Habitat-based models for predicting the occurrence of ground-beetles in arable landscapes: two alternative approaches

• Published in: Agriculture, ecosystems & environment ecosystems & environment, 2003-04, Vol.95 (1), p.19-28
• Main Authors: Petit, Sandrine, Haysom, Karen, Pywell, Richard, Warman, Liz, Allen, David, Booth, Roger, Firbank, Les
• Format: Article
• Language: English
• Subjects: Agronomy. Soil science and plant productions; Biological and medical sciences; Carabids; Expert knowledge; Fundamental and applied biological sciences. Psychology; General agroecology; General agroecology. Agricultural and farming systems. Agricultural development. Rural area planning. Landscaping; General agronomy. Plant production; Generalities. Agricultural and farming systems. Agricultural development; Logistic regressions; Rule-based models; Species monitoring; Species prevalence
• ISSN: 0167-8809; 1873-2305
• DOI: 10.1016/S0167-8809(02)00173-1

The potential of habitat-based models was explored to predict the occurrence of carabid beetles in arable conditions. It was hypothesised that: (i) the habitats surrounding a location were good predictors of the occurrence of the most common carabid species; (ii) the current knowledge on habitat associations for some individual species was sufficient to develop accurate predictive models. The performance of knowledge-based models was assessed for eight well-studied carabid species. Rule sets were produced using an extensive database describing the nature and condition of the habitats recorded within a 50 m radius of the sampling sites. The performance was compared to a more classical approach based on logistic regression (LR) models, using the same original information summarised into 19 variables by correspondence analysis (CA). The performance of the rule-based (RB) models was higher than expected by chance for species occurring in less than 70% of the sites ( k>0.4) and was relatively consistent across the three areas of England where they were tested. Models developed for widespread species had a high prediction success (PS) but no discriminatory ability (low k value). LR and RB approaches gave comparable results for species of average prevalence (30–70%) while for species occurring in less than 30% of the sampled sites, the RB approach performed significantly better than the LR one. It is suggested that knowledge-based approaches could be used more widely to predict the distribution of invertebrate species. The effect of species prevalence and the potential application of knowledge-based habitat-models in the context of biodiversity assessment are discussed.