Harnessing Machine learning to quantify ecosystem services in coastal Wetlands: A case study of the Bohai economic Rim

Coastal wetland ecosystems are pivotal for the sustainable economic development of coastal regions. While prior studies have explored interactions between wetlands and human activities across multiple spatiotemporal scales, current ecosystem service (ES) assessments predominantly rely on existing land cover data products with only ∼ 80 % spatial accuracy, insufficient for capturing fine-scale ES spatial variations. This limitation hinders the accurate characterization of complex relationships in wetland environmental science and constrains the reliability of subsequent conservation and management decisions. To address this critical gap, this study aims to improve the accuracy of coastal wetland delineation and ES quantification to ≥ 90 % via an integrated technical method. In this work, the Bohai Economic Rim (BER), a region with intensive interactions between climate change and human activities, was selected as the study case for high-precision coastal wetland mapping and quantification of four key ES types (carbon storage, habitat quality, soil conservation, and water yield). By integrating knowledge-based rules implemented on the Google Earth Engine (GEE) platform with a machine learning algorithm, we identified ES spatial hotspots and cold spots, analyzed trade-off and synergy relationships among the four ESs, and further employed a random forest model to evaluate the relative importance of driving factors (including climate variables and human activity indicators). Results show the total ES index ranges from 0.02 to 0.82 (normalized to 0–1), with all six pairs of ESs exhibiting significant positive correlations, indicating strong synergistic effects. The random forest model reveals that forest proportion and precipitation are the top two influential factors shaping ES spatial variation. This study provides a scalable high-precision ES assessment framework for complex coastal wetland systems with accuracy improved from ∼ 80 % to 93.76 %, offering valuable scientific insights and targeted policy recommendations for the conservation and sustainable management of coastal wetland ecosystems in the BER.