Giant Kangaroo Rat Species Distribution Model: William Bean (2014)

We created a distribution model for GKR using Maxent (Phillips & Dudik 2008). In addition to the 120 GKR presence points acquired from our trapping data, we used the gbif function in the dismo package in r (Hijmans et al. 2012) to obtain 38 spatially referenced records for GKR from museum collections. Records were restricted to those obtained after 1950, in order to match the temporal range of environmental variables. We then obtained 19 climate layers (Hijmans et al. 2005) frequently used in distribution modelling as independent variables (Graham & Hijmans 2006). Bioclim layers are estimated as mean conditions from 1950 to 2000. From the initial 19 layers, we limited our models to six that we believed to be most important in describing GKR distribution and that were only partially correlated with each other. These layers were annual mean temperature (BIO1); annual precipitation (BIO12); minimum temperature of the coldest month (BIO6); precipitation of the driest month (BIO14); and precipitation of the driest quarter (BIO17). In addition to climatic layers, we included soil particle size and slope as predictor variables. Giant kangaroo rats require medium-sized soil particles in order to construct their burrow systems and are generally restricted to areas of less than 10° slope (Grinnell 1932; Williams 1992). Study extent was limited to a buffer of 100 km from all occurrence records. The climate layers were the coarsest resolution layer (30 s), and thus soil particle size and slope were aggregated to match this resolution.

Maxent produces, as an output, an estimate of habitat suitability represented by a raster at the same extent and grain as input layers. Model values may be output in three formats: raw, logistic and cumulative (Phillips, Anderson & Schapire 2006). The logistic output may range from 0 to 1 and, if prevalence is well estimated in the model, may represent a probability of presence. Estimating prevalence with presence-only data may be difficult, and so the Maxent output is typically treated as a more general measure of habitat suitability, with suitability likely correlated with probability of presence. Performance was measured using the Area Under the Curve (AUC, Hanley & McNeil 1982).