A flood forecasting method in the Francolí River Basin by using a distributed hydrological model and an analog-based precipitation forecast

The recent flood event in Valencia (Spain) in October 2024 has revealed the need for real-time flood forecasts. Flood forecasts are based on meteorological forecasts that supply the feasible precipitation for the coming hours and a hydrological model to simulate the rainfall-runoff processes in the catchment. Distributed hydrological models require several parameters to simulate basin processes, though estimating their values accurately in each cell remains a challenge. Calibration processes that compare the hydrological model results with observations, in order to identify the best model parameter values, usually have an inherent uncertainty due to errors in the data, initial conditions and the simplified nature of the models. Furthermore, usually there is not a single set of parameter values that can characterise the hydrological response in all flood events. Therefore, the model calibration should consider diverse flood events, to optimize model performance under varying conditions.

This study presents the calibration and application of the Real-time Interactive Basin Simulator (RIBS) distributed hydrological model combined with precipitation forecasts based on the analog method to supply flood forecasts in the Francolí River Basin located in the Catalonia region in Northeast Spain.

First, observed rainfall and streamflow data recorded at gauging stations in the catchment for a set of real flood events have been used to calibrate the RIBS model. Five flood events were identified and used in the hydrological model calibration. The Nash–Sutcliffe Efficiency (NSE) coefficient showed good agreement between observed and simulated hydrographs for some events with values in the range 0.6179-0.9114.

Second, an ensemble of spatially distributed precipitation forecasts were used as input data to the calibrated hydrological model in the Francolí catchment. A set of five past events were used. A set of 10 meteorological analogs associated with the flood event was generated for each of the events analysed. The search for meteorological analogs was conducted using the 500 hPa and 1000 hPa geopotential height fields as predictors, and the similarity metric was based on a combination of Euclidean distance and Pearson spatial correlation. The generated set of analogs for each event can be used as an ensemble for generating a probabilistic precipitation field forecast for the region. The accuracy and reliability of the analog forecasts were assessed comparing the hydrological model outputs with the streamflow and precipitation observations at the gauging stations considered in the study. The best analog for each event obtained a Root Mean Square Error (RMSE) value ranging from 0.894 to 6.344, emphasizing performance variability.

The method proposed supplies a probabilistic flood forecast at the Francolí catchment outlet. This method improves the knowledge about the hydrological catchment response in flood events, supplying a probabilistic forecast. The method proposed enables more accurate flood predictions that can be used to supply informed response actions.