Prediction of butterfly diversity hotspots in Belgium: a comparison of statistically focused and land use‐focused models

• DOI: 10.1046/j.0305-0270.2003.00976.x
• Publication Date: 2004-12-26T19:41:12Z
• Authors: Dirk Maes; Marius Gilbert; Nicolas Titeux; Philippe Goffart; Roger L. H. Dennis
• Keywords: Generalized additive model

Abstract Aim We evaluate differences between and the applicability of three linear predictive models to determine butterfly hotspots in Belgium for nature conservation purposes. Location The study is carried out records located Universal Transverse Mercator (UTM) grid cells 5 × km. Methods first relationship factors correlated diversity by means modified t ‐tests principal components analysis; subsequently, we predict using based on land use, climate topographical variables well‐surveyed UTM ( n = 197). squares are divided into a training set an evaluation test model predictions. apply different models: (1) ‘statistically focused’ where entered descending order statistical significance, (2) ‘land use‐focused’ use known be related forced (3) ‘hybrid’ use‐focused model’ subsequently complemented remaining significance. Results A analyses reveals that climate, large extent, locked topography, topography most strongly with Belgium. In statistically focused model, biogeographical region alone explains 65% variability; other entering area coniferous deciduous woodland, elevation number frost days; 77% variability 66% set. region, mixed natural grassland, heathland bog, woodland edge, urban agricultural biotope model; 68% 57% hybrid all from covariates elevation, days grassland added grounds; 78% 67% Applying resulted delimitation spatially areas. Main conclusions best predictions obtained which relevant richness after climatic topographic were strictly grounds. does not satisfactory manner. When diversity, biologists need aware consequences applying such models. Although, biology, preferable models, one should always check whether applied makes sense ground. Predictive can target mapping efforts towards potentially species‐rich sites permits incorporation un‐surveyed conservancy policies. Species distribution maps produced modelling therefore used as pro‐active tools.