Modelling Procedure
To build the model, the authors referred to the protocols suggested by various studies (Bradie & Leung,2017; Halvorsen 2013; Kaky et al.,2020; Zhang et al., 2019), and selected several basic environmental factors, including terrain, vegetation, human disturbance variables, and climate data (Table 1), which are widely available for many locations around the world. MaxENT was adopted to develop the models (Elith et al., 2011; Phillips & Dudík, 2008), as it is a widely applied method to predict species distributions and proven to be reliable for numerous taxa (Kumar & Stohlgren, 2009; Onojeghuo et al., 2015; Khadka et al., 2017; Thapa et al., 2018). Before running MaxENT, all relevant accessible factors in the dataset were given full consideration and transformed into raster format (90m × 90m), using UTM coordination system. A total 19 climate variables, 2 terrain variables (elevation, slope), vegetation (forest), 2 disturbance variables (distance to towns, density of main roads) were included in the analysis. A jack-knife method based on 10 repetitions was used to evaluate model performance, using the AUC value to evaluate the performance of the models. Finally, we adopted the average results of the 10 repetitions. The threshold value to divide the suitability into suitable habitat or unsuitable habitat was based on the Maximum Youden index (Fluss et al., 2005).
To improve the accuracy of takin distribution map, a suitable habitat distribution map based on the final model was created. The raw takin habitat map was created after converting the resulting raster file into a shape file, which was overlayed with 5 kmx5km polygons with the resulting habitat suitability value. All known takin occurrence locations were used to generate a raster map based the euclidean distances to the locations and divided the raster value into 5 classes using 10km intervals. Based on the takin distribution map and euclidean distance map, polygons at the edge of suitable habitat were identified. The edge was selected where habitat present or distance to nearest location less than 20km, producing a species distribution polygon with jagged edges. To smooth the shape and cover as much potential habitat in the distribution range, the edge angles were modified to those less than 90 degrees or equal to 90 degrees by connecting consecutive corners with a straight line, making all angles larger than 90 degrees. After smoothing, this map was overlapped with CDR of takin from the IUCN and considered the overlapped area as the updated distribution range (UDR).