Snow leopard population assessment in Qilianshan National Park
Nianfan Ding, Luciano Atzeni, Mingjie Chang, Jian Fan, Yibin Li, Xuehan Hou, Wu Liji, Jucai Yang, Yongjun Se, Wen Pei, Duifang Ma, Kongtai Liao, Dazhi Hu, Juntao Zhang, Kun ShiContext
Conservation efforts should prioritise threatened species that are important to ecosystems and people. The snow leopard (Panthera uncia) is a prime example because it serves as an umbrella and flagship species in high-mountain Asia. However, a major obstacle to its conservation in China, where it has the largest range and population, is the knowledge gap on population parameters and the common sampling bias in existing data.
Aims
This study aimed to assess the population in Qilianshan National Park (QLSNP), a priority landscape for global snow leopard conservation, using a dataset of known bias towards suitable habitat.
Methods
Between 2016 and 2018, camera traps were deployed across five sites in the northern and western QLSNP. We estimated the probability of habitat use and density by using spatial occupancy models and spatial capture–recapture models respectively. To reduce the influence of sampling bias, we excluded bias-sensitive covariates and restricted spatial prediction to avoid over-extrapolation.
Key results
The probability of habitat use increased significantly with ruggedness. Density estimates varied from 0.15 to 0.94 individuals per 100 km2 across different sites, with an overall density of 0.24 individuals per 100 km2 (95% CI 0.12, 0.50) or a population size of 70 (95% CI 34, 143) in our habitat mask in QLSNP.
Conclusions
Our study provided population parameters that can be used as basis for protecting snow leopards in QLSNP. However, the steps taken to reduce sampling bias also limited our ability to model prey effects and to extrapolate results beyond the surveyed area. Additional work is needed to validate the surrogate methods used in the absence of prey covariates.
Implications
Our approach offers a cautious way to address known sampling bias in existing datasets, but its limitations must be recognised. Given the challenges we encountered, we recommend increased investment in high-quality data collection to enable reliable range-wide population assessment.