Performance analysis of real Time flood forecasting models for different forecast horizon using optimization algorithms

Hydrological models are often used as real-time flood forecasting tools. When used in this forecasting mode, these models rely on real-time observed rainfall and runoff data. Their parameters can be calibrated in simulation mode, but also in forecasting mode, i.e. for a particular forecast horizon, linked to the studied watershed response time. This second option is expected to produce better forecasting results at the calibration horizon than using model parameters calibrated in simulation. However, in operational conditions, forecasters often need to produce forectasts for various lead times (typically ranging from a few hours to a few days). Thus the use of model parameters calibrated for a specific horizon may not produce optimal results over the whole forecasting time window.