Map 1 Study Area

Data collection

The International Centre for Integrated Mountain Development’s (ICIMOD) land use map (http://geoportal.icimod.org/) was used to identify forest cover area within elevation range: 2000 m to 4000 m in potential red panda habitat. The DEM 90 m resolution image was used for elevation. The identified study area was divided into 504 grids of 9.6 km2 based on animal maximum home range (Yonzon 1989) using the Geospatial Modeling Environment built-in ArcGIS 10.2 version.
We selected 50% of grids which were further divided into 6 sub-grids (Area=1.6 km2). Altogether, 252 sub-grids were selected randomly for sampling across the habitat. We followed the red panda field survey and protocol for community base monitoring for data collection (MoFSC 2015). Ensuring these selected sub-grids cover the entire potential habitat, including elevation range and water availability, in the particular grid. All the available transects with an average length of 1 km at an interval of 100 m contour were traversed in each sub-grid (MoFSC 2015). We recorded the red panda presence evidences based on indirect signs, such as droppings, foot prints, foraging sign and remains of dead body parts, and direct sighting while walking along the transects. Additionally, we also recorded the occurrence data opportunistically when encountered beside designed transects. Additionally, the habitat variables were collected in a concentric sampling plots with a radius 10 m. Such sampling plots were also established in the red panda sign/sighting recorded site. Tree canopy cover and bamboo cover within a subplot of 10 m radius (A = 314.28 m2) and 1 m radius (3.14 m2) respectively also were recorded (MoFSC 2015). In total, we covered 1213 plots including 970 occurrence and 243 non-occurrence plots along 100.68 km long transects. All the field survey was conducted in June-July and September-October in 2016.

Data filtering and processing

Mostly, fecal pellets of red panda scat were used as indicative evidence of Himalayan red panda presence. The occurrence records (n=331) were used to predict distribution of Himalayan red panda using species distribution modeling techniques. Also, we categorized other field-collected data into three separate groups: topographic, habitat, and disturbance variables. All data were imported into excel spreadsheets and further statistical analysis was performed in R (Lê et al. 2008). Species absence record was nine times higher than the presence record, which could influence further statistical analysis. To address this inconsistency, we excluded those records consisting of more than 80% zero input values, and elevation below 2000 m and above 4000 m in further analysis.
Red pandas are relatively more abundant within these altitudinal range of (Choudhury 2001; Pradhan et al. 2001; Yonzon et al. 1991). At the same time, we also removed all the outliers from the data.

Potential habitat and corridors

​Occurrence data of red panda and bamboo species were extracted from the vegetation survey and used for distribution modeling based on the Maximum Entropy Algorithm (MaxEnt 3.3.3k). All 19 bioclimatic variables (11 temperature and 8 precipitation metrics) were downloaded from the WorldClim website (http://www.worldclim.org) (Hijmans et al. 2005). Our data were spatially distributed covering entire Western Nepal. All variables were converted into the ascii raster images with a cell size of 30 arc seconds (~1 km) and masked by study area boundary for the modeling process. We run 5000 repetitions with a convergence threshold of 0.00001, a regularization multiplier of 1, a maximum number of 100,000 background points, the output grid format as “logistic,” algorithm parameters set to auto features, and all other parameters at their default settings. Random test percentage was 25% of presence locations to test the performance of the model. The model outcome was evaluated by the Area Under Curve (AUC) of the Receiver-Operating Characteristic (ROC) plot. The training and test AUC above 0.75 indicated a reasonable to high model discrimination ability and good model performance (Elith et al. 2006). The habitat suitability map was built by combining the habitat model, bamboo distribution model, and forest cover using raster calculator in ARC GIS 10.2. We reclassified the habitat into three suitability classes: low (0.10.50), moderate (0.50–0.75), and high (>0.75) (Shrestha & Bawa 2014; Thapa et al. 2018). Forest and bamboo habitat within 2000 m and 4000 m that consist river/stream, (sign recoded with 0.5), occurrence of low human footprint (https://sedac.ciesin.columbia.edu/data/set /wildareas-v2-human-footprint-geographic), and away from human settlements (excluded cattle sheds) (http://sedac.ciesin.columbia.edu/data/set/gpw-v4-population-count-rev11), and habitat patches (>9.6km2) were considered as an area of potential corridor. All these layers were built and overlay using a spatial analysis tool in ArcGIS 10.2.