Seasonal Forecasts of River Flow in France Laurent Dubus 1, Saïd Qasmi 1, Joël Gailhard 2, Amélie Laugel 1 1 EDF R&D (Research & Development Division) 2 EDF DTG (hydro-meteorological forecasting division) laurent.dubus@edf.fr We acknowledge Météo-France, ECMWF and University of Cantabria @ Santander for providing seasonal forecasts
Importance of hydro-power in the french electricity system Hydropower is very important for the optimization of the balance between power production & demand, both at national and regional scale (e.g. Denmark/Sweden/Norway/Germany grid connections) Seasonal Forecasts of River Flow in France 2
Water cycle variability impacts Effects of floods on distribution stations Effects of floods on coal mines Effects of river water temperature and discharge on thermal units cooling capacity (e.g. summer 2003 in France) è Strong need of forecasts! Seasonal Forecasts of River Flow in France 3
Hydro-meteorological forecasts @ EDF / DTG: a long experience Monitoring Safety Deterministic forecasts Input: AROME, ARPEGE, ECMWF 24h Water resource management & system optimization Ensemble forecasts Input: EPS (ECMWF) + Analogs 14 days EDF s observation network Probabilistic forecasts Input: Precip. & Temp. climatology Reservoirs filling Forecasts of low levels 6 months Drainage basins with operational hydrological model There is some natural hydrological predictability! Seasonal Forecasts of River Flow in France 4 UVIG Workshop on Variable Generation Forecasting Applications - Lakewood, Feb 18-19, 2015 EDF
The MORDOR Hydrological Model Precipitation Atmosphere Évapotranspiration Rainfall Snow Surface U + Z L + Deep N + Watershed Évaporation Watershed Production Naturel River Flow Seasonal Forecasts of River Flow in France 5
Reference method Hydrological Model (MORDOR): Initial conditions + Atmospheric forcing = rainfall & temperature D to D+14: NWP (ECMWF EPS + Analogs) D+14 to M+6: climatology [~0 predictability] Mountain basins: strong relationship between winter snow stock and spring run-off è High natural predictability Plain basins: relationship between spring ground water stock and summer low river flows è Lower natural predictability Seasonal Forecasts of River Flow in France 6
The analog method Z700 & Z1000 D & D+1 Weather archive : NCEP 1962-2010 (Z700 & Z1000) è Analog dates Ensemble forecasts T2m, RR T2m, RR observations 1962-2010 Source :Jérémy Chardon sphd manuscript, 2014 Seasonal Forecasts of River Flow in France 7
Monthly forecasts of river discharge using ECMWF products + in house post-processing methods (analogs) ECMWF Monthly fcst (Z700 & Z1000) Analog Method Local T2m & Precip over 43 basins Hydrological Model Streamflow Prob fcsts Streamflowclimatology Hydro Model forced by T2m & Precip climatology (1969-2008) Hydro Model forced by Analog T2m & Precip Observation What about seasonal lead times? Dubus, 2014 Seasonal Forecasts of River Flow in France 8
Data & Forecasts - Hindcasts Z700/Z1000: ECMWF System 4 1981-2010 ; - Reanalysis Z700/Z1000: NCEP 1948-2010 ; - Forecasts T2m/RR : 1. ECMWF System 4 : 1981-2010 2. ARPEGE System 3 : 1979-2007 - 33 basins in France - Observations : T2m/RR : 1953-2010 ; Seasonal Forecasts of River Flow in France 9
T2m/RR forecasts : 33 watersheds, Run : Feb MAM MJJ ROC score for T2m in the upper tercile Initialization time : Feb System 4 raw Analogs ARPEGE 0.55 Improvement 0.5 Deterioration 10
T2m/RR forecasts : 33 watersheds, Run : May JJA ASO ROC score for T2m in the upper tercile Initialization time : May System 4 raw Analogs ARPEGE 0.5 Improvement Deterioration ARPEGE is good in spring Same results for precipitations 11
Streamflow forecasts: May for JJA ECMWF Seasonal fcst (Z700 & Z1000) Analog Method Local T2m & Precip over 33 watersheds Hydrological Model Streamflow Ens fcsts Durance@Serre- Ponçon, Initialization : May, Lead time : JJA Q90 Q50 Q10 Streamflow climatology Hydro Model forcedby T2m & Precip climatology (1981-2009) Hydro Model forcedby Analog T2m & Precip Observation 12
Streamflow forecasts, initialization : August MORDOR forced by T2m/RR analogs SON NDJ 13
Streamflow forecasts, initialization : May MORDOR forced by T2m/RR analogs JJA ASO 14
Streamflow forecasts, initialization : May ROCSS : Improvements w.r.t. the reference JJA ASO 15
Residual biases need to be adjusted Isere@Pizançon, Initialization : May, Lead time : JJA Climatology Hydro Model forced by T2m & Precip climatology (1981-2009) Hydro Model forced by Analog T2m & Precip Observation Bias in temperature Bias in precipitations 16
Summary q On monthly & seasonal time scales, hydrological predictibility = Ø Natural predictibility from initial conditions Ø + Atmospheric forcing predictibility q Recent developments in monthly forecasts (ECMWF) brought significant improvements in weather/climate forecasts, hence in hydrological forecasts. q Monthly forecasts in operations @EDF for more than 5 years q Seasonal time scales: some encouraging results in France, BUT ü Atmospheric forcing still not skillfull enough ü Residual biases after analog method: global (Z) or local (T2m, RR) adjustments to be tested Seasonal Forecasts of River Flow in France 17
Thanks to Météo- France, ECMWF, University of Cantabria, And EUPORIAS Partners Thank you for your attention Contact: Questions? laurent.dubus@edf.fr EDF R&D Expert Researcher Fluid Dynamics, Energies and Environment Department Applied Meteorology and Atmospheric Environment Group 6 Quai Watier - BP 49 78400 CHATOU CEDEX FRANCE Seasonal Forecasts of River Flow in France 18
Some skill, depending on season and location Low level flow forecasts [ Meuse@Chooz - 1990-2011] April October Predictability >30 days Predictability~20 days But what solution for other seasons & locations, and longer lead times? Seasonal Forecasts of River Flow in France 19
Geopotential heights Target date 1 st Analog Date 2 nd Analog Date 20
Z700/Z1000 Mean bias / 1981-2010 Init: May for JJASO Z700 : almost no bias Z1000 : mean bias : 10 % 21
Better scores for some events T2m, init November for DJF Cold winters 22
ROC / River Flow init: May for JJA 23