Spatiotemporal Heterogeneity of Ecosystem Service Interactions and Their Driving Factors Across Different Spatial Scales in China’s Coastal Cities: An XGBoost–SHAP Analysis

Accurately understanding the complex interactions among ecosystem services (ESs) and their driving mechanisms across multiple temporal and spatial scales is essential for effective ecological governance and regional sustainable development. To address the limited simultaneous consideration of long-term dynamics, cross-scale differences, and nonlinear driver effects in previous ES studies, this study develops an integrated multi-temporal and multi-scale framework to reveal the spatiotemporal dynamics and scale-dependent patterns of ES interactions and bundles and further identify their key drivers and threshold effects. The InVEST model was used to quantify six categories of ESs, while their trade-offs/synergies and bundles were examined across the 2 km, 5 km, and 10 km grids and county scales from 1990 to 2020. The XGBoost–SHAP model was further employed to identify the key drivers of ESs and their associated thresholds at each spatial scale. The main conclusions are as follows: (1) Six ESs exhibited spatial heterogeneity and broadly consistent declining trends across different spatial scales. From 1990 to 2020, soil retention (SR) showed the largest decrease across the four spatial scales (14.56–14.57%), followed by water yield (WY; 11.20–11.29%) and food production (FP; 7.52%), while landscape aesthetics (LA) declined the least (1.36%). (2) Interaction patterns among ESs were broadly consistent across spatial scales. Synergies were mainly observed among habitat quality (HQ), SR, carbon storage (CS), and LA, with Spearman correlation coefficients generally ranging from 0.49 to 0.92 (mostly p < 0.001), whereas trade-offs were predominantly observed between FP and other ESs. (3) ES bundles varied across spatial scales, with 8, 6, 6, and 5 bundle types identified across the four spatial scales, respectively; however, within a given spatial scale, their spatial distributions remained relatively stable across the four study periods. Transitions among different bundle types were also observed. (4) The relative importance of driving factors varied substantially among ESs, time periods, and spatial scales. The areal proportions of different landscape types were the primary drivers of habitat quality, carbon storage, food production, and landscape aesthetics, whereas soil retention and water yield were mainly influenced by biophysical indicators. The major drivers consistently exhibited relatively stable threshold effects across different temporal and spatial contexts. These findings deepen the understanding of multiscale interactions among ESs and their driving mechanisms and provide a scientific basis and decision-making support for the sustainable development of China’s coastal cities and other coastal regions worldwide.