Pitfalls of Validating Extreme Event Flood Models: The experience of RIFT for Hurricane Florence

Despite recent advancements in flood observation capabilities, validating flood models for large-scale extreme events remains a significant challenge. Common validation data sources include in-situ hydrometric measurements and remotely sensed water extent. Recently, social media and other crowdsourced information emerge as a new body of flood observation data. However, each dataset suffers from its unique sparsity and uncertainty. As a result, flood model validation is either biased or inconclusive. Furthermore, the lack of robust validation erodes confidence in using model predictions to support situation awareness during extreme events. The Rapid Infrastructure Flood Tool (RIFT) has been extensively used to support extreme flood event analysis and provide situational awareness. Most recently, RIFT was used to produce flood forecasts during Hurricane Florence 2018 to support multiple federal, state, and local stakeholders in response and recovery. A broad spectrum of reference data was then collected to support model validation, including both traditional and non-traditional observations. RIFT results were compared against reference in water extent and water surface elevation. Moreover, reference data were cross-compared to infer their data quality. In general, RIFT performed well in covering reference flood locations but overestimated water surface elevation. Using this case study, we will highlight the gaps in observational information, pose suggestions on best practices using currently available data, and identify strategies towards more optimal solutions for extreme event flood model validation.