Hopes and challenges for giant panda conservation under climate change in the Qinling Mountains of China

Abstract One way that climate change will impact animal distributions is by altering habitat suitability and habitat fragmentation. Understanding the impacts of climate change on currently threatened species is of immediate importance because complex conservation planning will be required. Here, we mapped changes to the distribution, suitability, and fragmentation of giant panda habitat under climate change and quantified the direction and elevation of habitat shift and fragmentation patterns. These data were used to develop a series of new conservation strategies for the giant panda. Qinling Mountains, Shaanxi, China. Data from the most recent giant panda census, habitat factors, anthropogenic disturbance, climate variables, and climate predictions for the year 2050 (averaged across four general circulation models) were used to project giant panda habitat in Maxent. Differences in habitat patches were compared between now and 2050. While climate change will cause a 9.1% increase in suitable habitat and 9% reduction in subsuitable habitat by 2050, no significant net variation in the proportion of suitable and subsuitable habitat was found. However, a distinct climate change‐induced habitat shift of 11 km eastward by 2050 is predicted firstly. Climate change will reduce the fragmentation of suitable habitat at high elevations and exacerbate the fragmentation of subsuitable habitat below 1,900 m above sea level. Reduced fragmentation at higher elevations and worsening fragmentation at lower elevations have the potential to cause overcrowding of giant pandas at higher altitudes, further exacerbating habitat shortage in the central Qinling Mountains. The habitat shift to the east due to climate change may provide new areas for giant pandas but poses severe challenges for future conservation.

Climate change may result in habitat loss and fragmentation and synergistically determine the shifting direction of population ranges, which degrade species fitness at the habitat and/or genetic levels (Brook, 2008;Peacock, 2011;Pyke, 2008). Many studies have projected the dynamics of habitat suitability under climate scenarios to uncover the role of climate change in shaping the distribution of giant pandas and found bamboo shortages, habitat loss, and northward shifts in population range (Fan et al., 2014;Li et al., 2014;Songer, Delion, Biggs, & Huang, 2012;Tuanmu et al., 2013). Few studies have quantified habitat shifts and patterns of habitat fragmentation induced by climate change despite this being a critical component to understanding the impacts of global change on giant pandas. Most calculations of species extinction risk assume that endangered species go extinct when they no longer have suitable habitats because of climate change (Hole et al., 2009). Species that depend on one specific plant species for survival and reproduction are expected to be more vulnerable to habitat fragmentation, and the giant panda is a typical example (Piessens, Adriaens, Jacquemyn, & Honnay, 2009;Zhu, Hu, Qi et al., 2013). Therefore, due to the risk of habitat loss and fragmentation, there is an urgent need to estimate giant panda habitat spatial patterns and fragmentation to clarify key areas and strategies for future conservation planning. Additionally, data in previous studies associating giant panda presence with environmental and climatic variables were based on the results of the Third National Giant Panda Survey (TNGPS) completed in 2000, and need to be updated with the most recent survey data.
The projected rates of climatic change indicate that range shifts over the next century are likely to be great (Hole et al., 2009). The most recent IPCC-CMIP5 climate scenarios and the Fourth National Giant Panda Survey (FNGPS) completed in 2012 provide a great opportunity to develop a complete understanding of the relationship between habitat spatial patterns and climate change using updated, high-quality population and climate data. Because fringe ecological patches are more sensitive to environmental change (Russell et al., 2012), we focused on the Qinling Mountains at the northern extent of the giant panda distribution. This study aimed to (1) understand the current status and future changes in giant panda habitat range, suitability, and fragmentation in this Qinling Mountains; (2) quantify the scale of habitat shift according to direction and elevation, and describe spatial dynamics of habitat fragmentation resulting from climate change; and (3) provide conservation recommendations for giant panda regarding population management, reserve networks, and habitat restoration.

| Study area
The Several major roads run north-south in the study area (national roads 108 and 210, Xihan expressway). We chose the Qinling Mountains for the following reasons. First, these mountains have a transitional climate between northern subtropical and warm temperate zones and represent a typical area to project the impacts of climate change due F I G U R E 1 Giant panda population range and nature reserves in the Qinling Mountains, China to high climatic variability. Second, the giant pandas in the Qinling Mountains are geographically and genetically isolated from other populations and this helps simplify the scenarios for population dispersal and climate change. Third, giant pandas inhabiting mountains should be the most sensitive to climate change and trigger a habitat change or migration based on future scenarios and the edge effects of ecology, high latitude, and high population density (Guralnick, 2006). This area has also been identified as a global biodiversity hot spot and a global conservation priority (Mittermeier, Myers, Mittermeier, & Robles Gil, 2002;Olson & Dinerstein, 1998). Due to the great value of this ecosystem, this area is designated as a major ecological function-oriented zone to protect its ecological function, nature forest protection area to protect its forest resource, and water storage area to protect water resource by the Chinese central government (Fan, Tao, Qing, & Ren, 2010;Yue, Long, Long, & Qian, 2012).

| Giant panda presence data and environmental variables
The Chinese State Forestry Administration conducted the FNGPS in the Qinling Mountains from April to June 2012 and recorded 1810 sign points indicating the presence of giant pandas including feces, dens, sleeping sites, and footprints (Table 1). These data were collected at a frequency of one transect per 200 ha across the giant panda range.
Latitude, longitude, elevation, slope, vegetation, and bamboo cover at each sign point were recorded. We derived elevation and slope data from a digital elevation model based on 1:50,000 topographic maps obtained from the Chinese Academy of Sciences (www.gscloud.cn).
Vegetation with 80% accuracy and bamboo with 70% accuracy in the study area were obtained from the survey and satellite images of Landsat 5 in 2000 and Spot5 in 2012 using the maximum likelihood classification algorithm in supervised classification by Erdas 9.2 (Leica Geosystems GIS and Mapping, 2003, LLC, Atlanta, GA, USA). All above data were applied and approved by the Chinese State Forestry Administration. As the main human-induced threats to giant pandas (Gong, Meng, Chen, & Song, 2012;Zhu, Hu, Zhang, & Wei, 2013), human communities and the latest road data, including national roads, highways, and high-speed railways, were taken from previous studies and field surveys.
All geospatial data were based on the UTM WGS 84 coordinate system. The raster data resolution was 30 × 30 m, and data were analyzed using ArcGIS10.0 (Esri, Redlands, CA, USA).

| Climate data
We obtained current and future bioclimatic variables at a 30-s reso-

| Modeling
We chose the maximum entropy modeling approach implemented in Maxent (Phillips & Dudík, 2008) to identify current habitat suitability, the importance of permutation, the contribution of each environmental variable, and future status under climate change by 2050.
T A B L E 1 The source and accuracy of study data with its modeling assumption with different random subsamples (70% training and 30% test data).
The AUC value is widely used as an indicator of a model's ability to discriminate between suitable and unsuitable habitats (Dan & Seifert, 2011) and the models were considered reliable when AUC > 0.75 (Rebelo, Tarroso, & Jones, 2010). Permutation importance depends on the final model and is better for evaluating the importance of a particular variable (Songer et al., 2012). Therefore, we evaluated the importance of the habitat variables based on permutation importance.
The main environmental variables incorporated into the simulation model were vegetation, bamboo, elevation, slope, resident community, and road disturbances including current and 2050 climatic variables (Table 1), based on previous studies of habitat selection (Feng, Manen, Zhao, Li, & Wei, 2009;Hu, 1987;Liu et al., 2014;Qi et al., 2012). Due to technical limitations and the complexity of prediction, future vegetation and bamboo for our study were derived at using following assumptions: 1. The location of the tree line ecotone is considered as a particularly sensitive bioclimatic indicator of climate and landscape change (Holtmeier & Broll, 2005), and climate change will drive the tree line upslope and poleward (Hansen, 2001;Kittel, Steffen, & Iii, 2010;Theurillat & Guisan, 2001).

2.
Although bamboo is the main food source for giant pandas, bamboo and flowering stochasticity has not devastated giant panda genetic diversity historically because this species has evolved a series of adaptive strategies to switch bamboo species and disperse long distances to forage (Zhu, Hu, Qi et al., 2013), including 314,807 ha of bamboo in the Qinling Mountains from FNGPS.
Furthermore, bamboo always occurs under forests or mixes with other types of vegetation, and its classification and mapping face technical challenges and no basis for projecting dispersal range is available (Tuanmu et al., 2013;Wang et al., 2010). With enough spatial range and limitations estimating dynamics, we integrated bamboo into habitat factors but assumed that bamboo remained stable during habitat prediction modeling.

3.
According to the zoning as major ecological function-oriented zone and water resource conservation area, we also made an assumption of modeling on the human disturbance that resident community and road network will keep static by banning large-scale construc-

| Habitat suitability, shift, and fragmentation assessment
We ran the Maxent model with current and future climatic and environmental variables to assess current habitat suitability and to project changes by 2050. All current and future habitats were divided into ordinary, subsuitable, and suitable based on the habitat suitability index. Using expert knowledge and experience in the identification of habitat suitability (Wood & Dragicevic, 2007), we calculated the percentage of giant panda signs in each type of habitat to indicate suitability. Then, we created criteria for habitat suitability assessment and set the habitat suitability index as follows: Habitat with 70% of the giant panda signs was suitable, habitat with 20% of the signs was subsuitable, and other habitat was ordinary. We were able to assess habitat suitability and spatial variations under climate change in 2050 using the proportions and spatial pattern of suitable and subsuitable habitats.
In order to quantify the shift in habitat spatial pattern and fragmentation, patches of suitable and subsuitable habitat were used to analyze dynamics. Latitude, longitude, and elevation at the centroid of all suitable and subsuitable habitat patches between now and 2050 were used to quantify how habitat suitability will shift in direction and  Both the training AUC (0.91) and test AUC (0.87) indicated reliable model performance (Table 3). Additionally, the correlation between the presence of giant pandas and habitat suitability index was significant (ρ = 0.56, p < .05), confirming that our methods captured the relationship between giant panda occurrence and habitat spatial distribution under current climatic conditions. The most important climate variable based on permutation importance was temperature seasonality (Bio4, 25%), followed by precipitation seasonality (Bio15, 23%). Slope was the most important habitat variable (5.8%) among abiotic and biotic habitat factors after climate factors. The disturbance on giant panda habitat from transportation is much higher than disturbance arising from resident community (Figure 2).

| Habitat fragmentation and elevation shift due to climate change
Our landscape indices indicated that fragmentation of habitat will de- to eastern regions will form (growth in MPS and LPS), and the landscape pattern of habitat will be more aggregated (Table 4; Figure 2).
Climate will greatly affect the suitable habitat elevation pattern.
Current suitable habitat is mainly around 2,100 m above sea level, around which 61% of total current suitable habitat occurs; future suitable habitat is mainly around 2,200 m above sea level and represents 60% of total future suitable habitat. The majority of suitable habitat loss will occur under 1,900 m from 32,678.1 ha (current) to 16,102.0 ha (by 2050) with 51% decline, and suitable habitat above 1,900 m will noticeably increase from 88,470.6 ha (current) to 159,210.1 ha (by 2050) with 80% growth following climate change (Table 5; Figure 2).
MPS and LPS dynamics for suitable habitat elevation showed similar qualitative and quantitative changes regarding fragmentation and loss as for TA, suitable habitat patch size will increase and become more concentrated with a distinct drop in landscape fragmentation at 2,000-2,400 m above sea level (also the most preferred altitude of giant pandas).

| DISCUSSION
In general, wildlife population ranges shift toward the poles and to higher elevations under climate change conditions (Hole et al., 2009).
Our results are coincided with previous studies at the spatial shift of  Songer et al., 2012), and the results from giant panda habitat selection studies that found fidelity for specific slopes override elevation and aspect because of the need to maintain balance between energy intake and expenditure (Hu, 1987;Schaller, 1985;Zhang et al., 2011). Moreover, our study provides a methodological case to analyze future climate variables with better accuracy and can be applied to other climate change studies.
Habitat fragmentation has always been the main threat to giant pandas (Gong, Yang, Yang, & Song, 2010)   requirements (10.62 km 2 )and provides hope for population expansion (Pan et al., 2014). Actually, giant pandas in the eastern reserves currently face a high probability of local extinction due to small population size and poor genetic diversity, and it is necessary to increase the population size via translocation or the reintroduction of captive individuals.
The changes in habitats modeled here mean that a new giant panda conservation plan is needed to respond to climate change.
Due to habitat integrity at the landscape scale (Loucks et al., 2001), the northeast part of the Qinling Mountains may become suitable habitat, and as it is currently outside the current reserve network, it should be included in current conservation planning and zoning ( Figure 2). A new habitat restoration program is also needed to welcome the climate change-driven migration of giant pandas by 2050 ( Figure 2 T A B L E 5 The elevation pattern of current and future (2050) suitable habitats