Real-Time Updating of Rainfall Threshold Curves for Flood Forecasting

Abstract

The rainfall threshold (RT) method is a nonstructural flood mitigation approach that is emerging as an effective flood forecasting tool. A critical RT value is the minimum cumulative rainfall depth necessary to cause critical water level or discharge at a cross section of a river. The major drawback of the RT approach is associated with the offline methods used for extracting critical RT values based on some fixed watershed characteristics and rainfall conditions. In this paper, a novel methodology is presented for real-time updating of RT curves for flood forecasting using a rainfall-runoff model and an artificial neural network. In this method, in addition to the rainfall depth, observed discharges are also used to update the rainfall threshold curves for real-time soil moisture and rainfall temporal and spatial patterns. The method was tested on the Walnut Gulch watershed with a 50-min time of concentration for selected historical rainfall events. It was shown that applying the proposed updating method can prevent the issuance of false warning, e.g., for the flood of August 2006, and in some cases increase the lead time of flood forecasting, e.g., 20-min increase in lead time for the flood of June 2008. Using data for 14 major historical rain events, it was shown that by applying the updating method, the hit rate is increased by an average of 28% and the false rate is decreased by an average of 51%.