MODELING BIRD DISTRIBUTIONS - A COMBINED GIS AND BAYESIAN RULE-BASED APPROACH

Models to predict the breeding distribution of three species of birds in north-east England are described. The models use readily available data from the ornithological literature on the habitat preferences and life-history characteristics of the birds, together with satellite (l and cover) and physiographic data. These data are linked via Bayesian decision-rules, and model predictions calculated at the landscape scal e using a raster-based Geographic Information System. Log-linear regre ssions of the predicted suitability of the landscape for the birds wit h observed sets of nest records were statistically significant for all three species. The robustness of the models to the effects of noninde pendence of predictor (habitat) variables on Bayesian predictions was investigated using a perturbation method, which gave minor improvement s to the accuracy of the predictions. The value of this modelling appr oach as a method of utilising published autoecological data to predict the landscape distribution of birds is discussed.