Uncertainty Assessment for Real-Time Stage Forecasting

The Flood Monitoring and Warning Systems (FMWS) operating in real-time represent the main non-structural measures to be actuated for reducing risk in flood-prone regions. During the last years significant efforts have been addressed to improve the FMWS reliability also involving the issue of forecast uncertainty estimation. In fact, a deterministic system provides the user or the decision-maker with an "illusion of certainty" of the forecast quantity. Therefore, it becomes fundamental to quantify the level of uncertainty of the future estimates provided by the forecasting models which does not eliminate the uncertainty about the future flood wave evolution but only reduces it. Quantifying uncertainty enables the authorities to set risk-based criteria for flood warning, furnishes information for making rational decisions and offers potential for additional economic benefits of forecasts to every rational decision maker. In this context, the purpose of this study is to assess the uncertainty level to be coupled with stage values provided by a simple stage forecasting model of Muskingum type, named STAFOM-RCM model. It is currently operative within the FMWS developed for the Upper-Middle Tiber River basin in Central Italy and is based only on flood routing process incorporating a correction procedure based on the `Rating Curve Model', allowing to relate local stage and remote discharge without the need of a flood routing methodology. The model is tested for two river reaches considering a large number of flood events occurred in the last decade assuming as performance evaluation measures the Nash—Sutcliffe efficiency coefficient, the error on peak stage and time to peak and the coefficient of persistence. Moreover, the model uncertainty estimate is addressed by applying one of the most direct technique based on the analysis of the statistics of the forecasting model errors for a significant number of historical recorded events. The forecast stage uncertainty level is expressed in terms of confidence interval (CI) with an associated probability equal to 90% or 95%. The CI is differentiated on the basis of the stage values. Results show that the STAFOM-RCM model is able to provide accurate forecast stage hydrographs both in terms of peak region and stage hydrograph reproduction and that the methodology for CI estimate based on the statistic analysis of the model error can be applied when an extended series of observed variables is available. When the database is limited, as for high stage values, the estimation of the CI width could be wrongly affected. Therefore, particularly for high stage intervals occurring during severe floods the error probabilistic methods for uncertainty estimation should be used with care.