Assessing vulnerability of giant pandas to climate change in the Qinling Mountains of China

Abstract Climate change might pose an additional threat to the already vulnerable giant panda (Ailuropoda melanoleuca). Effective conservation efforts require projections of vulnerability of the giant panda in facing climate change and proactive strategies to reduce emerging climate‐related threats. We used the maximum entropy model to assess the vulnerability of giant panda to climate change in the Qinling Mountains of China. The results of modeling included the following findings: (1) the area of suitable habitat for giant pandas was projected to decrease by 281 km2 from climate change by the 2050s; (2) the mean elevation of suitable habitat of giant panda was predicted to shift 30 m higher due to climate change over this period; (3) the network of nature reserves protect 61.73% of current suitable habitat for the species, and 59.23% of future suitable habitat; (4) current suitable habitat mainly located in Chenggu, Taibai, and Yangxian counties (with a total area of 987 km2) was predicted to be vulnerable. Assessing the vulnerability of giant panda provided adaptive strategies for conservation programs and national park construction. We proposed adaptation strategies to ameliorate the predicted impacts of climate change on giant panda, including establishing and adjusting reserves, establishing habitat corridors, improving adaptive capacity to climate change, and strengthening monitoring of giant panda.


| INTRODUCTION
Rapid climate change has been widely recognized as a major threat to biodiversity (Cramer et al., 2014). Compelling evidence has already been presented of the effects of climate change on geographic distributions (Ancillotto, Santini, Ranc, Maiorano, & Russo, 2016;Molina-Martínez et al., 2016), population dynamics (Auer & Martin, 2013;Lehikoinen et al., 2016), phenological phase (Lučan, Weiser, & Hanák, 2013;Yang & Rudolf, 2010), and evolution (Charmantier & Gienapp, 2014;Koen, Bowman, Murray, & Wilson, 2014), and these impacts are predicted to be exacerbated in future (Rinawati, Stein, & Lindner, 2013;Urban, 2015). Projected change rates of climate are now getting faster than they were in the past (IPCC, 2014). If global warming is not effectively controlled, a mean increase in global temperature of >2.0°C could be the result (2.0°C is defined as "dangerous"; UNFCCC, 2015), and 15%-35% of global species could be committed to extinction (Thomas et al., 2004). Although the impact of climate change on the extent and rate of species extinction is still controversial, it is clear that the trend of global warming will accelerate the extinction risk for species (Malcolm, Liu, Neilson, Hansen, & Hannah, 2006;Pereira et al., 2010;Urban, 2015).
The giant panda (Ailuropoda melanoleuca) is probably one of the world's most treasured endangered species (Wei et al., 2015). Its habitat is currently restricted to six isolated mountain ranges in Sichuan, Shaanxi, and Gansu provinces in south-central China (State Forestry Administration, 2015). The giant panda was listed as an endangered species by the International Union for Conservation of Nature (IUCN) in 1996 (IUCN, 1996) due to their limited geographic range, the risk of small and isolated populations, low reproductive rates, habitat loss, and diet specialization (Swaisgood, Wang, & Wei, 2016;Wang, Ye, Skidmore, & Toxopeus, 2010;Wei et al., 2015). A narrow geographic distribution makes them highly susceptible to climate change (Liu, Guang, Dai, Li, & Gong, 2016;Songer, Delion, Biggs, & Huang, 2012).
Over the past decades, the Chinese government implemented many conservation programs to protect giant panda, such as establishment of reserves (State Forestry Administration, 2015), the panda monitoring project (Wei et al., 2015), and the Grain-to-Green program (Li et al., 2013). From 1988 to 2015, the population of giant panda grew from 1,114 to 1,864 (State Forestry Administration, 2015), and the species has been downlisted from "Endangered" to "Vulnerable" in the IUCN Red List of Threatened Species (Swaisgood et al., 2016). The Chinese government announced that giant panda conservation programs will continue and will establish national parks in the giant panda's range to specifically strengthen further conservation of giant panda (State Forestry Administration, 2016a). Therefore, a major motivation for assessing the vulnerability of giant panda is to provide adaptive strategies for conservation programs and development of national parks to reduce effectively potential climate-related threats to the species.
In this study, we use the maximum entropy model (i.e., Maxent, Phillips, Anderson, & Schapire, 2006) to predict the habitat suitability, to assess vulnerability of the giant panda to climate change, and to identify the potential refuges and corridors. Furthermore, we propose the conservation strategies for the species and provide fundamental information for establishing giant panda national parks in the Qinling Mountains of China.

| Study area
The study area is located in the Qinling Mountains (106°30′-108°05′E, 32°40′-34°35′N) in Shaanxi Province in China. The Qinling Mountains are characterized by a specific geographic system in terms of topography and climate, and include the boundary between the temperate and subtropical zones (Zhao, Zhang, & Dong, 2014). The mountains rise from 222 m to 3,734 m, with a gentler gradient on the southern slope; however, their northern face is generally steep (Pan et al., 2001). Regarding the differences in climate between northern and southern China, the southern slope is generally warmer and moister than the northern face, and climatic conditions vary with elevation gradient (Pan et al., 2001).   (Baek et al., 2013). For the 2050s, the average increase in global temperature of 0.9-2.0°C under RCP4.5 would fall within a 2°C global warming limit (UNFCCC, 2015). The time horizon of the 2050s was selected for being a date far enough in future for significant changes to have occurred (Young et al., 2012).

| Data preparation
Other environmental variables were also used to construct the panda distribution models (Fan et al., 2011;Loucks et al., 2003).
The densities of rivers, roads, and settlements were obtained from a 1:1,000,000 map of China (National Geomatics Center of China, data are available at http://atgcc.sbsm.gov.cn). Elevation data were derived from a digital elevation model with a resolution of 30 s, obtained from the WorldClim database (Appendix 1). Because nonclimate variables (i.e., densities of roads, rivers, and settlements) were not available for the 2050s, we used the same variables in projections for the 2050s.
Thirteen variables (annual precipitation, annual temperature range, density of rivers, density of roads, density of settlements, elevation, mean diurnal range, min. temperature of coldest month, precipitation of warmest quarter, precipitation seasonality, precipitation of driest quarter, precipitation of driest month, and temperature constancy) which were the most biologically meaningful for giant pandas were retained (Appendix 2; Li, Xu, Wong, Qiu, & Li, 2015;Songer et al., 2012). Subsequently, we first input thirteen environmental variables layer into the Maxent model. Then, we input the set of most important variables based on permutation importance obtained from first model output, to construct the giant panda finally distribution model, and rerun the Maxent models.

| Habitat suitability model
We used the Maxent software (version 3.3.3k) to build the habitat suitability model for the giant panda. This approach is considered one of the best performing algorithms in predicting species distribution with presence-only data (Elith, Phillips, Hastie, Dudík, & Chee, 2011).
It has been extensively applied to project species range shifts under climate change (Li, Clinton, et al., 2015;Lei, Xu, Cui, Guang, & Ding, 2014;Songer et al., 2012). Maxent estimates species distributions by finding the probability distribution of maximum entropy, subject to the constraints of the data that are available (Phillips et al., 2006).
Maxent also estimates the importance of variables and contributions representing the degree to which each variable has contributed to the model, based on jackknife tests. We divided the occurrence data of giant panda into training sets (75%) for model building, and testing sets (25%) for model evaluation, and conducted a subsample F I G U R E 1 Distribution of giant pandas in Qinling Mountains procedure (Khatchikian, Sangermano, Kendell, & Livdahl, 2010;Wisz et al., 2008) to evaluate the habitat suitability model by performing 15 replications in Maxent.
Model performance was measured using the area under the receiver operating characteristic curve (AUC). An AUC value closer to 1 represents near perfect performance of the model (Phillips et al., 2006). The output of Maxent comprised continuous values between 0 and 1 that were considered as probabilities of species' occurrence. We then convert these values to presence and absence predictions, based on the threshold values that maximized training sensitivity plus specificity (Liu, Berry, Dawson, & Pearson, 2005;Songer et al., 2012). The cells with probability values above the threshold value were selected as suitable habitat for the giant panda. We then removed patches <4 km 2 and >0.5 km distance from the nearest patch based on the minimum home range size and the average daily dispersal ability of specie (Pan et al., 2001). A Mann-Whitney U test was used to examine the difference in mean elevation of suitable habitat between current and the 2050s. Statistical analyses were conducted using the SPSS 19.0 software (IBM Inc., USA).

| Gap analysis of nature reserves
The Gap analysis of protection of biodiversity is a powerful and efficient step to first assess the protection of biodiversity on a coarsefilter scale (Scott et al., 1993). The current and projected suitable habitat were overlapped with the boundaries of established nature reserve networks, to explore the conservation effectiveness of these   In these formulas, A f is area of projected suitable habitat for pandas under the 2050s' climatic scenario; A c is the area of projected current suitable habitat; A fc is the overlapped distribution space between current and the 2050s (Duan, Kong, Huang, Vaerla, & Ji, 2016;Levinsky, Skov, Svenning, & Rahbek, 2007;Thuiller, Lavorel, Araújo, Sykes, & Prentice, 2005).

| Species distribution model
In the Maxent model, 256 presence points and nine variables were finally used as model parameters to construct the giant panda distribution model. The average training AUC was 0.967 ± 0.001, and the average testing AUC was 0.961 ± 0.005. The permutation importance of variables in the model as ranked from the highest to the lowest were as follows: density of rivers (34.6%), annual precipitation (30.4%), mean diurnal range (15.4%), precipitation seasonality (5.3%), precipitation of warmest quarter (4.4%), density of roads (3.6%), annual temperature range (3.1%), density of settlements (2.0%), and temperature constancy (1.3%; Figure 2). The average threshold for the probability of presence at maximum training sensitivity plus specificity was .1434.
We then defined the cells with probability values greater than .1434 as suitable habitat for giant panda.

| Vulnerability assessment
Our predicted 3,823 km 2 of unchanged suitable habitat is mainly distributed in Foping, Ningshan, Taibai, Yangxian, and Zhouzhi counties. We predicted 987 km 2 (SH c = 20.52%) of current suitable habitat distributed in Chenggu, Taibai, and Yangxian counties is expected to become vulnerable habitat. Interestingly, our results also revealed that there was an increase in the extent of suitable habitat (706 km 2 , SH f = 15.89%) in Ningshan country ( Figure 5).

| DISCUSSION
Over the past several decades, giant pandas have been exposed to several threats to their survival, such as bamboo flowering, extensive poaching, and habitat destruction (Li, Guo, Yang, Wang, & Niemelä, 2003;Pan et al., 2001). However, the Chinese government has conducted giant panda conservation programs, and many of the key threats have been mitigated (Wei et al., 2015;Zhu et al., 2013). At present, human disturbances (e.g., roads, construction projects, ecotourism, and environmental pollutants) and climate change are considered as the paramount threats that degrade and fragment panda habitat (Wei et al., 2015). Particularly, the impacts of climate change on giant panda may have negative impacts on current conservation efforts (Shen, Pimm, Feng, Ren, & Liu, 2015). Therefore, assessing vulnerability is the key step to develop proactive strategies to reduce the impacts of climate change on the giant panda.
Based on the model output, under a mild climate change scenario (RCP 4.5), 20.52% (SH c ) of current suitable habitat of giant panda is projected to transfer to unsuitable habitat, particularly in the southwestern region of the Qinling Mountains (i.e., Chenggu and Liuba counties).
Climate change associated with suitable habitat fragmentation would present another conservation challenge for this species (Holyoak & Heath, 2016;Li, Clinton, et al., 2015). Current habitat connectivity in southwestern portion of Qinling Mountains is relatively low, and these areas are predicted to experience greatest loss by the 2050s due to climate change, thereby emphasizing the need for a regional conservation strategy for giant panda conservation to protect these areas, and con- An assessment of the impact of climate change on species is a critical initial step in implementing the adaptation planning process (Rowland et al., 2011). Some nature reserves, among which planning had been done decades in advance, need to be re-evaluated considering climate change (Bellard, Bertelsmeier, Leadley, Thuiller, & Courchamp, 2012;Hansen, Hoffman, Drews, & Mielbrecht, 2010).
Our results revealed that the loss of giant panda suitable habitat would affect the conservation effectiveness of the existing giant panda reserves. These reserves do not adequately protect the current suitable habitat of giant panda, nor will they protect future potential suitable habitat. Coping strategies to deal with potential threats, particularly in those nature reserves (i.e., Banqiao, Motianling, Panlong, Sanyuan) that would suffer the greatest loss of suitable habitat under future climate change, require further indepth study. Meanwhile, three provincial nature reserves located in Ningshan county are also urgent need to improve their conservation effectiveness against climate change, because they currently support a small population of giant panda (Sun et al., 2007), but are isolated from the network of large reserves, and have low levels of protection ( Figure 4).
Vulnerability assessments can provide information about the locations that are vulnerable to climate change (Levinsky et al., 2007) and broadscale guidance to direct conservation efforts (Dubois, Caldas, Boshoven, & Delach, 2011;Rowland et al., 2011). Based on our vulnerability assessment, protection needs to prioritize habitat in which the maximum effects of climate change are predicted to occur, namely the vulnerable areas. These regions are predicted to suffer from large range contractions under climate change and present the greatest risk to the persistence of giant panda in the 2050s. Furthermore, vulnerability assessments are able to identify the potential climatic refuges for giant panda within Qinling Mountains range, namely unchanged and new increased suitability habitat (Ashcroft, 2010;, and these areas may facilitate species persistence during periods of climatic stress.

| Establishing new reserves
Gap analysis showed the distribution of current suitable habitat in Foping, Ningshan, and Taibai counties is largely unprotected, leaving significant gaps in the conservation network, and suitable habitat distributed in these areas will be discrete and fragmentated by the 2050s ( Figure 6). Therefore, new reserves need to be established in these regions to improve the connectivity of habitat.

| Establishing habitat corridors
Establishing migration corridors in juncture of Chenggu, Taibai and Yangxian counties (C1 and C2), and Ningshan county (C5; Figure 6) to increase chances for the small population of these areas to larger suitable areas, and enable giant panda to escapes from unsuitable climatic conditions. We also need to establish habitat corridors in Ningshan county (C3 and C4) to enhance habitat connectivity in these areas.

| Improving adaptive capacity to climate change
Reducing nonclimate stressors (such as invasive species, human activities, pollution, disease, and other stressors) will improve the impact on the ability of specie to adapt to climate change (Gross, Watson, Woodley, Welling, & Harmon, 2015). Such as invasive species, anticipatory actions might focus on identifying invasive species likely to expand their ranges in response to climate change, and establishing early-detection and rapid response protocols designed to keep them from invading sensitive areas.

| Strengthening monitoring on giant panda
Many nature reserves just started to consider strategies to adapt to climate change when they made their master plans. We do not fully understand how giant panda will respond to those strategies and what management measures might be effective. Therefore, a standardized monitoring program is necessary for nature reserves to collect information of climate change impacts on panda and monitor the responses of the species to the strategies.