A ‘Peak‐Box’ approach for supporting interpretation and verification of operational ensemble peak‐flow forecasts

Abstract Uncertainty communication is an important topic within hydrological forecasting. Hydrological ensemble prediction systems are established tools for the generation of forecasts including uncertainty information. The interpretation of such forecasts requires new visualization and verification tools to help forecasters and end‐users in their decision making. While the visualization of hydrographs is important for estimating flood volumes, little support is provided for the interpretation of peak‐flow forecasts. We introduce the ‘Peak‐Box’, a novel visual support that envelops all ensemble peak timings and peak discharge, from which specific verification metrics are defined. A user‐defined metric quantifies the sharpness concerning peak timing and peak discharge and allows communicating a‐priori, if the spread of peak forecasts is acceptable. Eighteen months of ensemble predictions for four basins have been evaluated. A probabilistic verification which relies on the relationships between the spread and the relative operating characteristic area indicates the quality of the ensemble predictions in all basins. A sub‐sample of 485 events was used for exploring the value of the ‘Peak‐Box’. The combined spread of forecast peak time and peak discharge is higher than a tailored reference sharpness for most of the considered events. We show that, depending on the lead time of the forecasts, 30% to 55% of the observed peaks are found outside the predicted range. Most correct forecasts (hits) were obtained for forecasts having a lead time of 2 or more days. Further analyses indicate that the median of the ensemble peak forecast provides reliable estimates on either peak timing or peak discharge in more than 80% of the events evaluated. Finally, a score system was defined in order to combine different verification measures and obtain an overall assessment on the quality of both the peak discharge and peak ‐timing of the ensemble forecasts. We demonstrate that the ‘Peak‐Box’ can be adopted in different ways in order to obtain quantitative and qualitative insights on the quality of peak forecasts. Copyright © 2012 John Wiley & Sons, Ltd.