as opposed detailed from a standardized habitat survey (RHS). 2. Using entire data, overall success at was only slightly greater (89–100%) than either (75–92%) analysis (81–95%), this criterion all gave good performance. 3. When prediction averaged 71–80%, marginally significantly out-performing other methods. Encouragingly for researchers limited model jack-knife tests faithfully represented more rigorous validations where (n = 119) (n = 61) were separate geographical regions. 4. All three predicted true absences (83–92% success) better presences (31–44%). Results most variable across species, positive declined increasing rarity method. 5. Applications illustrated that significant varied between sets within species. Hypotheses about causal effects by structure thus difficult erect test. Logistic also showed substantially improved comparison altitude alone. 6. We conclude differ when distributions. Model choice should depend nature needs any particular whether assumptions satisfied. All share drawbacks due systematic measures. They limitations correlative often spatial scales required macro-ecology biology. Tests wider range measures those traditionally, will be important examining models testing hypotheses such applications.