Nonlinear Drivers, Lagged Mechanisms, and Spatial Disparities in Logistics Green Innovation Performance: A DLIA Integrated Framework

Against the backdrop of intensifying climate change and stricter carbon-neutrality targets, improving logistics green innovation performance (GIP) is essential for low-carbon transformation. However, existing studies largely emphasize contemporaneous linear effects and insufficiently address nonlinear interactions, lagged responses, and regional heterogeneity. This study develops a DLIA framework that integrates driver screening, lag-response diagnosis, integrated learning validation, and SHAP-based attribution. Using panel data from 30 Chinese provinces over 2011–2024, logistics GIP is measured with the Super-SBM model, while complementary correlation diagnostics identify significant drivers and their optimal lag structures. Four ensemble-learning algorithms are then compared, with XGBoost selected for explainable attribution analysis. Results show that national logistics GIP increased from 0.405 to 0.467, although pronounced spatial disparities persist and high-performance provinces remain concentrated in eastern China. Twenty-seven variables jointly influence GIP through nonlinear relationships. Highly qualified talent produces the fastest response, with an average lag of 0.60 years, whereas capital stock requires a longer accumulation period of 5.29 years. SHAP results identify economic concentration as the largest contributor (14.0%), followed by highly qualified talent (9.6%) and capital stock (8.4%). These drivers also display threshold effects and substantial provincial heterogeneity. The findings extend innovation-ecosystem research by demonstrating that logistics green innovation depends on nonlinear interactions, differentiated temporal transmission, and regional absorptive capacity. Green logistics policies should therefore shift from uniform linear interventions toward time-sensitive and place-based strategies aligned with local factor endowments and knowledge capacities.