Review of statistical methods and data requirements to support post market environmental monitoring of agro ecosystems

We used generalized linear mixed models (SAS macro Glimmix) with Poisson error distribution and log-link function to analyse the effects of farm practice and landscape type on butterfly species richness. The butterfly abundance (individuals per 50 m of transect) was logtransformed [ln(x+0•1)] to achieve residual Normal distribution, and analysed using general mixed models with Normal error distribution. Data were analysed at the segment level to account for the slightly unequal sampling effort at different farms (all results were qualitatively the same if analysed at transect level). The fixed factors in the models were year, landscape type, farm practice and landscape type × farm practice; the random factors were farm pair and farm identity; the repeated factor was visit (nested within year). We selected the covariance structure for the repeated factor based on AIC (Akaike Information Criterion), which in all cases resulted in a first-order autoregressive structure being used. We used the Satterthwaite method (Littell et al., 1996) to approximate denominator degrees of freedom. Pearson correlation was used to assess the association between proportion of organic arable land and productivity of the arable land (yield of spring barley; kg ha−1). All statistical analyses were performed in SAS 9•1 for Windows. Although we can conclude that both farming practice and landscape heterogeneity significantly affects butterfly species richness and abundance, the most interesting result is the effect of the interaction between the two. A similar relationship has been proposed for arable weeds (Roschewitz et al., 2005). 1.4.6.3. Example 3: Application of GAM Fewster, R. M., Buckland, S. T., Siriwardena, G. M., Baillie, S. R. and Wilson, J. D. (2000) Analysis of population trends for farmland birds using generalized additive models. Ecology, 81: 1970-1984.