Real time prediction approach for floods caused by failure of natural dams due to overtopping

This paper presents a real time prediction approach for floods caused by failure of natural dams due to overtopping. The approach adopts the observed outflow data of the preceding failure process for calibrating a simulation model, and the calibrated model is then implemented to predict the remaining failure process and flood characteristics. A widely used parametric model of dam failure is adopted in consideration of practicability and computational simplicity. The problems raised by interrelation among the model parameters and impeding the model calibration are analytically identified, and a simple but effective solution method is proposed. The approach was examined through two idealized cases where there exist no model inadequacy and measurement errors. Its effectiveness of and applicability to predicting the peak discharge and the time to peak of various outflows were exhibited. The real world case of Tangjiashan Quake Lake in China was further analyzed. The outflow peak discharge and the time to peak were reasonably predicted with one and a half hours ahead, demonstrating its potential for practical applications. Multistage features of the breach growth in nature may lower its performance due to raising difficulties in the identification of reasonable predictions. Future work of improving the model adequacy and observation accuracy would enhance its applicability to natural environments.

► We develop a real time prediction approach for dam-failure floods due to overtopping. ► We identify and solve the problem resulting from model parameter interrelation. ► Multistage breach growth complicates real time predictions.