A PCA-Granger-Random Forest Framework for Assessing Landscape Fragmentation and Economic-Social System Interactions in Three Major River Basins

This study investigates the causal dynamics between landscape fragmentation (LF) and the socioeconomic system (ESS) in three major river basins. Using a comprehensive dataset spanning multiple years, we first constructed composite indices for both LF and ESS through Principal Component Analysis (PCA) with indicator direction correction. We then employed Granger causality tests to examine bidirectional causal relationships, complemented by rolling window analysis to capture temporal variations in causality strength. To further explore the driving mechanisms, we applied machine learning approaches—specifically Random Forest models optimized via Bayesian optimization—coupled with SHAP (SHapley Additive exPlanations) analysis to quantify variable importance and interaction effects. The results reveal significant causal linkages between socioeconomic development and landscape fragmentation, with notable heterogeneity across different land types (forestland, cropland, and grassland) and river basins. Key socioeconomic drivers include GDP, urbanization rate, and nightlight intensity, while landscape metrics such as patch density and edge density significantly influence socioeconomic outcomes. This research provides a robust methodological framework for understanding human-environment interactions and offers valuable insights for sustainable land use planning and ecological conservation policy-making.