作者: Lesley Gibson , Brent Barrett , Allan Burbidge
DOI: 10.1111/J.1472-4642.2007.00365.X
关键词: Altitude 、 Ecology 、 Habitat 、 Spatial distribution 、 Geography 、 Generalized linear model 、 Biodiversity 、 Vegetation 、 Regression 、 Species distribution
摘要: In the development of a species distribution model based on regression techniques such as generalized linear or additive modelling (GLM/GAM), basic assumption is that records presence and absence are real. However, common concern in many studies examining distributions absences cannot be inferred with certainty. This particularly case where rare, difficult to detect and/or does not occupy all available habitat considered suitable. The western ground parrot (Pezoporus wallicus flaviventris) southern Western Australia, point, only it rare detect, but also unlikely suitable habitat. A recent survey parrots provided opportunity develop predictive model. As data were susceptible false absences, these replaced randomly selected ‘pseudo’ modelled using GLM. comparison, presence-only information was relatively new approach, MAXENT, machine-learning technique has been shown perform comparatively well. performance both models, assessed by receiver operating characteristic plot (ROC) high (AUC > 0.8), MAXENT performing marginally better than These approaches indicated prefers areas altitude, distant from rivers, gently sloping level habitat, an intermediate cover vegetation there mosaic ages. this case, use resulted identification important environmental attributes defining occurrence parrot, additional factors account for inability bird should component refinement.