Habitat fragmentation can reduce population viability, especially for area-sensitive species. The Minimum Area Requirements (MAR) of a population is the area required for the population’s long-term persistence. In this study, the response of occupancy probability of giant pandas against habitat patch size was studied in five of the six mountain ranges inhabited by giant panda, which cover over 78% of the global distribution of giant panda habitat. The probability of giant panda occurrence was positively associated with habitat patch area, and the observed increase in occupancy probability with patch size was higher than that due to passive sampling alone. These results suggest that the giant panda is an area-sensitive species. The MAR for giant panda was estimated to be 114.7 km2 based on analysis of its occupancy probability. Giant panda habitats appear more fragmented in the three southern mountain ranges, while they are large and more continuous in the other two. Establishing corridors among habitat patches can mitigate habitat fragmentation, but expanding habitat patch sizes is necessary in mountain ranges where fragmentation is most intensive. The Minimum Area Requirements (MAR) of a population is the area required for the population’s long-term persistence. In this study, the response of occupancy probability of giant pandas against habitat patch size was studied in five of the six mountain ranges inhabited by giant panda, which cover over 78% of the global distribution of giant panda habitat. The probability of giant panda occurrence was positively associated with habitat patch area, and the observed increase in occupancy probability with patch size was higher than that due to passive sampling alone. These results suggest that the giant panda is an area-sensitive species. Many studies have been conducted on the habitat of giant panda, and most of them focused on habitat selection 36 , 37 , 38 , habitat quality assessment 39 , 40 , 41 and the impacts of human activity 27 , 42 , 43 . However, little work has been done on the effects of habitat patch size on giant panda populations. Giant pandas are generally regarded as an area-sensitive species in previous studies 44 , but empirical evidence is still lacking. This study seeks to understand how giant panda population presence depends on patch size and to test whether the giant panda are area-sensitive. Our study went further, comparing the occurrence-patch size curve with a null response curve, demonstrating that area-sensitivity in giant pandas is real and not just an effect of patch size sampling. The MAR was estimated to be 114.7 km 2 using a logistic regression of habitat patch size on giant panda occupancy. MAR can also be estimated from Population Viability Analyses (PVAs) 19 . A MAR of 156–248 km 2 for giant panda was estimated from its PVA (40 individuals) 60 and home range size (3.9–6.2 km 2 ) 29 . The MAR value estimated from PVA and home range size is higher than that estimated from occupancy patterns and patch area. This discrepancy may arise because giant panda home ranges often overlap 61 , 62 , which will inflate estimates of required space in PVAs. MAR estimated from occupancy patterns are believed to be sensitive to transient dynamics, while MAR estimated using the PVA-based approach enable us to consider time horizon and extinction probability 19 . As a mechanism-driven model, the plausibility of the PVA-based approach is dependent upon the reality of the modeling assumptions and the robustness of model behavior when population and environment parameters cannot be accurately determined, both of which are, often untested. Several habitat patches, each smaller than the MAR but still containing giant panda presence records, were distributed around Patch E. Those records may stem from ‘spillover’ of giant panda from patch E rather than separate viable populations 63 . Therefore, we urge connection of these small habitat patches to Patch E to enhance the persistence of giant panda in the region. Though habitats are fragmented in the DXL and LS Mountains 44 , four habitat patches (Patches G-J) exceed the giant panda’s MAR. However, Veitch 65 suggested that revealing and repairing the original cause of population degrading is the most important prerequisite to successful Nevertheless, isolation should be considered in future studies of giant panda occupancy that explicitly adopt a more mechanistic perspective on giant panda dispersal. We determined the value of MAR based on the response of giant panda occurrence probability against habitat patch size. The landscapes and forest habitats in which giant panda live have changed considerably in the five mountain ranges over the last few decades, and the impacts of these changes on populations may lag behind the physical changes themselves. For example, the variable of slope is strongly correlated with altitude ( r = 0.8469, P < 0.01) at the scale of the five mountain ranges, whereas slope is apparently indispensable for predicting giant panda presences at finer scales. We measured model performance using the area under the receiver-operator curve (AUC) 85 and the True Skill Statistic (TSS) 86 . A value of AUC greater than 0.90 was considered to be excellent, 0.8 to 0.9 good, 0.7 to 0.8 fair, and 0.6 to 0.7 poor 87 . For TSS values, greater than 0.75 was considered to be excellent, 0.4 to 0.75 good, and less than 0.4 poor 88 . The null response curve is defined by the function , where a b (corresponding to the minimum home range size of a giant panda 29 , namely 3.9 km 2 ) and a (a > a b ) is the patch size, q b is the occupancy rate in a patch with size of a b . The null response curve corresponds to the probability of a habitat patch of size a being occupied by giant panda, provided that patch is ecologically and functionally equivalent to independent patches with areas of a b and shows no area-related ecological advantages over the set of small patches.

Abstract The giant panda is an example of a species that has faced extensive historical habitat fragmentation, and anthropogenic disturbance and is assumed to be isolated in numerous subpopulations with limited gene flow between them. To investigate the population size, health, and connectivity of pandas in a key habitat area, we noninvasively collected a total of 539 fresh wild giant panda fecal samples for DNA extraction within Wolong Nature Reserve, Sichuan, China. Seven validated tetra‐microsatellite markers were used to analyze each sample, and a total of 142 unique genotypes were identified. Nonspatial and spatial capture–recapture models estimated the population size of the reserve at 164 and 137 individuals (95% confidence intervals 153–175 and 115–163), respectively. Relatively high levels of genetic variation and low levels of inbreeding were estimated, indicating adequate genetic diversity. Surprisingly, no significant genetic boundaries were found within the population despite the national road G350 that bisects the reserve, which is also bordered with patches of development and agricultural land. We attribute this to high rates of migration, with four giant panda road‐crossing events confirmed within a year based on repeated captures of individuals. This likely means that giant panda populations within mountain ranges are better connected than previously thought. Increased development and tourism traffic in the area and throughout the current panda distribution pose a threat of increasing population isolation, however. Maintaining and restoring adequate habitat corridors for dispersal is thus a vital step for preserving the levels of gene flow seen in our analysis and the continued conservation of the giant panda meta‐population in both Wolong and throughout their current range. Nonspatial and spatial capture–recapture models estimated the population size of the reserve at 164 and 137 individuals (95% confidence intervals 153–175 and 115–163), respectively. Relatively high levels of genetic variation and low levels of inbreeding were estimated, indicating adequate genetic diversity. Surprisingly, no significant genetic boundaries were found within the population despite the national road G350 that bisects the reserve, which is also bordered with patches of development and agricultural land. We attribute this to high rates of migration, with four giant panda road‐crossing events confirmed within a year based on repeated captures of individuals. We used the identification of different individuals through the fecal genetics data to build a CMR model and estimate the giant panda population size in Wolong using the “CAPWIRE” package (Pennell, Stansbury, Waits, & Miller, 2013 ) in the R programming environment. CAPWIRE performs population size estimation as well as or better than other abundance estimators when the data contain multiple observations of an individual within a session and there are <200 individuals (Miller, Joyce, & Waits, 2005 ; Mumma et al., 2015 ). Because our fecal collection efforts focused on all the giant panda's suitable habitat, we inferred that recapture probability was even among all individuals. We thus ran models under the assumption of equal capture (ECM) probabilities in CAPWIRE to estimate the population size. Because our study population was not closed during the study period and there was likely migration across the Northern and Southern borders, we also used the R package “secr” to employ spatially explicit capture–recapture (SECR) methods to estimate a density of pandas per square km across our study area (Efford, 2013 ). We used the polygon We then multiplied the estimated density by the sample area to get an estimate of the number of pandas in Wolong. 2.5. Population genetics analysis The number of alleles ( A ), observed heterozygosity (Ho), expected heterozygosity (He) and polymorphic information content (PIC) were calculated at individual loci and across loci using the software CERVUS 3.0 (Marshall, Slate, Kruuk, & Pemberton, 1998 ). A Bayesian clustering method implemented in Structure 2.3.1 (Pritchard et al., 2000 ) was used to determine the most likely number of genetic clusters. DISCUSSION Previous capture/recapture studies of giant pandas that have used genetic markers have resulted in population size estimates that have exceeded those of other methods, with Zhan et al. ( 2006 ) estimating nearly double the population size in Wanglang Nature Reserve compare to the 3rd national survey. These methods have been criticized in the past for potential violations of CMR model assumptions, including population closure and genotyping error (Garshelis et al., 2008 ). Our CMR model estimate of giant panda population size within Wolong was also larger than results from the latest national survey, though not as drastically so (slightly over 50%). The potential for genotyping errors was explicitly addressed in our analysis (see 2.3), but open borders could be a source of bias in the estimation of population size. Because of this, we used SECR methods to model the density of pandas/km 2 and estimated a total population of 137 individuals. Because SECR estimates density in a spatially explicit manner and thus avoids the assumption of spatial population closure, this is likely a more accurate estimate of the concurrent number of pandas residing in Wolong at a given time. These results are in agreement with previous genomic inbreeding and relatedness metrics calculated using SNP markers from the whole panda genome: Pandas in larger populations like those in the Qionglai Mountains have relatively low levels of inbreeding compared to other mountain ranges (Garbe, Prakapenka, Tan, & Da, 2016 ). Our findings that the pandas were able to cross the national road bisecting Wolong, and that this road has not resulted in genetic differentiation between the populations on either side, differ from road effects found in the Xiangling Mountains. Zhu et al.