Modelling the distribution of shorebirds in estuarine areas using generalised additive models

In this study we modelled the occurrence of the shorebirds in their intertidal feeding areas using Generalised Additive Models (GAMs). The data used for the modelling exercise consisted of regular winter counts of shorebirds in one hundred and five 50×50 m plots, arranged in a 1050×250 m area. Several physical characteristics were obtained from each plot and some other variables were derived using a Geographical Information System (GIS). These variables were used to predict the probability of occurrence of shorebirds using GAMs. The three most influential variables in the distribution of the shorebird species assemblage were the distance to channels, type of sediment and area of oyster beds. The smooth response curves of the species occurrence along the gradient of the variables were biologically meaningful, and generally consistent with previous descriptions of the habitat preferences of each species. The quality of the fits was generally very high, as assessed by the significance of the models and by comparing the observed frequencies with the predicted probabilities of occurrence. The performance of GAMs was compared with that of the equivalent GLMs, and we concluded that extended flexibility offered by GAMs resulted in better overall fits. We suggest that GAMs represent a convenient framework for modelling the large-scale distribution of shorebirds in intertidal areas from their physical characteristics.