Mapping the abundance of birds in Cheshire and Wirral
Maps of abundance were produced using the densities estimated for surveyed 1-km squares, derived as a by-product of the population estimates work above, and using a statistical approach known as simple co-kriging. Simple co-kriging is one of a number of geostatistical methods that model the statistical relationship between surveyed sites. This approach weights the surrounding counts at surveyed sites to derive a prediction for unsurveyed locations. In these, the weights are based on the distance between measured sites and the prediction location, but also on the overall spatial arrangement in the weights (the spatial autocorrelation). For a full discussion of geostatistics and geostatistical methods see Chilès & Delfiner (1999). These techniques allow us to produce a prediction surface over the entire area of interest, in this case Cheshire and Wirral.
Because the addition of habitat in the modelling process may improve the resulting predictions of abundance, we first carried out some preliminary analyses to examine which habitat or habitats best predicted abundance of each bird species. For this we used the bird densities and 8 aggregate CEH landcover 2000 classes (Haines-Young et al. 2002: inland water, moorland / heath and bog, broad-leaved woodland, coniferous woodland, improved grassland, semi-natural grassland, arable and human habitats) to identify the best habitat predictors for each species. These analyses were performed for each species using the GENMOD procedure in SAS (SAS Institute 2001), specifying poisson errors and using a logit link function. The dependent variable used in each model was the density estimates and aggregate CEH landcover data habitat classes as categorical independent variables. This allowed us to identify, which habitats were the best predictors of bird densities, and for each species up to three habitats were included in the subsequent co-kriging models. The Geostatistical analyses were implemented with the Geostatistical Analyst extension of ArcGIS (Johnston et al. 2001) and resulted in contour maps of abundance. Each map was then scaled to best reveal any patterns present in abundance.
S.E. Newson, British Trust for Ornithology