Can we extend flood forecasting lead-time by optimising precipitation forecasting based on analogs ? Application to the Seine river basin

This study presents the application of a precipitation forecasting technique based on analogs to forecast discharges at the Seine river basin. Given a target meteorological situation, the analog approach aims at searching in a meteorological archive a sample of past situations that are similar to the target one. From this sample, it is possible to extract the associated precipitation amounts, and then provide a probabilistic estimation of the expected daily rainfall. Precipitation forecasts were obtained up to eight days in advance through a selection of analogs based on two levels of analogy (method A : pressure fields and humidity) and four levels of analogy (method B : by adding temperature and vertical velocity). These forecasts were introduced in the hydrological model GR3P to derive probabilistic discharge forecasts. Performances were evaluated in a perfect prognosis context (target situation comes from a meteorological reanalysis) and in an operational forecasting context (target situation comes from weather forecasts). Results show a significant improvement of precipitation forecasts when method B is used, with reduced uncertainties and lower false alarm rates. In an operational forecasting context, the highest gain obtained by method B is observed during the two first days of lead-time. Flood cases were studied and results show that it is possible to anticipate a flood event up to one week ahead. The approach here proposed can contribute to the methods currently used by operational flood forecasting services, by providing quantitative precipitation forecasts for longer lead-times.