Predicting distributional change, with application to bird distributions in northeast Scotland

Methods for modelling distributional change over time of birds in a heterogeneous environment are illustrated using data from northeast Scotland. Ten species of bird that are currently undergoing changes in their range are selected to demonstrate the methods. Habitat suitability is estimated using bird presence/absence data at one particular time as the dependent variable, and by carrying out a logistic regression in which the covariates are principal component scores obtained from environmental data. It is assumed that some environmental variables are available at the same resolution as the bird distributional data, while others are available at higher resolution. A two step procedure is used, to allow estimation of habitat suitability at the higher resolution. Future distributions are predicted by fitting a spatio-temporal model to successive distributions obtained from atlas studies using maximum likelihood. Under this model, the probability of future site occupation is assumed to depend on habitat suitability of the site and on its effective distance from occupied sites at a previous time point. Effective distance is assumed to depend both on the actual distance between sites and on the suitability of habitat between the sites. Confidence limits for the predicted range of each species and estimated probabilities of extinction are obtained using bootstrap methods.