**3. Results**

The AUC and pRCOC values of the final distribution consensus model produced from ten low correlated bioclimatic variables and nineteen spatially rarefied occurrences was 0.925 and the pROC value was 1.874, indicating that the model has a predictive ability that is better than random, and that overlapped well with known occurrences for *Manihot walkerae* (Figure 2). Areas of high suitability shown in red are found primarily along the Texas–Mexico boundary and extend towards the southeastern portion of the Tamaulipan thornscrub ecoregion (Figure 2). The variables shown to contribute the most to the model from the jackknife test were Annual Mean Temperature (BIO 1), Temperature Annual Range (BIO 7), Precipitation of Coldest Quarter (BIO 19), Annual Precipitation (BIO 12), and Precipitation of Seasonality (BIO 15), which collectively contributed 96.5% to the model (Table 1). The bioclimatic variables that contributed least were Mean Diurnal Range (BIO 2), Isothermality (BIO 3), Precipitation of Wettest Month (BIO 13), and Precipitation of Driest Month (BIO 14), which collectively contributed 3.4% to the model (Table 1).

The present geographic distribution binary consensus model was used to compare the percent change of geographic distribution with the future climatic models (Figure 2b). For the year 2050, a change in distribution is predicted to occur in the northeastern portion of the Tamaulipan thornscrub study area (Figure 3). Both the CM3 and CMIP5 GCM, at a 4.5 RCP emission scenario, projected an increase of 7.20% and 7.42% in distribution, respectively, primarily in the northeastern portion of the study area (Figure 3a–c). However, at a more severe emission scenario RCP 8.5, CM3 predicted a reduction in distribution, most notably in the north and southwestern portion of the study area (Figure 3b) and a slight decrease in the southernmost portion of the study area for the CMIP5 (Figure 3d). The HadGEM predicted a loss of distribution at both emission scenarios, but notably it is higher for the 4.5 RCP −5.60% than the 8.5 RCP −1.19% (Figure 3e,f and Table 2).

All future climatic models for the year 2070 predicted a loss of potential distribution with notable differences in percent lost between the models (Figure 4 and Table 2). For CM3, a loss of −13.63% of geographic distribution was seen in all areas especially in the northeastern area, northwestern portion along the border, and in the southernmost portion of the Tamaulipan thornscrub ecoregion (Figure 4b). The largest reduction of distribution is shown by the CMIP5 at the 4.5 RCP scenario in the northeastern, northwestern portion along the border, and the southernmost of the study area (Figure 4c) The HadGEM GCM at both emission scenarios show slight decreases of distribution in the southernmost portion of the study area (Figure 4d,e) with a greater loss calculated for the most severe emission scenario −4.13% than the intermediate scenario −2.61%.

**Figure 2.** Consensus models of present geographic distribution for *Manihot walkerae* based on ten bioclimatic variables and 19 spatially rarefied occurrences. (**a**) The color scale ranges from blue to red, with blue depicting areas of unsuitable distribution (value: 0) and red areas with highest potential of distribution (value: 1). (**b**) Binary model, blue areas are potentially suitable. The calculated area under the curve (AUC) and partial receiver operating characteristics (pROC) values for this model are 0.925 and 1.874, indicating good performance.

**Figure 3.** Future potential geographic distribution for *M. walkerae* for the year 2050 using ten bioclimatic variables and nineteen spatially rarefied occurrences. Panels (**a**,**b**) correspond to the CM3 GCM at an RCP of 4.5 and 8.5, respectively. Panels (**c**,**d**) correspond to the CMIP5 GMC at an RCP of 4.5 and 8.5. Panels (**e**,**f**) correspond to the HadGEM GMC at an RCP of 4.5 and 8.5.

**Figure 4.** Future potential geographic distribution for *M. walkerae* for the year 2070 using ten bioclimatic variables and nineteen spatially rarefied occurrences. Panels (**a**,**b**) correspond to the CM3 GCM at an RCP of 4.5 and 8.5, respectively. Panels (**c**,**d**) correspond to the CMIP5 GMC at an RCP of 4.5 and 8.5. Panels (**e**,**f**) correspond to the HadGEM GMC at an RCP of 4.5 and 8.5.
