Using a seamless forecast ensemble to force the CTRIP river routing model in order to simulate hydrological drought indicators useful to decision-makers in France.

Seasonal hydrological forecasts have become an essential tool for water resources management, especially in the context of increasing droughts in the 21^st century. As part of the CIPRHES project, the purpose here is to assess the capacity of a hydrological forecast modelling chain to simulate low-water flows over France, in order to extract relevant indicators of hydrological droughts for decision-makers, such as the anticipation, i.e., the start date of a drought event, and the precision, i.e., the lowest observed flow for 10 consecutive days (VCN10). Seamless meteorological forecasts, combining 10-days ECMWF forecasts with 134-days forecasts simulated by the ARPEGE model using the Ensemble Copula Coupling method, are used to force the SURFEX land surface model coupled with the CTRIP river routing model to simulate 144-days river hydrological forecasts. To bring this study into real-time conditions, data assimilation is performed on a 7-days simulation prior to each forecast using the observed discharges at the gauged stations from the CAMELS database, to correct the internal states of the CTRIP model. The results show that data assimilation significantly improves the simulations over the assimilated period, and its persistence (i.e., the duration of the effect of the data assimilation) is over 30 days for the largest rivers but close to 0 days on the smaller ones. This last point leads to a poor effect of data assimilation on the CAMELS database catchments, most of them having a surface lower than 1000 km². However, the modelling chain simulates a good anticipation for 70% of the used stations from the CAMELS database, and a precision deviation closed to 0 for the large majority of the stations. A post-bias correction procedure based on the Empirical Quantile Mapping (EQM) method at each station allows to improve the estimations of these indicators, e.g., good anticipation for 86% of the stations.