Decoding the impact of urban form on energy efficiency in shrinking cities: dynamic evolution, spatiotemporal transitions, and SHAP analysis

Analyzing the evolutionary patterns and influencing factors of energy efficiency (EE) is of significant importance for ensuring energy security and addressing urban sustainable development. This study, based on data from 159 shrinking cities (SC) in China, employs exploratory spatiotemporal data analysis to investigate the dynamic evolution and spatialtemporal transitions of EE. Additionally, it integrates SHapley Additive exPlanations (SHAP) and Random Forest (RF) to explore the impact mechanisms of urban form (UF) on EE. The findings reveal that the rate of EE improvement initially accelerates and then decelerates, following an “S-shaped growth curve”. Significant regional disparities exist, with the lowest levels observed in northeast China. The transition capability of EE is relatively weak, characterized by strong stability in local spatial transfer directions and the presence of path dependency. Explainable machine learning identifies that UF indicators such as patch distance, patch perimeter-area ratio, patch fragmentation, and total patch area are the primary determinants of EE. Among these, the positive effect of patch distance is dominant, exhibiting a spatial differentiation pattern where its intensity decreases from the southwest to the northeast. This research provides multiple perspectives for optimizing urban spatial construction, reducing energy waste, and enhancing overall efficiency.