The effect of the extent of the study region on GIS models of species geographic distributions and estimates of niche evolution: preliminary tests with montane rodents (genus Nephelomys) in Venezuela

Abstract

Aim Various techniques model a species' niche and potential distribution by comparing the environmental conditions of occurrence localities with those of the overall study region (via a background or pseudoabsence sample). Here, we examine how changes in the extent of the study region (ignored or underappreciated in most studies) affect models of two rodents, Nephelomys caracolus and Nephelomys meridensis.Location North-central South America. MethodsWe used Maxent to model the species' potential distributions via two methods of defining the study region. In Method 1 (typical of most studies to date), we calibrated the model in a large study region that included the ranges of both species. In Method 2, we calibrated the model using a smaller study region surrounding the localities of the focal species, and then applied it to the larger region. Because the study region of Method 1 is likely to include areas of suitable conditions that are unoccupied because of dispersal limitations and/or biotic interactions, this approach is prone to overfitting to conditions found near the occupied localities. In contrast, Method 2 should avoid such problems but may require further assumptions ('clamping' in Maxent) to make predictions for areas with environmental conditions beyond those found in the smaller study region. For each method, we calculated several measures of geographic interpredictivity between predictions for the species (cross-species AUC, crossspecies omission rate, and proportional geographic overlap).Results Compared with Method 1, Method 2 revealed a larger predicted area for each species, less concentrated around known localities (especially for N. caracolus). It also led to higher cross-species AUC values, lower cross-species omission rates and higher proportions of geographic overlap. Clamping was minimal and occurred primarily in regions unlikely to be suitable.Main conclusions Method 2 led to more realistic predictions and higher estimates of niche conservatism. Conclusions reached by many studies depend on the selection of an appropriate study region. Although detailed information regarding dispersal limitations and/or biotic interactions will typically be difficult to obtain, consideration of coarse distributional patterns, topography and vegetational zones often should permit delimitation of a much more reasonable study region than the extremely large ones currently in common use.

Detecting the potential sympatric range and niche divergence between Asian endemic ungulates of Procapra

“…To examine how changes in the extent of the study region affect the results of SDMs, we applied two methods to define the study region extent (Anderson and Raza 2010). In method 1, we modeled each species' distribution in a large study region that included the whole range of Procapra (28-51°N and 75-123°E; Fig.…”

“…Then, we projected the respective model for method 2 to the larger study region applied in method 1. Disadvantages are considered to exist in both methods of defining the extent in SDMs (Anderson and Raza 2010). The models may be prone to overfitting to environmental conditions present in the region where the species is known to occur in method 1; concurrently, the values for one or more environmental variables in some pixels of the larger study region may be uncovered by the model when projecting in method 2.…”

“…We tested model performance using the area under the receiver operating curve (AUC) statistic, which gives a measure of ability to discriminate between observed presences and absences, and has been widely applied (Acevedo et al 2010;Anderson and Raza 2010;Phillips et al 2006; but see Lobo et al 2008). It ranges from 0 to 1, with a score of 1 representing perfect discrimination and a score of 0.5 representing a model that performs no better than random chance.…”

“…To identify any artefactual differences that derive from discrepancies between the two methods of defining the extent, we assessed interpredictivity using two approaches (Anderson and Raza 2010). First, cross-species omission rates (a threshold-dependent measure indicating how well the models of each gazelle species predict localities of the other two gazelle species) were calculated.…”

“…Compared with spatial resolution, which has attracted substantial attention (e.g., Braunisch and Suchant 2010;Hu and Jiang 2010), the extent of the study region has until recently been ignored, or at least has not been considered explicitly in most studies (e.g., Costa and Schlupp 2010; but see Acevedo et al 2012;Anderson and Raza 2010;Barve et al 2011). Because patterns observed on one scale may not be apparent on another, it is essential to understand the theory and processes driving the observed distribution patterns in order to avoid a mismatch between the scale used for modeling and the one at which key processes occur (Guisan and Thuiller 2005).…”