Dam-Break Flood Model Uncertainty Assessment: Case Study of Extreme Flooding with Multiple Dam Failures in Gangneung, South Korea

Dam safety programs are informed by numerical model solutions of dam-break flood depth, extent, and timing that are uncertain owing to an imperfect mathematical and numerical representation of system dynamics as well as uncertain model parameters and input data, yet model uncertainty is rarely reported. The most extreme and damaging events of greatest interest are also the most infrequent to occur and, thus, seldom studied and poorly understood from a model uncertainty perspective. Here an extraordinary event is considered that occurred in Gangneung, South Korea, on August 31, 2002, when Typhoon Rusa dropped nearly 90cm of rainfall over a 24h period, causing two dams in tributary valleys to fail only a few moments apart. A two-dimensional (2D) hydrodynamic flood model is developed using the best available data, calibrated, and validated, and uncertainty is systematically examined. Sources of uncertainty include topographic and bathymetric data, breach geometry data, precipitation data, storm surge data, resistance parameters, and structural model errors that represent deficiencies in the flood model formulation. An uncalibrated solution yields a flood height root-mean square error (RMSE) of 0.4m and a flood-extent agreement of 86% when the best available measurements are used as data input, including precipitation. Additionally, the calibration of resistance parameters reduces the flood height RMSE to 0.33m, which is close to the estimated uncertainty in flood height measurements (0.2m) and RMSE of the best available topographic data (0.2m). Results show that many factors contribute to solution uncertainty and in different ways depending on whether the focus is flood height, flood extent, or flood discharge and timing, and strategies to reduce uncertainties are presented. (C) 2016 American Society of Civil Engineers.