Exploring the impact of socio-ecological drivers on ecosystem services (ESs) is critical to ES conservation and restoration. Although a considerable amount of the literature has focused on this topic, few studies have investigated the non-linear impact thresholds of socio-ecological drivers on ESs from a global perspective. In this study, multisource geospatial data was integrated with ecological modeling to quantify six typical ESs in Wuhan, China. Dominant drivers were identified through random forests, and non-linear relationships and thresholds were analyzed by partial dependence analysis. The results revealed that elevator, distance from rivers, soil organic carbon content, aggregation index, and Shannon diversity index were the dominant drivers of most ESs. Moreover, three types of non-linear impact thresholds exist in the relationship between ESs and their socio-ecological drivers: “single threshold” effects; “monotonic impact” effects; and “complex curve” effects, including “S-shape”, “inverted U-shape” and “inverted S-shape” effects. Based on these findings, we proposed policy guidance to inform ecological protection and restoration aimed at enhancing ES provision and promoting sustainable development.

Understanding the mechanisms that drive ecosystem services (ESs) is critical to reversing the trend of ESs deterioration and allocating targeted restoration measures. However, previous studies have not utilized machine learning methods to explore the impacts and thresholds of social-ecological drivers on ESs from a global perspective. Therefore, in this study, we employed a random forest model to identify the dominant drivers of ESs. We then utilized partial dependence analysis to investigate the non-linear impacts and thresholds of these dominant drivers on ESs. The results revealed that elevator, distance from rivers, soil organic carbon content, aggregation index, and Shannon diversity index were the dominant drivers of most ESs. Moreover, three types of non-linear impacts existed in the relationship between ESs and their social and ecological drivers: the majority of drivers exhibited “single threshold” effects on ESs, the minority showed “monotonic change” effects, and the rest had “complex curve” effects in the forms of “S-shape”, “inverted U-shape” and “inverted S-shape” effects, etc. Based on these findings, we proposed policy guidance to inform ecological protection and restoration aimed at enhancing ESs provision and promoting sustainable development.