THE ABUNDANCE OF RIVER CORRIDOR BIRDS IN RELATION TO THEIR HABITATS - A MODELING APPROACH

1. Abundance of territories on 837 500-m sections of river corridor distributed throughout England and Scotland, collected as part of the British Trust for Ornithology Waterways Bird Survey, were analysed in relation to measured habitat characteristics at these sites. 2. The variation in habitat types and the abundance of birds across the range of river sections was analysed using detrended correspondence analysis. The first axis of both ordinations was similar, showing a major trend from upland fast flowing to lowland slow flowing rivers in both cases. 3. The first axis of the habitat characteristics ordination was used as a measure of habitat 'type' as an independent variable and the abundance of territories of each species used as dependent variables in series of log-linear regression models with a restricted data set of English and Welsh rivers. Twenty-three species showed significant relationships with water quality, 20 to the first DCA habitat variable, 10 to the second DCA habitat variable, 22 to Northing, 21 to Easting and 17 to river altitude. 4. These single variable models were extended to joint variable models using a parsimonious approach to variable selection. In analyses of deviance, water quality followed by the first DCA habitat variable, were the most important factors affecting the abundance of bird territories. For 13 species where water quality was included as a significant factor in the model, the parameter estimates were negative, indicating that rivers with poor water quality had fewer territories, with redshank, grey and pied wagtail being the species most sensitive to water quality. 5. The numbers of territories of all species predicted from the above models, based on measured habitat and environmental characteristics, were compared with observed numbers determined by field survey of three river systems not included in the original analysis. Predicted territories of mallard, moorhen, coot and reed bunting matched observed numbers reasonably. In the case of shelduck, lapwing, oystercatcher and whitethroat, more were observed than predicted, whereas for little grebe, pied wagtail and sedge warbler, more were predicted than observed. Those species for which territory data were overdispersed were those for which model predictions were worst, relating to the poor fit assumed by the Poisson error term in the model. 6. The value of an integrated modelling approach to policy formulation and river management prescriptions using these results is discussed.