Abstract from BTO Research Report No 345:

Newson, S. & Noble, D. (2003, published 2005)

Comparison of population trends abundance and species richness on land affected by SRDP agri-environment schemes with non-agreement land.

1. INTRODUCTION

The Scottish Rural Development Plan (SRDP), which is aimed at improving the economic, environmental and social conditions of rural communities in Scotland, was announced by the government in 2000. One of its components is an agri-environment scheme programme, including the Rural Stewardship Scheme, the Organic Aid Scheme, the continuation of the Woodland Grant Scheme and the Farm Woodland Premium Scheme. The purpose of this preliminary study is to investigate the use of existing bird survey data to evaluate the effectiveness of the Scottish Rural Development Plan (SRDP) agri-environment scheme in Scotland. As holders of many large-scale datasets of counts of birds across the UK, the BTO was contracted to extract and analyse data from the BTO/JNCC/RSPB Breeding Bird Survey (BBS) in relation to their participation in the SRDP agri-environment programme

1.1 Data extraction

The BTO/JNCC/RSPB Breeding Bird Survey (BBS) was introduced in 1994 to monitor population trends of a broad range of breeding birds across the UK. The BBS, which uses a line transect method for recording birds, calculates annual population indices for all species where there is sufficient coverage, and population trends are reported annually (e.g. Noble et al. 2001). The use of volunteers is maximised through a stratified random sampling design. BBS survey squares are randomly selected from a list of all 1-km squares in the National Grid, excluding coastal squares with less than 50% land. Within each region, squares are selected randomly, and allocated to volunteers through a network of voluntary regional organisers.

Fieldwork involves two counting visits by a volunteer observer to each survey square. These line transects should follow the ideal, a straight lines through the square as closely as possible, but allow for minor deviation due to impracticalities for the observer caused by land features. The habitats of each 200-m section along line transects are recorded using a hierarchical coding system formulated from Crick (1992). Early morning bird counts are carried out between early April and mid-May and between mid-May and late June respectively. Observers record all birds that they see or hear as they walk along their transect routes within each 200m section in one of three distance categories from the line (0-25-m, 25-100-m, 100-m or more) or in flight. The flight category, excludes birds obviously associated with the square (e.g. display flight of Skylark Alauda arvensis), which is recorded in the appropriate distance band. Distances are estimated at right angles to the transect line. Juvenile and immature birds are recorded in the field, but are excluded from the computerised data and subsequent analyses. Thus, it is assumed that sightings relate to breeding, or potentially breeding adults only.

A list of the locations of all BBS squares surveyed in Scotland between 1994 and 2002 was provided to the partnership. Squares that coincided with land under SRDP agreement were matched with squares that were not under SRDP agreement, based on a suite of characteristics including geographical location (Ordnance Survey easting and northing), mean annual rainfall, average temperature, distance of the square from the sea, and the mix of land cover types in the square
(J Wright, pers comm.).

1.2 Estimating population trends

Examination of species coverage in the matched squares revealed that 17 species were recorded on a mean of at least eight squares for both SRDP and control land (Table 1). Count data were analysed using log-linear Poisson regression models fitted in SAS (SAS 1996). Annual indices were generated by modelling a matrix of annual site counts, with site and year effects (ter Braak et al. 1994). The year effect is an annual index of total numbers, whilst the site effect describes how species abundance at sites differ from one another. The first index of a run of years is set to an arbitrary value 1 and other indices are measured relative to this.

A second set of analyses examines the significance of the underlying trend by removing the year effect from the model and formally testing whether the trend is significantly different from zero. We then determine whether linear trends on SRDP land and non-SRDP control squares are significantly different from another. For these analyses, an additional variable ‘type’ was employed where ‘type’= 1 for a site on land affected by SRDP schemes and 2 on non-agri-environment scheme land. Adding to the model a year*type interaction enables a formal assessment of the significance of the difference in linear trends, between two geographic areas for the time period 1994 to 2002.

1.3 Estimating mean densities

Although bird counts using standardised protocols can be used to monitor trends over time, estimates of absolute abundance encounter the problem that raw counts represent only observed individuals. It is extremely unlikely that all individuals in the sampled population will be detected, and hence the undetected proportion must be estimated. Distance sampling provides a method for estimating this proportion (Buckland et al. 2001). This problem is further complicated by the fact that detectability varies between habitats. For example a species in open country is likely to be detectable at a greater distance from the transect line than in woodland or scrub. For an unbiased comparison of species abundance on and outside SRDP agreement land, it is necessary to take detectability in different habitats into account. Using distance sampling the distribution of detected individuals in relation to their distance from the transect line allows the probability of detection of this species to be modelled in relation to distance from the line. Assuming that all individuals near the transect line are detected, the proportion of individuals missed can be estimated, and the true number of individuals estimated.

As recommended by Buckland et al. (2001), birds recorded in the final distance band (100-m or more) were excluded from the analyses. Birds in flight were also excluded, leaving the count data from the two remaining bands (0-25-m, 25-100-m). We modelled the detectability of each species in each of seven broad habitat classes (Table 2). To take these habitat-specific detectabilities into account, we included habitat as a covariate in the analyses using the distance sampling software; Distance 4 (Thomas et al. 2002). Exploratory analysis suggested that detection functions based on half-normal models gave the best fit of the model to the data. Comparison of model fit were based on Akaike’s Information Criterion (AIC), which assesses the most parsimonious model based on maximising fit and minimizing the number of model parameters used. In all analyses we assume that whilst encounter rate (‘occurrence’) varied between squares, detectability is species-specific and varies by habitat, but not between years or between agri-environment and non agri-environment scheme land. Using this methodology we estimated mean abundance on 1-km BBS squares for each year (1994 to 2002) in each land category. To generate estimates of variance we used bootstrapping, resampling from BBS squares within years for agri-environment and non agri-environment scheme land separately. We restricted density estimates to the same 17 species included in the trend analyses.

For the same 17 species we also perform paired t-tests on the raw count data to examine whether there was a significant difference between SRDP agreement squares and their corresponding matched control squares. For an additional suite of 24 species occurring on five to eight BBS squares, a simple comparison of the raw counts on land affected by SRDP agri-environment schemes and non-agreement land was made using a Wilcoxon Rank Sum test (Zar 1999).

1.4 Species richness

A simple comparison of species diversity on land affected by SRDP agri-environment schemes and non agreement land was made by determining the number of species recorded on each BBS square within these two groups for each year (1994 to 2002) using Mann-Whitney U to test whether the medians differed significantly.

Back to Research Reports 345 onwards