Modeling the effect of uncertainties in rainfall characteristics on flash flood warning based on rainfall thresholds

Abstract

This study proposes a risk assessment framework for quantifying the reliability of the rainfall threshold used in flash flood warning, which should be influenced by the uncertainties in the rainfall characteristics, including rainfall duration, depth, and storm pattern. This risk assessment framework incorporates the correlated multivariate Monte Carlo simulation method, the Sobek 1D–2D hydrodynamic model, and a logistic regression equation to establish a quantile relationship of the rainfall threshold for quantifying reliability of the rainfall threshold. The Shuhu Creek catchment locates in East Taiwan, and its historical hourly rainfall records on eight typhoon events are used as the study area and data. The results from the proposed framework indicate that the variation in the rainfall threshold declines with the duration; 12-h duration associated with a stable coefficient of variance of the rainfall threshold appears to be appropriate for the flash flood warning in the Shuhu Creek catchment. Moreover, the issued rainfall thresholds in the Shuhu Creek catchment by Water Resources Agency in Taiwan have a low exceedance probability. This infers that inundation might occur as the observed rainfall depth approaches the threshold, so that it is necessary to lower the rainfall threshold in accordance with higher exceedance probability in order to achieve the goal of early flood warning.