Ecosystem service interactions and their driving factors based on a geospatially explainable framework: A case study in the Yangtze River Basin, China

Understanding the interactions (synergies and trade-offs) among ecosystem services (ESs) and their driving factors is crucial for sustainable ecosystem management under intensifying climate change and anthropogenic disturbances. In recent years, machine learning approaches have demonstrated strong potential in capturing nonlinear relationships and exploring the driving mechanisms of ES interactions. However, most existing studies provide unified explanations at the global scale and often overlook the spatial heterogeneity and spatial dependence inherent in geographic locations, thereby limiting the ability to reveal the differentiated effects of the same driving factors on ES synergies and trade-offs across regions. This gap becomes particularly critical in large river basins, where pronounced environmental gradients, spatial connectivity, and heterogeneous human activities jointly drive strong spatial differentiation in ecosystem processes and services.

In this study, we develop a geospatially explainable machine learning framework to more explicitly characterize the spatial variability of ES interactions and their formation mechanisms in the Yangtze River Basin, China. Specifically, six key ESs, including food supply (FS), water yield (WY), water purification (WP), soil conservation (SC), carbon sequestration (CS), and habitat quality (HQ), were quantitatively assessed for the period from 2000 to 2023. Spearman correlation analysis and geographically weighted regression (GWR) were then employed to identify the ES relationships and their spatial distribution patterns. Furthermore, the GeoShapley method was introduced to incorporate geographic location into the model interpretation process, thereby enhancing the transparency and interpretability of machine learning decisions. From a spatial interaction perspective, this approach enables the analysis and visualization of the differentiated driving effects of climate conditions, topography, land use, and human activities on ES synergies and trade-offs across different spatial locations.

This study shows that the geospatially explainable framework enhances insights into the formation mechanisms of ES interactions and provides scientific support for implementing zoned ecosystem management and targeted regulation strategies under ongoing global environmental change.