Combining threat and occurrence models to predict potential ecological traps for Andean bears in the Cordillera de Mérida, Venezuela

Abstract

Ecological traps occur when rapid environmental change makes organisms' habitat selection cues misleading and leads them to prefer poor quality habitats. Such traps can threaten the persistence of affected populations, so techniques to predict and map potential traps are of great conservation interest. Here we present a novel method for visualizing such traps and their uncertainty at large scales in a natural landscape, by combining a spatially explicit model of anthropogenic threats with one of occurrence probability. We began with poaching and occurrence data for Andean bears in the Cordillera de Mérida, Venezuela, and applied a partitioning procedure to generate 10 replicates of three partially independent data subsets. To the first subset, we fit a previously developed model of poaching probability, while we used the second and third subsets to fit, validate and select the best of four occurrence probability models. We then combined replicates of the poaching probability model with those of the best occurrence probability model to predict the spatial distribution and uncertainty of potential ecological traps. The best occurrence model predicted high probabilities in the center and in the northern parts of Cordillera de Mérida, with variation among replicates in the same areas. Predicted areas of occurrence covered 10 217 ± 2762 km 2 (24%) of the study area. However, more than a third of this area had a high probability of being an ecological trap. Furthermore, these potential ecological traps were next to or within the largest national parks and were surrounded by large areas with high occurrence probability and low poaching probability. Future research should focus on independent verification of potential occupancy and ecological traps, as well as on bear dispersal behavior. In areas where ecological traps are confirmed, targeted education and law enforcement will be most effective, while in confirmed safe harbor areas, increasing connectivity will be equally important. Our approach will be useful to identify potential ecological traps at the landscape level created by hunting and other human activities elsewhere in the world.

Ecological traps: current evidence and future directions

“…This is an important knowledge gap, as effective management needs to consider how patches fit within the intrinsic variations in habitat quality likely to occur across the landscape [22]. Recent work mapping potential traps for Andean bears demonstrates the advantages of such a landscape approach [44].…”

Andean bear Tremarctos ornatus natural history and conservation

“…At the moment, most initiatives rely on subjective measures of success if success is measured at all, but it is paramount that an objective process of evaluation is put at the core of new and ongoing actions, in order to assess their real impact. Educational strategies have efficiently promoted the species as a flagship for the Andes, but changes in attitudes have proven more difficult to achieve (Sánchez‐Mercado et al. 2010).…”

Geographic distribution of the short-tailed river stingray (Potamotrygon brachyura): assessing habitat loss and fishing as threats to the world’s largest obligate freshwater elasmobranch

“…Presence data for RdlP included trawl stations with positive catches of P. brachyura, plus other records. The second approach was a presence-background technique, called MaxLike (Royle et al 2012), that uses presence data plus environmental predictors to estimate the probability of occurrence of a species, and is being increasingly used in species distribution modelling (e.g. Sarre et al 2013;Fitzpatrick et al 2013;Ferrer-Paris et al 2014;.…”

“…The mean of the ten runs of the best model was our final estimation of the geographic range of P. brachyura in the UP. In MaxLike models of both regions, we included a quadratic term for each predictor in order to account for potential non-linear effects (Royle et al 2012). For both modelling approaches, the best model was selected by minimising the Akaike Information Criterion (AIC) (Anderson et al 2000).…”

“…3.1.1 (R Core Team 2014), using the packages MASS (Venables and Ripley 2002), raster (Hijmans 2014), mgcv (Wood 2006) and maxlike (Royle et al 2012).…”