Unraveling drivers of land use efficiency in rapidly urbanizing areas: A hybrid SBM-DDF and explainable machine learning framework

Abstract

With the acceleration of urbanization in China, improving land use efficiency has become a key priority for sustainable development. This study, based on a multi-source panel dataset covering 35 representative rapidly expanding cities from 2007 to 2022, proposes a hybrid analytical framework that integrates a Slack-Based Measure Directional Distance Function (SBM-DDF) model for measuring urban land use efficiency (ULUE) with an explainable machine learning pipeline combining CatBoost, SHAP, and Generalized Additive Models (GAM) to identify and interpret its key drivers. The results show that ULUE has generally increased over time, with more rapid improvements observed after 2015, particularly in eastern cities. Nevertheless, significant regional disparities persist: the eastern region has the highest average efficiency (0.749), followed by the central (0.737), western (0.727), and northeastern regions (0.691). Some developed cities have seen declines in efficiency, while several less-developed ones are approaching the efficiency frontier. Based on the average SHAP contributions, economic level (38 %), social development (23 %), and environmental conditions (18 %) emerge as the dominant drivers, all exhibiting strong threshold effects. ULUE in eastern cities is mainly driven by economic and industrial growth, while other regions rely more on improved public services and environmental management. Land policies and market openness show a negative impact overall. These findings underscore the need to incorporate non-linear thresholds into policy design, promote regionally differentiated land use strategies, and align spatial planning with both economic development and environmental sustainability goals.