Citizen flood imagery enhances urban aftermath monitoring and exposes reporting inequities

Urban flood management requires street-level situational awareness, yet existing monitoring remains limited by overhead sensing constraints and socially uneven reporting. Here we show that citizen-generated flood imagery can complement official data by providing localized and interpretable evidence of flood impacts. We develop a two-stage framework that turns social media flood images into spatially explicit aftermath information. First, a flood-masked strategy improves street-level localization by focusing on stable urban features rather than temporary changes caused by flooding. Second, a fine-tuned vision-language model converts the images into standardized measures of flood aftermath. Applied in San Francisco, the framework shows strong agreement with institutional reporting in dense urban areas while revealing additional hotspots that deviate from official records. Furthermore, institutional reporting intensity declines with increasing social vulnerability, whereas social media signals do not show the same monotonic decline. These results demonstrate the potential of citizen-generated imagery to broaden flood visibility and reduce informational gaps in urban aftermath monitoring. Citizen-generated flood images complement official records by revealing localized flood aftermath and additional hotspots, improving visibility in socially vulnerable areas according to flood-masked geolocation and a fine-tuned vision-language model applied to social media imagery.