Coupling high resolution meteorological models with neural networks for flash flood forecasting: implementation on a Southern France basin

Flash floods are an important hazard that particularly affects the Mediterranean region. Flood forecasting using simulation tools adapted to this context is therefore a crucial issue. In exposed regions, the difficulty of measuring and forecasting the spatial variability and intensity of rainfall, as well as the difficulty of identifying processes at the necessary time and space scales, has often led to the use of highly conceptual - or even statistical - models that make few assumptions about hydrological processes. Among these, neural networks have proven their relevance for flash flood forecasting. However, without hydrometeorological coupling, flow forecasting is often limited to the response time of the basin, i.e. a few hours in general. The purpose is to find a way of increasing this lead time, which is often too short for crisis management.

A flood forecasting model for the Gardon de Mialet basin (Southern France) is being developed as part of the HydIA joint laboratory funded by the ANR (French National Research Agency) and the Synapse company, with the aim of developing a range of hydrometeorological forecasting services based on artificial intelligence approaches. The use of gridded observed data, like in a meteorological model, has enabled the neural network model implemented (Multilayer Perceptron) to reduce its sensitivity to support change.

In the absence of rainfall forecasts, performance decreases with the lead time. With perfect forecasts (observed data used as future data), performance remains high for lead times up to 24h. The model has been coupled with two high resolution weather models, AROME and ARPEGE (2.5km and 10km respectively), implemented by Météo-France for short-range numerical weather prediction. The use of forecasts from these meteorological models for the 49 events in the database enables us to identify the error generated by the hydrological model and that generated by the meteorological model, in comparison with perfect forecasts. Analysis of these errors opens operational perspectives for crisis management. It also makes it possible to improve model training based on perfectible forecast data, and to correct rainfall forecasting biases to achieve higher performance.