ML-based Flexible Flood Inundation Mapping and Intercomparison Framework

Abstract:

Flood Inundation Mapping (FIM) is a critical component in the flood forecasting and planning paradigms, translating rainfall and streamflow predictions and observations to flood inundation extent over an impacted area. FIM forecasting combines predictions by a hydrological model, which estimates streamflow and overland runoff, and a hydraulic model which predicts the movement of water over the landscape. The selection of FIM methodology for an operational framework is based on trade-offs between numerical complexities, resolutions, and computational costs, typically resulting in the adoption of a single forecasting model. This can considerably limit the accuracy of FIM predictions across a wide gamut of flooding conditions (e.g. riverine, coastal, urban). Emerging Artificial intelligence and Machine Learning (AI/ML) techniques show great promise for hydrological predictions, particularly those based on the mechanistic underpinning of the relevant processes. AI/ML approaches have not yet been extensively explored for improving FIM. The project’s overarching goal is to advance FIM prediction accuracy and efficiency within large-scale operational frameworks. This project will (1) develop ML-based and ML-supported FIM tools, (2) develop a modular and expandable FIM intercomparison framework based on a large observational database, (3) conduct an intercomparison analysis employing a wide range of FIM solutions over diverse geographic and hydrologic settings, and (4) developing a proof-of-concept AI/ML FIM optimization framework, which employs the most suitable FIM solution(s) based on flood event characteristics. The project’s scientific and operational outcomes have a high transformative potential for the NOAA Office of Water Predictions (OWP) operational FIM framework and the broader scientific and operational communities.