Examining the rationality of Giant Panda National Park's zoning designations and management measures for habitat conservation: Insights from interpretable machine learning methods

Abstract

Examining the rationality of zoning designations and management measures in the initial establishment of national parks in China is of great significance for supporting decision-making regarding habitat conservation. There exists a research gap in exploring the threshold effects of both environmental and human-related factors on habitat distribution in the context of national parks. However, it may be a challenge because of the limited species distribution data. Our study aims to put forward an analytical framework that integrates species distribution models (SDMs) with interpretable machine learning methods. A case study was performed in the Sichuan region of the Giant Panda National Park (GPNP). We constructed a SDM based on the Random Forest algorithm and made use of accessible remote sensing and big data to predict the distribution of giant panda habitat (GPH) in 2020. Interpretable machine learning methods, namely Partial dependence plots (PDPs) and SHapley Additive exPlanations (SHAP), were utilized to uncover the underlying mechanisms of environmental and anthropogenic variables influencing the GPH distribution. Through GIS overlay analysis, areas where conflicts between human settlements, transportation infrastructure, and GPH exist were identified. Our findings indicated a potential 28.44 % decrease in GPH from 2014 to 2020. Environmental factors such as temperature, topography, and vegetation type, as well as anthropogenic factors including distance to built-up areas and transportation infrastructure, notably distance to national roads, provincial roads and city arterial roads, influenced the GPH distribution with threshold effects significantly. The overlay analysis revealed escalated conflicts between human settlements, transportation infrastructure, and GPH in 2020 compared to 2014. Currently, the Sichuan region of the GPNP implements two zones: a core protection zone and a general control zone, covering 63.71 % of the GPH, while 36.29 % remains outside the management scope. Drawing from the analysis above, this study provided suggestions for the adjustment of zoning designations and management measures in the GPNP.

Keywords: Habitat conservation; Interpretable machine learning; Management measures; National park; Species distribution model; Zoning designations.