3.2 Study area
3.3.3 Ecological niche models
I used the maximum entropy algorithm (MaxEnt 3.3.3 K; Phillips et al., 2006; Phillips & Dudík, 2008; Elith et al., 2011; Warren & Seifert, 2011) to construct red panda ENMs. In order to evaluate the importance of biological variables (i.e., forest cover and water occurrences), relative to climatic variables, for predicting red panda distribution, I developed five ENMs: (1) climate model, using only climatic variables; (2) climate—water model, using climatic variables and water occurrences;
(3) climate—forest model, using climatic variables and forest covers; (4) climate—water—forest model, using climatic variables, water occurrences and forest covers; and (5) water—forest model, using water occurrences and forest covers. I compared the predicted distribution of climate model (model 1) to the remaining four models to assess how adding biologically-informed variables (i.e.
forest covers, water occurrences) alters predictive red panda distributions.
I retained only one of the predictive variables that are highly correlated in model building (|r| > 0.85; Table S3.2), which helps to reduce multi-collinearity among the variables in model building process (Graham, 2003). Among a set of highly correlated variables, I retained the one that is biologically important and/or has the highest contribution to model fit based on a jackknife
analysis in MaxEnt. For example, mean temperature of the coldest quarter (BIO-11 in WorldClim) may be considered biologically important for red pandas, because their metabolic rates were found to increase in winter at environmental temperature between 5.3°C and 7.6°C, but not in summer at environmental temperature between 15.5°C and 20.2°C (Fei et al., 2017). I used linear and quadratic features to constrain the variance of the predictors (Elith et al., 2011; Merow et al., 2013). I divided the occurrence data into subsets of 75% and 25% records as the training and test data sets respectively. I compared the area under the receiver operating characteristic curve (AUC) at different numbers of background points (background points are used as pseudo-absence data points) and selected the lowest number (500) at which the AUC approaches asymptote (Fig. S3.1).
I ran 20 replicates for each model (see Fig. S3.2 for standard deviation rasters of the replicates).
I converted MaxEnt outputs (i.e., suitability scores) to binary values (i.e.
suitable/unsuitable) using the threshold rule of ‘maximum training sensitivity plus specificity’, which minimizes the mean of the error rates for presences and pesudo-absences and is appropriate for presence-only data (Liu et al., 2016). For model validation, I calculated AUC (Hajian-Tilaki, 2013), true skill statistic (TSS), sensitivity and specificity (Allouche et al., 2007; Lobo et al., 2008). For model comparison, I quantified differences in the amount of predicted suitable area between climate model and each of the other four models (i.e. climate—water model, climate—
forest model, climate—water—forest model, water—forest model). To identify gaps in protection, I overlaid the predicted red panda distributions of all five models with the protected areas in the mountain regions of Nepal (http://www.wdpa.org). These protected areas are composed of national parks, wildlife reserves, conservation areas and hunting reserves.
3.4 Results
All models performed reasonably well (Table 3.1). Across the four metrics of model performance, the climate—forest and climate—water—forest model performed the best and the water—forest model the worst. The water—forest model predicted substantially larger suitable area than other models, c. 29,952 km2 (Fig. 3.2d). Other models predicted a similar size of suitable area from c.
13,088 km2 (climate—forest model; Fig. 3.2b), to c. 15,360 km2 (climate—water—forest; Fig.
3.2c), c. 17,168 km2 (climate model; Fig. 3.2), and c. 17,424 km2 (climate—water model; Fig.
3.2a). Using the climate model as a baseline, the addition of water occurrences alone did not substantially alter predicted suitable areas (Fig. 3.2a) whereas the addition of forest covers alone reduced the predicted suitable areas in the Annapurna region (western Nepal; Fig. 3.2b).
Furthermore, the addition of both water occurrences and forest covers yielded new predicted suitable areas in the far-western region although the Annapurna region remained unsuitable (Fig.
3.2c). Finally, without the constrains of climate variables, the water—forest model produced a very liberal prediction of suitable areas (Fig. 3.2d). Overall, the climate—forest model provided the best performance with a better balance between sensitivity and specificity, and the climate—
water—forest model uniquely revealed potential suitable areas for red pandas in the far-western region of Nepal.
Fig. 3.2 Model comparisons on predicted potential distributions of red pandas Ailurus fulgens: (a) Climate vs climate—water model, (b) Climate vs climate—forest model, (c) Climate vs climate—
water—forest model, (d) Climate vs water—forest model. Pink color indicates the suitable areas predicted only by the climate model, light blue the suitable areas predicted only by the other model being compared to the climate model (i.e. climate—water model, climate—forest model, climate—water—forest model, water—forest model), and dark blue the suitable areas predicted by both models. Map on upper right corner represents developmental regions of Nepal. Grey:
eastern; brown: central; blue: western; yellow: mid-western and red: far-western, and pink overlaying on western Nepal is Annapurna Conservation Area.
Fig. 3.3 Predicted potential distribution of red pandas Ailurus fulgens overlaying existing protected areas in Nepal. The predicted suitable areas for red pandas are in green; solid blue lines delineate the boundary of existing protected areas with buffer zone. (a) Climate model, (b) Climate—water model, (c) Climate—forest model, (d) Climate—water—forest model, (e) Water—forest model.
Map on upper right corner represents developmental regions of Nepal. Grey: eastern; brown:
central; blue: western; yellow: mid-western and red: far-western, and pink overlaying on western Nepal is Annapurna Conservation Area.
Approximately 7,728 km2 (c. 44%) of the predicted area in the climate—water model are currently protected, followed by c. 7,520 km2 (c. 44%) in the climate model, c. 7,040 km2 (c. 24%) in the water—forest model, c. 6,208km2 (c. 40%) in the climate—water—forest model, and c.
red panda habitat is outside the protected area system of Nepal (Fig. 3.3). Furthermore, there are some isolated suitable areas for red pandas in the western and far-western regions predicted by the climate—water—forest model but are not currently protected (Fig. 3.3d).
Table 3.1 Model performance based on the area under the receiver operating characteristic curve (AUC), true skill statistics (TSS), sensitivity and specificity.
Model AUC TSS Sensitivity Specificity
Climate 0.911 0.862 0.967 0.895
Climate—water 0.919 0.870 0.960 0.910
Climate—forest 0.928 0.913 0.964 0.949
Climate—water—forest 0.922 0.895 0.978 0.917
Water—forest 0.883 0.728 0.920 0.809
3.5 Discussion
This study demonstrated that adding biologically-important variables (e.g., water occurrences, forest covers) to red panda ENMs could reveal potential suitable habitats not predicted by the climate models. Specifically, the models predicted suitable red panda habitats in the far-western region of Nepal. Although red panda occurrences have not been recorded in the far-western region, Jnawali et al. (2012) suggested that this region could contain red panda habitats. Indeed, forest habitats with bamboo understory are available in this region (HPS, per. obs.). Therefore, the lack of red panda occurrences in this region may simply be the results of low survey efforts. Field surveys are urgently needed in the far-western region to confirm red panda occurrences. Another area needing attention is the Annapurna region, which is currently protected but the suitable red panda habitats might fall just outside the boundaries of the protected areas (Bista et al., 2017).
Given that different models yielded different predictions, particularly at high elevation (Fig. 3.3;
also see Kandel et al., 2015; Bista et al., 2017), red panda presence in the Annapurna region warrants additional surveys. "
All of the red panda ENMs performed well except the water—forest model, suggesting that climates could be an essential factor determining red panda distribution. There are several different and non-mutually exclusive reasons why climates are important to red panda niche distribution.
First, climates can directly impact red panda’s ecophysiology. Red pandas can tolerate temperature between 5.3°C—20.2°C (Fei et al., 2017). However, heat stress could play a role limiting their distribution, as in the case found for the giant panda (Zhang et al., 2017). Second, climatic conditions such as temperature and precipitation determine primary productivity (Nemani et al., 2003; Del Grosso et al., 2008), thereby influencing food availability for animals. Third, the red panda is a bamboo specialist, using Abis spectabilis and Betula utilis for sheltering (Sharma &
Belant, 2009), and feeding on specific bamboo species, such as Arundinaria spp., H. aristata, and Thamnocalamus spp. (Yonzon, 1989; Panthi et al., 2012; Sharma et al., 2014b). Because bamboo distribution can be sensitive to climatic conditions (Caccia et al., 2009; Lin et al., 2018), red panda distribution may be tracking the distribution of bamboos, making climate conditions useful predictors for red panda niche models.
It is well recognized that for the giant panda, it is important to understand bamboo distribution and its possible shifts under climate change are important factors of their predicted distribution (Tuanmu et al., 2013; Li et al., 2015a,b; Tang et al., 2018). Bamboo has synchronized mass flowering (Janzen, 1976), and climate change might have negative effects on bamboo seedlings (Lin et al., 2018). Climate change effects have been reported for tree species such as A.
spectabilis and B. utilis (Tiwari et al., 2017a,b), which are also used for sheltering by red pandas
(Yonzon et al., 1997; Sharma & Belant, 2009). Increased rainfall intensity due to climate change could lead to more landslides (Crozier, 2010), which may reduce bamboo distribution in the mountain regions (Wang et al., 2009). Any negative effects of climate change on bamboo distribution will likely influence the current and future red panda distribution.
Red pandas are prone to habitat fragmentation due to their limited mobility (Yonzon &
Hunter, 1991). In fact, even inside the protected areas, forest resource extraction and natural barriers such as high mountains and rivers can isolate their populations (Roberts & Gittleman, 1984; Yonzon & Hunter, 1991; Crooks et al., 2017; Panthi et al., 2017). Outside the protected areas, anthropogenic impacts such as forest resource extraction, grazing and land use change are more intense, further reducing available habitats for red pandas. Approximately 56%–76% of the predicted suitable areas for red pandas across the five models fall outside the current protected areas, which is in agreement with the findings of smaller-scaled studies (Yonzon et al., 1997; Bista et al., 2017). To connect all suitable habitats and improve connectivity among red panda populations, we recommend extending current protected areas to include adjacent suitable red panda habitats and/or utilizing existing community forests as corridors to connect suitable red panda habitats and current protected areas. Community forests in Nepal are forested lands in close proximity to villages. The Nepal government encourages local people to co-manage community forests such that natural resources in these forests can be utilized in a sustainable way (Bhattarai, 2016). A total of 18,135 km2 of forest is managed as community forests outside the protected areas of Nepal (MOFSC, 2017), and deforestation has been decreasing in the past 15 years (2005–2014) owing to effective community forest programmes (Reddy et al., 2017). Furthermore, people in Nepal generally have positive attitudes towards red panda conservation (Sharma et al., 2017).
Therefore, using community forests as red panda habitats or corridors is a real possibility.
Biological corridors are known to facilitate animal migration, recolonization, and dispersal (Tischendorf & Fahrig, 2000; DeCesare & Pletscher, 2006; Beier et al., 2011). If more community forests in Nepal can be managed for wildlife conservation through working with local people (e.g., awareness programmes, incentive schemes, ecotourism promotion), not only red pandas but other species such as Himalayan black bear Ursus thibetanus, musk deer Moschus chrysogaster, common leopard Panthera pardus, ghoral Naemorhedus goral, Himalayan tahr Hemitragus jemlahicus and barking deer Muntiacus vaginalis may also benefit (Jnawali et al., 2011).
There are two limitations to this study. First, the number of red panda occurrences used in constructing the ENMs was relatively small considering the size of the study area. Fortunately, the model performances were acceptable, suggesting that a modest sample size might not be a critical problem in this case (Elith et al., 2006; Wisz et al., 2008). More field surveys, particularly in the areas that have been identified in this study as potential red panda habitats (e.g., far-western region), will add valuable occurrence data if able to confirm their presence and further improve red panda niche models. Another limitation is the incompleteness of biologically-relevant data layers. For example, even if obtaining a comprehensive bamboo distribution is difficult, the reliability of using forest covers as proxies for bamboo distribution needs to be more carefully examined in future studies. Furthermore, layers of other biological factors that might be also important to red panda survival, such as predation pressure from common leopard Panthera pardus or livestock-grazing intensity (Yonzon & Hunter, 1991), remain unavailable. Despite the data limitation, this study illustrates the importance of incorporating biologically-relevant data other than the basic climate data set in the niche models of threatened species such as the red panda.
Chapter 4
Conservation of the red panda Ailurus fulgens: a review of the current state-of-knowledge 4.1 Red panda ecology
The red pandas Ailurus fulgens, is a bamboo specialist (Panthi et al., 2012; Sharma et al., 2014b), inhabiting sub-tropical and temperate montane forests in the Himalayas and southern China (Roberts & Gittleman, 1984; Yonzon & Hunter, 1991; Yonzon et al., 1991; Glatston et al., 2015).
The red panda is separated into two sub-species A. f. fulgens and A. f. styani by Nujiang river in Yunnan, China (Wei et al., 1999; Groves, 2011), with A. f. fulgens distributing in Myanmar, Bhutan, India and Nepal and A. f. styani in Sichuan and a small portion of Yunnan. Due to their remote geographic locations, isolated populations across multiple countries, and arboreal and nocturnal life styles (Roberts & Gittleman, 1984; Johnson et al., 1988; Glatston et al., 2015), field studies on red pandas have been sporadic (Yonzon & Hunter, 1991; Glatston et al., 2015), and some aspects of their ecology is studied using zoo populations.
In the 1980–90’s, three groups of researchers attempted radio telemetry on adult red pandas (3 males and 3 females in Langtang National Park, Nepal; Yonzon, 1989; 1 female in Wolong Reserve, China; Johnson et al., 1988; 1 male and 1 female in Wolong Reserve, China; Reid et al., 1991b), and they were able to provide some estimates on red panda’s home range (0.94–11 km2) and mobility (daily movement distance c. 100–540 m). Red pandas use elevated objects such as fallen logs, cut tree stumps, shrub branches, to access bamboos (see Chapter 2; Wei & Zhang, 2011a). Red pandas are less active during snow fall period while more active during the time of bamboo shoot production and fruiting seasons (May–June) (Reid et al., 1991b). Red pandas’
temperature tolerance is estimated to be between 5.3°C—20.2°C (Fei et al., 2017), which may restrict their distribution in the Himalayas. Red pandas give birth once a year, mostly singletons
or doubletons (Yonzon & Hunter, 1991). They reach sexual maturity at c. 18 months old based on the studies of zoo populations (Roberts & Kessler, 1979). Captive red pandas breed during January–March, and both captive and wild red pandas are found giving birth during June–August (Yonzon, 1989; Northrop & Czekala, 2011). Red pandas use hollow trees as nests (Roberts &
Gittleman, 1984; HPS, per. obs.).