生态学报2025，Vol.45Issue(24)：12306-12323,18.DOI:10.20103/j.stxb.202503270704
基于可解释性机器学习的四川生境质量时空演变特征及其驱动因子
Spatiotemporal evolution characteristics and critical driving factor threshold analysis of habitat quality in Sichuan province based on SHAP-interpretable machine learning
摘要
Abstract
Habitat quality constituted the fundamental basis for maintaining ecosystem stability and healthy functioning.Revealing its spatiotemporal evolution patterns,key influencing mechanisms,and critical thresholds was essential for achieving regional sustainable development.Taking Sichuan Province as a case study,this research quantitatively assessed land use pattern changes(2000-2020),spatiotemporal differentiation characteristics of habitat quality,and critical influencing factors with their thresholds by integrating the InVEST model with XGBoost(eXtreme Gradient Boosting)-SHAP(Shapley Additive Explanations)coupling methodology.The results indicated that:(1)From 2000 to 2020,Sichuan Province experienced an expansion of construction land and forested areas alongside declines in cultivated land and grassland,with negligible changes in unused land and water bodies.The mean habitat quality showed a slight decrease,forming a"west-high-east-low"spatial pattern,with Chengdu as the core demonstrating outward radiative improvement.(2)Population density,elevation,NDVI(Normalized Difference Vegetation Index),annual mean temperature,and slope emerged as dominant factors,where population density and elevation exerted primary control—the former surpassing the latter in contribution magnitude.(3)Habitat quality-factor relationships displayed nonlinear dynamics:positive correlations with NDVI and slope but negative correlations with annual mean temperature and road proximity.Threshold effects manifested as single-segment(slope gradient)and multi-segment(elevation,annual mean temperature)patterns.Proposed management measures leveraging dominant factors and thresholds could optimize governance efficiency under constrained investments.These findings provided scientific support for Sichuan's high-quality ecological development.关键词
生境质量/驱动因子/时空特征/XGBoost-SHAP/四川省Key words
habitat quality/driving factors/spatiotemporal characteristics/eXtreme Gradient Boosting algorithm-Shapley Additive exPlanations/Sichuan Province引用本文复制引用
QIU Dae,ZHANG Junyi,QI Kelu,YANG Xiaoxue..基于可解释性机器学习的四川生境质量时空演变特征及其驱动因子[J].生态学报,2025,45(24):12306-12323,18.基金项目
国家社会科学基金一般项目(23BJY156) （23BJY156）
地理与旅游学院学生创新研究项目(DL25YJSZD01) （DL25YJSZD01）
