Field-level bird abundances were assessed across 29 farms selected from within the area covered by the environmentally sensitive area (ESA) AESs in the Peak District (for details of AES implementation in England, see Hodge & Reader 2010
). Although closed to new entrants, the long-running ESA remained the dominant AES in the region at the time of our surveys, making it an ideal case study. Fields were surveyed on two separate early-morning visits at least six weeks apart between 28 March and 5 July 2007. A transect was walked through each field and the presence of all birds recorded. Fields were small (median 2.1 ha) with few obstacles obstructing vision, so field-level abundances were taken to be the total number of birds recorded without the necessity of estimating detectability. Total
avian abundance for each field was defined as the higher of the abundances recorded on the two visits. We further divided this into two assemblages of greater relevance to conservation, upland specialist
and conservation concern
abundance (Dallimer et al. 2009
Each field was classified according to whether it was improved or semi-improved grassland during surveys. We quantified the landscape-scale habitat composition in a 500 m buffer around each field using a GIS based on the Land Cover Map 2000. Two classes of land use were defined: (i) seminatural (seminatural grassland, scrub, bracken, moorland, woodland) and (ii) intensive use (improved grassland, arable land, urban areas). Data pertaining to all AESs operating in the study system were taken from a GIS layer provided by Natural England, which included reference to the payment made (range £18–260 ha−1), if a given survey field was part of the ESA scheme. The proportion of the buffer around each field that was included in any AES was also determined.
We modelled avian abundance for the three different assemblages, using Poisson errors (corrected for over-dispersion where necessary), against AES and habitat explanatory variables at both field and landscape scales. All possible model combinations were constructed for the predictor variables, using AIC comparisons to identify the most parsimonious model (Burnham & Anderson 2002
). We used a generalized linear mixed model to account for the lack of independence between fields within the same property. Field area was forced into each model as a covariate of abundance. We anticipated that the relationship between avian abundance, habitat and AES provision may not take a simple linear form. Hence, we included interaction and quadratic terms in the modelling process.
For each assemblage, we determined: (i) model weights for candidate models, (ii) parameter estimates for each explanatory variable calculated by averaging across all models, (iii) the relative importance of each variable in explaining field-level avian abundances, by calculating wi, the Akaike weight, and (iv) model explanatory power, by assessing the correlation between predicted and observed avian abundances; conventional r2 measures are not appropriate in our modelling approach. All analyses were carried out using lme4 in R 2.9.2.