Identifying ecological security patterns across Nanshan National Park and its adjacent counties via an integrated XGBoost-MCR framework
Year: 2026
Type: article
Abstract: National parks anchor biodiversity conservation, but ecological connectivity in mountainous regions also depends on surrounding county landscapes, where land-use change, transport infrastructure, and landscape fragmentation may constrain cross-boundary ecological processes. Existing ecological security pattern studies are often confined to protected-area boundaries, while resistance surfaces based on expert scoring or linear weighted overlay may inadequately represent nonlinear relationships between ecological resistance and multiple environmental and anthropogenic factors. We treated the four-county analytical extent centered on Nanshan National Park in Hunan Province, China, as an integrated park–county study area and developed an ecological indicator-based extreme gradient boosting–minimum cumulative resistance (XGBoost–MCR) framework. Multi-source spatial datasets were harmonized to a common 30 m grid, from which 12 natural and anthropogenic resistance indicators were derived. Ecological-source areas were used to define the relatively low-resistance sampling domain, whereas areas in the very-low integrated ecosystem-service importance class were used to define the relatively high-resistance domain for seven binary classifiers. XGBoost probabilities were used as normalized relative ecological resistance. Resistance-derived relative ecological security was subsequently calculated as E S i = 1 − R i and classified into five levels, while the resistance surface served as the MCR cost layer for corridor extraction and subsequent breakpoint identification. Among the seven models, XGBoost achieved the highest testing area under the receiver operating characteristic curve (AUC = 0.9985), with a testing accuracy of 0.9837, whereas k-nearest neighbors achieved the highest testing accuracy (0.9865). Because the testing subset was randomly held out rather than spatially blocked, these metrics indicate sample-level discrimination rather than spatially independent predictive performance. Shapley additive explanations identified elevation, distance to roads, and population density as the three most influential predictors, accounting for 31.68%, 27.05%, and 15.24% of the normalized mean absolute SHAP importance, respectively. The framework identified 15 ecological sources covering 1730.45 km 2 (16.54% of the study area), 18 ecological corridors totaling 824.37 km, and 66 merged ecological breakpoints, of which 41 were located outside the national park. The resulting ecological security pattern comprised one core, four zones, four axes, and multiple nodes. By avoiding expert-assigned indicator weights in resistance-surface construction, representing nonlinear resistance relationships, and quantifying predictor contributions, the framework provides a reproducible and interpretable spatial basis for corridor protection, breakpoint mitigation, and cross-county ecological governance in mountainous national park regions.
Source: Ecological Indicators
Authors: Zhiyang Gao, Zhiqiang Zhang, Ang Zhou
Institution: Hunan University of Technology
Language: English
Subfield: Ecology
Field: Environmental Science
Domain: Physical Sciences
SDG: Life on land
Open Access status: gold
Estimated APC paid: $3,650
Awards: G9000368076, G9000369334
OpenAlex ID: W7213767322