The urban greening paradox: Measuring day-night perceived safety divergences from paired street view images using generative AI

Whether nighttime is the better half of life varies across cities depending on how safe residents perceive the neighborhood environment. Although street view imagery (SVI) is instrumental in auditing perceived safety, city-wide nighttime SVI does not exist. Consequently, the extent to which day-night safety perceptions diverge remains unclear. Although emerging studies use day-to-night (D2N) translations for nighttime SVI, the model convergence and transferability are unknown. Using paired day-night SVIs from U.S. and Chinese cities, we confirmed the converging size (~2,000 pairs) and the cross-cultural transferability of the D2N model. Medium-density SVIs facilitated cross-density transferability, and US-trained models outperformed Chinese-trained models, suggesting moderate limitations in the training datasets when applying D2N models across cultures and countries. Interpretable modeling reveals a clear day–night asymmetry in perceptual mechanisms, with daytime safety perception dominated by pedestrian-oriented configuration, whereas nighttime safety perception relies more on visibility-related cues. Brightness emerges as a key positive predictor in both periods. Greening-related features contribute disproportionately to day–night divergence, amplifying differences between daytime and nighttime safety perceptions rather than uniformly enhancing perceived safety. This pattern reflects urban greening paradox, whereby tree and plant support perceived safety under daylight yet are associated with greater instability after dark. City-scale mapping in Boston reveals pronounced spatial heterogeneity, with high-perception central and north-west corridors remaining relatively safer after dark, while south areas experience the most significant nighttime declines. Our scalable GenAI framework extends urban scene studies to nighttime and enables city-wide mapping of perceived safety after sunset, informing more inclusive nighttime urban planning. • GenAI training converges with ~2,000 paired day–night street view images • Pedestrian-oriented cues shape daytime safety perception • Visibility cues shape nighttime safety perception • Greening enhances daytime safety perception but weakens it after dark • Urban greening contributes most to day–night perception divergence