Knowledge-based models for predicting species occurrence in arable conditions

This paper explores the potential of rule-based habitat models to predict the occurrence of some common species in arable conditions. Models were developed for 10 arable plant species, 7 Hemiptera species, 8 carabid species and for 5 bird species whose ecology was sufficiently known. Rule sets linking species occurrence to environmental variables were produced using available literature and expert knowledge about ecological requirements of the selected species. Environmental variables described the nature and condition of habitats at various scales, ranging from vegetation quadrat to the landscape in a 1 km radius of species sampling sites. The performance of the 34 models developed was assessed in two areas of England. Results show the rule-based habitat models developed for arable plants and birds were not very successful with Cohen's k values often <0.4 for plants and very close to 0 for all bird species. Conversely, rule-based models performed surprisingly well for carabids and Hemiptera with k values on average >0.4. This suggests that ecological knowledge on these invertebrate species is more complete than we expected. The effect of species prevalence on model performance and the potential application of knowledge-based habitat models in the context of biodiversity assessment are discussed.