A terrain-aware approach for image-based urban flood monitoring

Urban nuisance flooding is widespread, yet quantitative observations of its magnitude and spatial variability remain limited. Most image-based approaches provide only relative measures of flooded image fraction, while quantitative methods require surveyed ground control and three-dimensional reconstruction. We introduce a terrain aware, perspective weighted framework that converts flooded image fractions directly into estimates of water level and inundation area using only topography and camera geometry. Our approach combines high resolution digital terrain models with a simple depression-based flood fill model to generate camera specific reference curves that relate water level to a perspective corrected flooding index. Applied to more than 350,000 images from eight cameras deployed across a community in Cahokia Heights, Illinois, the method shows strong agreement between modeled and observed flooding indices, allowing quantitative reconstruction of multi-year flood dynamics. Estimated maximum water levels range from 25 to 86 centimeters, and maximum flooded areas vary by more than an order of magnitude across sites, revealing pronounced neighborhood scale variability in flooding patterns. Our method provides a proof of concept for scalable, low cost, image-based approaches for distributed urban flood monitoring.