Two statistical methods to validate habitat suitability models using presence-only data

摘要: Predicting species occurrence using a modelling approach based on geographic information system (GIS) represents new methodological tool which can be used to endorse conservation policies, condition that models are tested for reliability. Habitat suitability often predict through the of proper environmental variables. A major constraint in building large-scale distribution is availability data and therefore deductive adopted. This suitable bio-diversity assessment as it applied great number species, does however require statistical validation, both test accuracy input deal with presence (often affected by high spatial temporal variability). Available are, large majority, represented small sample collected without an ad hoc sampling design absence. Reliable absence not easily obtained animal due elusive behaviour, poorly accessible habitats their activity patterns. Therefore, considered ambiguous. Hence need develop tools presence-only data. Moreover, rarely geo-referenced precision. Often spatially degraded referred polygon within exact location cannot recovered. Consequently, validation procedure should statistically consider this uncertainty. In paper we suggest compositional procedurebe when strong variability, multinomial(based) precision during field survey. The two procedures have been built represent vertebrate Italy. Results show allow calibration properties models. reliability methods confirmed model rejection pattern, determined ecological factors, scale issues, size. Both wide range applications being threshold independent. © 2004 Elsevier B.V. All rights reserved.