A Multi-Source Data Fusion Framework for Sustainable Drunk-Driving Risk Governance in Megacities: Evidence from Shanghai

Alcohol-impaired driving in megacities is both a road safety problem and a public safety governance challenge. This study proposes a multi-source data fusion framework for identifying the spatio-temporal risk of publicly recorded drunk-driving violations and translating risk estimates into enforcement-oriented priority areas. Using Shanghai as a case study, we integrate 1814 publicly recorded drunk-driving violations with dining and entertainment POIs, 400–800 m annular POI measures, road networks, metro stations, public parking facilities, meteorological conditions, and temporal attributes. Kernel density estimation and ST-DBSCAN are first used to identify observed spatio-temporal clusters. Time-specific grid-based Logit models are then estimated for daytime, evening, and post-midnight periods, with complementary log-log models used for robustness checks. The results show strong nighttime concentration, with T2 showing the highest predicted risk and weekend scenarios further increasing risks in T2 and T3. In-grid dining and entertainment POI variables are mostly negatively associated with risk, whereas surrounding dining POIs show stable positive associations, indicating a spatial mismatch between drinking venues and road spaces where violations are observed. Road exposure, metro accessibility, parking facilities, and weekend context also display time-varying effects. Finally, integrating model-predicted risk grids with observed clusters produces Priority, Warning, and Background areas, supporting evidence-based, time-specific, and location-specific drunk-driving risk governance in megacities.