The core challenges in ecosystem service valuation revolve around the inherent difficulties of quantification, limitations in assessment accuracy, and the complex interplay of ecological components within diverse environments. These issues become especially evident in nature reserves, where intricate habitats and a rich array of species amplify the difficulty of reliably measuring ecosystem service values. Such complexities also hinder the straightforward integration of Gross Ecosystem Product（GEP）into national economic accounting frameworks, thereby curtailing its role in ecological compensation initiatives and market—driven environmental trading mechanisms. In this study, Jinyun Mountain National Nature Reserve was selected as a representative site to design a systematic, on—site accounting indicator system capable of precisely calculating GEP. By employing the Integrated Valuation of Ecosystem Services and Tradeoffs InVEST model, remote sensing data were rigorously processed and analyzed, thus enabling the derivation of the Ecosystem Service Index（ESI）. The subsequent steps entailed a thorough examination of critical trade—offs, synergies, and underlying drivers that shape various ecosystem services. To refine the initial estimations and mitigate potential biases, the research incorporated both Geographically Weighted Regression（GWR）and Random Forest（RF）models as part of a robust calibration and optimization procedure. Several pivotal findings emerged from these analyses. First, the total GEP of Jinyun Mountain National Nature Reserve was calculated to be 2.841 billion CNY, with the ESI displaying consistently high and stable values between 1984 and 2024. Notably, 81% of forest ecosystems demonstrated pronounced synergistic relationships across multiple services. Second, spatial cluster analysis revealed four distinct service clusters and three functional zones, making it evident that elevated road density and larger population density negatively influenced overall ecosystem service quality. Third, initial biases in the GEP and ESI estimates encompassed 48.79% of the reserve’s area, and the RF model substantially surpassed the GWR model by achieving a 61.56% reduction in bias. Through the integration of detailed field measurements and advanced modeling methodologies, this work advances a cohesive framework for evaluating ecosystem services in mountain—urban composite（KMCs）nature reserves. The study underscores the importance of uniting remote sensing data with reliable statistical instruments to bolster the precision and consistency of ecosystem service evaluations. High-caliber input data prove indispensable for producing accurate model outputs, particularly when investigating heterogeneous terrains. The study provide a more nuanced perspective on quantifying ecological service capacity and offer concrete recommendations for refining ecosystem service assessments in complex environmental contexts, thereby guiding targeted policy decisions and conservation measures.