River instantaneous peak flow estimation using daily flow data and machine-learning-based models

Abstract

Estimation of the design flood flow for hydraulic structures is often performed by adjusting probabilistic models to daily mean flow series. In most cases, this may cause under design of the structure capacity with possible risks of failure because instantaneous peak flows may be considerably larger than the daily averages. As there is often a lack of instantaneous flow data at a given site of interest, the peak flow has to be estimated. This paper develops new machine-learning-based methods to estimate the instantaneous peak flow from mean daily flow data where long daily data series exist but the instantaneous peak data series are short. However, the presented methods cannot be used where only daily flow data are available. Developed methodologies have been successfully applied to series of flow information from different gauging stations in Iran, with important improvements compared to traditional empirical methods available in the literature. Reliable results produced by the machine-learning-based models compared to the traditional methods show the superior ability of these techniques to solve the problem of inadequate measured peak flow data periods, especially in developing countries where it is difficult to find sufficiently long instantaneous peak flow data series.