Evaluation Studies of Real-Time Flood Forecasting: A Review of Issues

Producing streamflow predictions in real-time remains a challenging task for hydrologic research and operational forecasting. Early warning systems providing these forecasts are very complex because all the components required for transforming accurately rainfall into runoff are subject to large uncertainty. These systems generally integrate estimates of past and future rainfall with hydrologic modeling, allowing the prediction of flood events across locations of the drainage network. A recurrent problem in the literature that explores the performance of these systems is that most of the studies assessing the skill of these hydrologic predictions are limited to reporting simulation performance at specific sites and particular events. These limitations are due to many factors. One of the more important is the lack of a research environment infrastructure that allows mimicking past conditions of states of the basin system, serving as basis for running hydrologic models that emulate real-time operation. Other limitation is due to the absence of tools that allow exploring the propagation of errors in hydrologic modeling in a spatial and dynamical context. The Iowa Flood Center has built during recent years a data-driven computational modeling infrastructure, in parallel to its real-time modeling system. This infrastructure facilitates the development of offline experiments that allow examining the effect of modifying individual components of the modeling system on their overall performance (e.g. rainfall inputs, model structure, density of drainage network). This study summarizes the results of a collection of experiments that helped IFC to assess the skill of their hydrologic forecasts, and the contribution of individual components to the associated uncertainty. The authors discuss examples of research experiments focused on improving the routing component of the model. Finally, they illustrate a novel method that shows to provide hydrologic evaluation in a spatial context that accounts for the connectivity of the drainage network.