Boundary estimation versus discrimative methods for ecological niche modelling. Simulated data illustrates why modelling a species’potential distribution is a problem for boundary estimation not classification. A. The left most panel represents the habitat in two environmental dimensions (e.g., precipitation and temperature) in locations at which a species is known to occur. The heavy curve depicts the true niche of the species. The dashed line is the convex hull of the sample, a naive estimate of the species niche. The black cross represents the center of the species niche, which is the most probable set of environmental conditions at which the species occurs. B. The center panel represents samples of environmental conditions at locations taken at random from the background distribution of environments. The green cross indicates the mean environment. The arrow is a vector of “niche displacement”. C. The right most panel depicts both occurrence and background data. The dashed line is the estimated optimal classification boundary between occurrence and background points. The blue-green color gradient depicts the conditional probability that a given instance is an occurrence points (blue: P (occurrence) = 1; gray: P (occurrence) = 0. 5; green: P (occurrence) = 0). Inset plots illustrate the region of environmental space in which each fit model makes Type I (α) or Type II (β) errors.