Improvement of early warning systems for flood risk with a distributed hydrological model and an analog-based precipitation forecast

The Spanish Mediterranean region is frequently impacted by flash floods, driven by intense convective rainfall, the presence of torrential basins, and dense urbanization in flood-prone areas. This situation can be aggravated by climate change, as demonstrated in recent studies. In this context, one way to reduce risk is to decrease vulnerability by improving both early warning systems and public preparedness. The recently completed Next Generation Flood2Now project aimed to address these two points through an interdisciplinary approach that brought together experts in hydrology, meteorology, sociology, databases, and complex systems, complemented by the participation of a private company. Two basins that suffer frequent flooding were chosen for the project: the torrential basin of the Francolí River, which can be dry in some reaches and times of the year but often experiences catastrophic flash floods as a result of intense rainfall, and the Arga River basin, characterized by a permanent flow produced by snowmelt and rainfall.

The ultimate goal of the project was to develop a hydrometeorological prediction chain that can be used operationally to aid decision-making in the face of potential floods. This was achieved using the INUNGAMA and PIRAGUA_flood (Llasat et al., 2024) databases, which contain all recorded floods in Catalonia and the Pyrenees, respectively, between 1980 and 2020. Based on these databases, an analogue model was developed (Guzzón et al., 2025) that considers the geopotential fields of 1000 and 500 hPa and weather types classified according to the method of Beck et al. (2007). In parallel, the RIBS hydrological model (Garrote and Bras, 1995) was calibrated at a set of points in the respective basins, where streamflow data recorded at gauging station are available, and subsequently fed with rainfall fields corresponding to flood events and their analogues, generating a probabilistic flow output that allows estimating the uncertainty of a given flood event causing damage (Carril-Rojas et al., 2025). For this purpose, flood compensation payments were taken into account, based on information from the Insurance Compensation Consortium. The data generated by the analogues, as well as the predictions obtained from the GFS, constituted the input for the Delft-FEWS platform (https://oss.deltares.nl/web/delft-fews/), which generates a set of flow rates and an exceedance warning as output. To update the flood databases, an AI-based methodology was created that extracts information from the press, analyzes it, and inputs it into the database. At the same time, citizen science campaigns, workshops and exhibitions have been developed, both to raise awareness among the population in both basins and to obtain more real-time observations of the river level. The contribution presented here shows the methodological synthesis of the project and the main results.

Carril-Rojas, et al., 2025. A Flood Forecasting Method in the Francolí River Basin (Spain) Using a Distributed Hydrological Model and an Analog-Based Precipitation Forecast. Hydrology, https://doi.org/10.3390/hydrology12080220

Garrote, L.; Bras, R.L., 1995. A Distributed Model for Real-Time Flood Forecasting Using Digital Elevation Models. JoH, https://doi.org/10.1016/0022-1694(94)02592-Y

Llasat, M.C., et al., 2024. Floods in the Pyrenees: A global view through a regional database, NHESS, https://doi.org/10.5194/nhess-24-3423-2024