Regional Estimation of Floods for Ungauged Sites Using Partial Duration Series and Scaling Approach

One of the main obstacles in making reliable predictions of extreme events is the apparent shortness of the time series available in hydrology. There is a common advantage of both regional flood-frequency analysis and partial duration series (PDS) in which both of them allow a reduction of uncertainty by introducing more data to the site of interest. Therefore, a new regional index flood method for ungauged sites based on the PDS model is presented. The PDS model considered in this case assumes a Poisson-distributed number of threshold exceedances and generalized Pareto-distributed peak magnitudes. A new objective approach for the selection of the threshold in the context of regionalization is introduced. This approach estimates a range of reasonable thresholds (or an average annual number of events) for every site. Consequently, the regional average annual number of events can be determined as a common value for all sites in the homogeneous region. The delineation of hydrologically homogeneous regions is determined using the scaling approach. The feasibility of the proposed method was assessed using the available daily flow series from 57 watersheds in Quebec (Canada). Furthermore, the new method is compared with two existing methods for regionalization: the region of influence and canonical correlation analysis methods. Results of the numerical application indicate that the quantile estimates obtained from the new method provide the best values of the performance criteria (e.g.,root-mean-square error). Hence, the new method not only eliminates subjective decisions but also greatly improves the predicted floods for ungauged sites. (C) 2016 American Society of Civil Engineers.