Abstract: Understanding long-term vegetation dynamics in subtropical mountainous and hilly regions and their mechanistic responses to climate change and human activities is pivotal for assessing regional ecological security and guiding strategic management.Taking Lishui City in Zhejiang Province as a representative case study,this investigation integrated kernel Normalised Difference Vegetation Index(kNDVI)data spanning 2000—2025 with a multi-source suite of climatic,topographical,socio-economic,and infrastructural variables.Methodologically,it synthesised Theil-Sen trend analysis,the Mann-Kendall test,and seasonal dynamics,while concurrently benchmarking multiple linear regression against the XGBoost machine learning algorithm,augmented by SHAP(SHapley Additive exPlanations)for interpretability.This integrated framework was deployed to systematically disentangle the spatiotemporal evolution of vegetation and the relative contributions of its underlying drivers.The results demonstrated that:over the 26-year observational window,the areal mean kNDVI for Lishui City exhibited a statistically significant increasing trajectory(Mann-Kendall τ=0.502,P<0.001).The most conspicuous increases were localised in Qingtian,Jinyun and Suichang counties;nevertheless,marked interannual variability and pronounced spatial heterogeneity persisted across sub-regional scales.At the intra-annual scale,kNDVI followed a unimodal distribution,attaining its peak between June and September and registering its minimum in December.Both multiple regression and SHAP analyses consistently identified population density,settlement density,and per capita GDP as the most potent suppressive drivers of kNDVI.Conversely,slope,minimum temperature and elevation emerged as significant positive facilitators.The XGBoost model substantially outperformed multiple linear regression in terms of goodness-of-fit,demonstrating a markedly superior capacity to capture non-linear ecological interactions and threshold effects inherent to the coupled human-natural system.Collectively,these findings furnish a robust scientific foundation for devising differentiated ecological management strategies and targeted forest conservation initiatives within subtropical mountainous urban agglomerations,thereby supporting the operationalisation of regional carbon neutrality goals.