A stochastic event-based approach for flood estimation in catchments with mixed rainfall/snowmelt flood regimes

The estimation of extreme floods is associated with high uncertainty, in part due to the limited length of streamflow records. Traditionally, flood frequency analysis or event-based model using a single design storm have been applied. We propose here an alternative, stochastic event-based modelling approach. The stochastic PQRUT method involves Monte Carlo procedure to simulate different combinations of initial conditions, rainfall and snowmelt, from which a distribution of flood 5 peaks can be constructed. The stochastic PQRUT was applied for 20 small and medium-sized catchments in Norway and the results show good fit to the observations. A sensitivity analysis of the method indicates that the soil saturation level is less important than the rainfall input and the parameters of the PQRUT model for flood peaks with return periods higher than 100 years, and that excluding the snow routine can change the seasonality of the flood peaks. Estimates for the 100- and 1000-year return level based on the stochastic PQRUT model are compared with results for a) statistical frequency analysis, and b) a 10 standard implementation of the event-based PQRUT method. The differences between the estimates can be up to 200% for some catchments, which highlights the uncertainty in these methods