Local Polynomial-Based Flood Frequency Estimator for Mixed Population

Floods are often generated by more than one physical mechanism, e.g., rainfall and snowmelt. Consequently, traditional flood frequency methods that use a single distribution may not adequately describe the observed flood variability. Mixed distribution models have been proposed but they have two major drawbacks when applied to observed data: (1) determining the appropriate number of components or flood mechanisms and (2) identifying the probability distribution to be used for each component. Further, available flood data are often not sufficient for detecting mixture populations. As a result, mixed-distribution models can be difficult to apply in practice. In this paper we present a nonparametric approach based on local polynomial regression for estimating a flood quantile function that is data driven, flexible, and can capture any arbitrary features present in the data, alleviating the drawbacks of the traditional methods. We applied the proposed method to a suite of synthetic data from mixture of conventional distributions and to flood records that exhibit mixed population characteristics from Gila River basin of southeast and central Arizona. It is found that the proposed method provides a better fit to both the synthetic and historical data. Although the proposed method is presented in the context of mixed population flood frequency estimation, the data-driven nature of the method lends itself as a simple, robust, and attractive alternative to traditional flood frequency estimation.