Threshold-oriented ecosystem-service relationships from interpretable machine learning to spatial management: A case from Shandong Province, China

Ecosystem-service relationships (ESRs) – the synergies and trade-offs that arise when multiple services respond to the same biophysical and human drivers – are decision-relevant outcomes, but most existing studies have treated them only indirectly and with predominantly linear tools, making it hard to expose real nonlinear ecological processes. In this study, six services in Shandong Province (carbon storage, grain providing, soil conservation, water conservation, habitat quality, and recreation service) were pairwise combined into 15 ESRs, and an interpretable “spatial ESR – nonlinear driver inference” framework was built by coupling GWR with transparent machine-learning models. Among the 27 “nature–landscape–human” factors, climate variables (temperature, precipitation, sunshine) together with elevation and nighttime light intensity consistently dominated ESR variation, showing that background environmental energy and human disturbance jointly constrain whether services co-benefit or compete. More importantly, ESR–driver linkages were clearly nonlinear: by superimposing the partial-response curves of the 15 ESRs, we identified probability-oriented optimal intervals – for example, annual mean temperature above about 15.5 °C, precipitation around 484–549 mm, nighttime lights lower than about 1.03, elevation higher than about 90 m, and sunshine between roughly 2379 and 2427 h – within which the likelihood of comprehensive synergy is maximized. Compared with MLR, GAM and GBDT, XGBoost achieved the highest accuracy, confirming that ESRs are better described by high-dimensional, thresholded responses than by global linear specifications. The main contribution of this work is thus to turn mapped ESR patterns into operational, probability-oriented, interval-based guidance that can be directly used for zoning, agricultural upgrading, urban–rural interface control and biodiversity-friendly conservation in fast-developing provinces.