Streamflow assimilation for improving ensemble streamflow forecasts G. Thirel (1), E. Martin (1), J.-F. Mahfouf (1), S. Massart (2), S. Ricci (2), F. Regimbeau (3), F. Habets (4). (1)CNRM-GAME, Météo-France, CNRS, GMME, France, (2)CERFACS, France, (3)Direction de la Climatologie, Météo-France, France, (4)UMR SISYPHE, UPMC, ENSMP, CNRS, Paris, France +33 (0) )

Introduction 2 ensemble streamflow prediction systems (ESPS) at a short- and mid-term range at Météo-France –Based on the distributed hydrometeorological model SIM –ECMWF-based ESPS (10-day range, 1.5°, 51 members) –PEARP-based ESPS (60-h range, 0.25°, 11 members) Need to improve the initial states by an assimilation system First validation of the ESPSs against streamflows observations

ISBA Physiographic data for soil and vegetation + MODCOU Qr Qi E H G Aquifer Daily Streamflow Surface scheme Snow SAFRAN Observations + NWP models Precipitation, temperature, humidity, wind, radiations Hydrological model Poor Weak to moderate Good Nash Habets et al. (2008) Meteorological analysis The SIM hydro-meteorological model

The SIM based ESPS Observations Meteor. models ANALYSIS RUN (daily) SAFRAN 10-year climatology Wind, Rad., Humidity SOIL WAT. TABLES RIVERS FINAL STATE ECMWF/PEARP Ensemble forecasts 51/11 members, 11/2-day forecasts ENSEMBLE FORECASTS T+ Precip Spatial DESAGGREGATION ISBA MODCOU ENSEMBLE FORECAST SOIL WAT. TABLES RIVERS FINAL STATES ISBA MODCOU SOIL WAT. TABLES RIVERS STATE Initial states of ESPS : need for improvement Adjusted by BLUE

Strategy 186 stations assimilated over France –Low human influence –Good quality of observations –Not too bad results given by SIM Aim : to use observed streamflow in order to improve streamflow simulation, by adjusting the ISBA soil moisture

The BLUE equations Analysed state Background state Innovation vector Jacobian H : H determines the sensitivity of streamflows to soil moisture variations Hypothesis : linearity of the model -> H is computed with SIM runs initialized by perturbed soil moisture states (perturbation around 0.1%) Observed streamflows streamflows x : control variable

Experiments (10 March 2005 / 30 September 2006, 186 stations) 6 experiments : 3 variable states * 2 physics of the model Daily assimilation, daily observations

Jacobian matrix filling 3 gauging stations Q1, Q2 et Q3. w1, w2 et w3 moderated sums of soil moistures on the basins Jacobian matrix : basins stations 186 stations

Principle of the assimilation system

IS2 will be retained IS2 combines the best Nash and rmse scores, and the lowest increments The Doubs at Besançon Scores for a selection of 148 stations

The Garonne at Portet-sur-Garonne

An exemple of the effect on ensemble forecasts PEARP ECMWF

Some statistical scores Scores for a selection of 148 assimilated stations for the 10-day ECMWF-SIM

RMSE Scores are presented against streamflow observations

Brier Skill Score day 1

Brier Skill Score day 10

Ranked Probability Skill Score

Decomposition of Brier

BSS for PEARP-SIM and ECMWF-SIM BSSs are unbiased with the Weigel et al. (2007) method because of the impact of the number of members PEARP is slightly better, but without the unbiasing, ECMWF wins!

Conclusions and perspectives A streamflow assimilation system has been implemented and validated for the SIM suite –Better simulation of flows and initial states for the ESPSs (Thirel et al., submitted to the Journal of Hydrology) Significative improvement of ensemble streamflow forecasts when initialized by the assimilated SIM suite –Lower RMSE, better BSS and RPSS –Few differences between SIM-PEARP and SIM-ECMWF –It is the first time that the ensemblist SIM is compared to observations, not a reference run Perspectives : –Optimizing computing costs and the quality of the assimilation system –Using another operator (EnKF?) –Implementing the assimilation system into the SIM-ECMWF operational suite (2012?)

Thank you for your attention!