Operational fl ood risk management based on meteorological ensemble predictions (case study: Mulde) Jörg Dietrich1, Eva Lechthaler1, Andreas Schumann1, Frank Voß1, Yan Wang1, Michael Denhard2, Sebastian Trepte2, Bernd Pfützner3, Ingo Michels4, Michael Redetzky4, Jörg Walther4, Uwe Büttner5

1 Ruhr-University Bochum, Institute for Hydrology, Water Resources Management and Environmental Engineering 2 Deutscher Wetterdienst (German National Weather Service), Offenbach 3 BAH Institute for Applied Hydrology, Berlin 4 WASY GmbH Institute for Water Resources Planning and Systems Research, Berlin Funding: Bundesministerium für Bildung und Forschung. Projektträger: 5 Sächsisches Landesamt für Umwelt und Geologie ( State Offi ce for Environment and Geology), Dresden BEO, Forschungszentrum Jülich, FKZ 0330694

Project Goals The general aim of this project consists in a fl ood risk management possible consequences and provides the information which is based The following sub-results will be provided: system which combines numerical weather forecasts, hydrological on a probabilistic assessment of the uncertainties of the input values • determination of options and limitations of ensemble precipitation modelling with a GIS-based estimation of resulting fl ood risks within as well as uncertainties of the hydrological model used resulting from predictions for fl ood forecasting, a river basin of the mesoscale. Under consideration of the fast reac- model and parameter uncertainties. Measured data of precipitation and • evaluation of the fl ood protection capabilities of reservoirs in real- tion of mountainous watersheds and the uncertainties of numerical runoff will be used for data assimilation in meteorological and hydro- time, weather forecasts ensemble predictions will be used to consider the logical models and for a probabilistic re-assessment of fl ood forecasts • development of an integrated fl ood management system, uncertainties of the meteorological inputs. The transformation of the based on Bayesian statistics. The main result of this project will consist ensemble predictions of rainfall into runoff ensembles is done by a in a probabilistic assessment of fl ood risk in real-time which is derived • development of a methodology to consider forecast uncertainties spatial distributed fl ood forecast model. The result will be an ensemble from a chain of the following three model types: based on ensemble predictions and hydrological uncertainties, of fl ood forecasts at each time step. The different discharge values at • meteorological models • hydro-meteorological data assimilation to reduce model uncertain- the fl ood gauges are related to specifi c local risks which are specifi ed ties in real-time and • hydrological models by so-called fl ood protection concepts based on a detailed estimation • probabilistic assessment of fl ood risk based on Bayesian statistics. of fl ood-prone areas. The risk management system determines the • hydraulic models

Flood Forecast Uncertainties Study Area One main focus of the project is to quantify the different uncertainties of the fl ood fore- Legend The Mulde catchment (Ger- Main Rivers casting: many, Czech Republic) has Bad Düben 1 Gauges • the input uncertainty (precipitation) and DEM a size of 6178 km². It con- • the hydrologic uncertainty (all other uncertainties, e.g. uncertainties of the model, its Elevation [m] sists of several parallel sub- 69 - 100 catchments. Therefore the parameters or of the measured data used in real-time forecasts). M Thallwitz u 100 - 200 ld spatial uncertainties of the e 200 - 300 ensemble forecasts for precipitation Nemt 1 precipitations forecasts 300 - 400 precipitation forecast measured precipitation data Neichen lead to large uncertainties weather radar data 400 - 500 precipitation forecast of fl ood warnings with re- Golzern 1 500 - 600 precipitation forecast 600 - 700 gard to the locations of a precipitation forecast probabilistic Großsermuth 700 - 800 possible fl ooding. Within evaluation Erlln 800 - 900 the project the uncertain- Mahlitzsch Niederstriegis 1 900 - 1.000 Kriebstein UP Nossen 1 ties of forecasts shall be u a 1.000 - 1.100 measured parameter flood forecast model p Böhrigen o quantifi ed. Wechselburg 1 h discharge tracking ArcEGMO c Krummenhennersdorf 1 1.100 - 1.200 Göritzhain s Z

e ld Naundorf discharge forecast u M Lichtenwalde Worlkenburg Oberschöna e Flöha F discharge forecast u r a Berthelsdorf 2 e k Chemnitz 1 Hetzdorf 2 i ic b w e discharge forecast Z Mulda Licrhgtenberg 2 Kunnersdorf e Niederlungwitz Altchemnitz 1 F r Harthau l M ö WolfsgrundBurkersdorf 2 discharge forecast h u

u l aBorstendorf d Niedermülsen 1 1 a e discharge forecast p Zwickau-Pölbitz Burkhardtsdorf 2 o h Deutschgeorgenthal 1/2

ensemble forecasts for discharge c Pockau 1/2 TS ZP Hospfgarten Neuwernsdorf Z Zöblitz 2 Rauschenbach 1/2 Niederzwönitz alternatives in flood control Streckewalde reservoir operations Kirchberg Niederschlema Wiesa Rothenthal by reservoirs Wolfersgrün Aue 3 Tannenberg Aue 1 Annaberg 1 probabilistic Schwarzenberg Schmalzgrube 2 Markersbach 2 evaluation of Neidhardtsthal Jöhstadt 4 Markersbach 1 Jöhstadt 1 Schönheide 3 Cranzahl forecasts 2 Rittersgrün Cranzahl H RautenkranzTS Sosa 2 predicted spatial extension flood risk assessment, 3/4 O TS Carlsfeld UP 1 Johanngeorgenstadt 2 Sachsengrund and time of flooding decisions about flood alerts Muldenberg 1 TS Carlsfeld ZP 1 0 5 10 20 30 km flood protection plans

Fig. 1: Conception of an ensemble-based fl ood management Fig. 2: Mulde river basin with gauges for fl ood forecasts

Precipitation Forecast Hydrological Model and Parameter Uncertainties Ensemble predictions of precipitation are used to assess the uncertain- Arc EGMO, a deterministic rainfall-runoff model is used in this investiga- ties of precipitation forecasts. These ensemble forecasts are based on tion. The parameter uncertainty will be considered by the application different mesoscale systems (ECMW-Ensemble and Lagged Average of several parameter sets generated based on the GLUE-method (Ge- Forecast Ensembles). neralized Likelihood Uncertainty Estimation; Beven and Binley, 1992). IT-Tool Flood Management System Global prediction LMK The evaluation of indi- systems GME o LM The fl ood management system integrates hydrological models, para- 2,8 km The parameters of the model will be re-calibrated with discharge data 40 - 80 km vidual forecasts will be transmitted online. Special emphasis will be given on the interaction meter tracking methods and modules for reservoir control and decisi- carried out by means Mesoscale ensemble Assimilation LMK-Ensemble on support. An intelligent bidirectional interface enables the real-time “Lagged Average Forecast” between sub-catchments. The application of three feedback mecha- systems Measurements 7 - 22 km Ensemble of „Bayesian Model Up to +18 hours nisms will be checked: communication between models and databases (fi g. 6). A new GIS- Averaging“(BMA). Real based component for the automated calculation of fl ood areas and Superensemble: COSMO-LEPS, SRNWP-PEPS, LMK-Ensemble (up to +5 days) • the error in fl ood forecasts will be considered as a stochastic variab- time radar measure- le, expected fl ood damages is tested in a pilot area. ments of precipitation Calibration (BMA) Measurements • single parameters and also the will be used for data as- Mulde food management system Precipitation development scenarios in the Mulde catchment • initial parameters will be updated. similation and assess- for hydrological ensemble forecasts up to +5 days ments of the uncer- data server WISYS client probabilistic assessment : DSS DLL tainty by probabilistic Fig. 3: Making of the precipitation forecast ensembles geodatabase : ESRI-GDB decision support : means. DSS GUI szenario tables model scenarios : Model GUI

parameter tracking : data transfer database Flood Management - Technical Aspects Decision Support Model GUI interface : ISSNEW reservoir control: The potential of fl ood protection of selected reservoirs is analyzed ba- The possible consequences of fl ood forecasts will be assessed by pro- wrapper Model GUI sed on the relationships between the storage capacities in retention babilistic means. Results from the probabilistic assessment of the en- data transfer model structures and fl ood risks downstream. Therefore diverse fl ood event ty- semble members and the assimilation of measured data are evaluated ArcEGMO pes, different control strategies and technical with a belief function reasoning approach. wrapper precipitation-runoff model : parameters of the reservoirs are considered. Comprehensive fl ood risk analyses and maps ArcEGMO DLL (fi g. 5) are related to the fl ood forecasts. As a call On this basis the fl ood prediction model will reservoir module : be used to develop event based control stra- result ensemble predictions of the potential ArcEGMO Routine tegies taking into account past fl ood events fl ooding areas will be provided. as well as ensemble-based scenarios. Addi- A decision support system provides recom- Collaboration of the main components of the fl ood management system tionally the options and limitations of using mendations for issuing alerts by multiple cri- Fig. 5: Exemplary fl ood risk map Fig. 6: Fig. 4: Eibenstock Reservoir (http://www.umwelt.sachsen. (Foto: Foto-Design E. Hertel, ensemble-based predictions for decision sup- teria assessment of the current hydrological de/de/wu/umwelt/lfug/lfug-in- Quelle: http://www.gemeinde- port concerning the fl ood control of reservoirs ternet/wasser_13439.html) schoenheide.de/talsperre.htm) situation and scenarios of future develop- are determined. ments.

Reference Beven, K. & A. Binley (1992), The future of Distributed Models: Model Calibration and Uncer- tainty Prediction, Hydrological Processes, 6, 279-298.