Modelling the distribution of wadersin the Westerschelde : what is the predictive power of abiotic variables?

Describes various models that predict the distribution of birds, mainly waders, feeding on the intertidal flats of the Westerschelde, SW Netherlands, during low tide. Abiotic predictor variables used in the models are e.g. emersion time, current velocity, silt content and salinity. Feeding densities of birds were determined from seven counts covering an entire low water period in 71 study plots at four locations in 2003 and 2004. Abiotic predictor variables were either measured in the study plots or derived from GIS-maps. Generalized linear modeling (GLM) was employed to fit models with total number of foraging hours of a particular bird species as the dependent variable. The predictive power, as judged from the explained deviance, was high for many of the models, but they performed poorly when the predictions were validated with low tide counts in other areas and high tide counts covering the entire Westerschelde. This mismatch was partly due to the fact that the models often predicted very high bird densities for abiotic conditions that were relatively rare in the Westerschelde and did not occur in the study plots. It was also investigated whether adding information on actual macrozoobenthos densities could improve the predictive power of the models. Generally, this was not the case, but the distribution of Oystercatchers showed a good correlation with the Cockle stocks.