基于机器学习的长株潭城市群生态敏感性及其驱动因子研究  

Study on ecological sensitivity and driving factors of Changsha-Zhuzhou-Xiangtan urban agglomeration based on machine learning

作　　者：陈艳君[1] 谢炳庚[1] 李俊翰 黄军林 CHEN Yanjun;XIE Binggeng;LI Junhan;HUANG Junlin(School of Geographical Sciences,Hunan Normal University,Changsha 410006,China)

机构地区：[1]湖南师范大学地理科学学院,长沙410006

出　　处：《生态科学》2025年第1期28-39,共12页Ecological Science

基　　金：国家自然科学基金联合基金项目(U19A2051)。

摘　　要：为探究生态敏感性驱动因子的区域差异,以长株潭城市群为研究对象,从自然要素和社会经济要素两个方面共选取12个因子,采用自组织神经网络模型进行生态敏感性聚类,结合随机森林算法与SHAP(Shapley additive explanations)方法,量化因子总体重要性程度,并探究每个生态敏感性类的主要驱动因子及驱动因子间的交互作用关系。研究结果表明:(1)长株潭城市群可分为环境质量敏感类、远离水源敏感类、土壤–林地敏感类、土壤–农田敏感类和人类活动敏感类共5个生态敏感性类型区;PM_(2.5)含量、土壤容重、距河流距离、GDP以及生态环境质量为最重要的5个影响因子。(2)各生态敏感性类的主要驱动因子具有一定差异且存在复杂的交互作用,整体上自然要素对社会经济要素有增强贡献的效果,环境质量敏感类的主要驱动因子为生态环境质量和PM_(2.5)含量;远离水源敏感类主要受制于距河流距离的影响;土壤–林地敏感类与土壤–农田敏感类均受土壤容重和PM_(2.5)含量的影响较大,但土壤–林地敏感类还受距河流距离的影响,土壤–农田敏感类对土壤酸碱度更为敏感;人类活动敏感类主要受社会经济要素影响,驱动因子为PM_(2.5)含量和GDP。研究可为生态敏感性评价与精细化国土空间规划及治理提供借鉴。In order to explore regional differences in driving factors of ecological sensitivity,taking Changsha-Zhuzhou-Xiangtan urban agglomeration as the research object,this paper selected 12 factors from two aspects of natural factors and human factors,by using self-organizing map model for ecological sensitivity clustering,combining with the analysis method of random forest algorithm and SHAP(Shapley additive explanations),quantified factors overall importance degree,and explored the main driving factors and their interaction relationship of each ecological sensitivity type.The results show that:(1)The Changsha-Zhuzhou-Xiangtan urban agglomeration can be divided into five ecological sensitivity types:environmental quality sensitive category,remote water source sensitive category,soil–forest sensitive category,soil–farmland sensitive category and human activity sensitive category.PM_(2.5) content,soil bulk density,distance to river,GDP and ecological environment quality are the most important factors.(2)The main driving factors of each ecological sensitivity category have differences and complex interaction.On the whole,natural factors have enhanced contribution to human factors.The main driving factors of environmental quality sensitive category are ecological environmental quality and PM_(2.5) content.Remote water source sensitive category is mainly influenced by distance from river.The soil–forestland sensitive category and soil–cropland sensitive category are greatly affected by soil bulk density and PM_(2.5) content,but the soil–forestland sensitive category is also affected by distance to river,and the soil–cropland sensitive category is more sensitive to soil PH.Human activity sensitive category is mainly affected by human factors,and the driving factors are PM_(2.5) content and GDP.This study can provide a reference for ecological sensitivity assessment and refined territorial spatial planning and governance.

关 键 词：自组织神经网络 随机森林 SHAP 生态敏感性 驱动因子 

分 类 号：X171.1[环境科学与工程—环境科学]