The European Flood Forecasting system
HEPEX Workshop, 08-10 March 2004 The EFFS: the facts
• Project EFFS was sponsored by the 5th Framework Programme of the European Commission. • Duration 01/03/2000 – 31/09/2003 (42 months). • EC financial contribution: 1.8 M€ over 3,5 years. • 11 institues + 8 NAS institutes= 19 partners in total. EFFS Project Objectives
• EFFS aims at developing a prototype of a flood forecasting system for the European countries for up to 10 days ahead. • The main emphasis is on the medium-range lead time, i.e 4 to 10 days. • The prototype is designed at providing pre-warning to local water and flood forecasting authorities across Europe. • The system should permit encapsulation of pre- existing hydrological and river routing models already tested and used by local authorities. EFFS “Mother” Consortium
• Original Consortium: 11 institutes • 3 Weather services (DWD, DMI, ECMWF) • 3 Universities (Lancaster, Bristol, Bologna) • 2 Research Centres (Delft Hydraulics, JRC) • 2 National Hydro-meteorological Services (SHMI, BAFG) • 1 National Water Management Authority (RIZA) EFFS “Mother” Consortium
DMI SMHI WL|Delft, DWD RIZA
GRDC/BFG Univ Lancaster Univ. Bristol ECMWF
JRC Ispra Univ. Bologna EFFS NAS-1 amendment
• In June 2002 an EFFS amendment is approved to include 8 additional institutes from the Newly Associated States (NAS) in view of the enlargement of the EU. • Amendment financed with ~ 4 k€ over 1 year • New opportunities: Data from river systems in Eastern Europe are made available for sharing within the consortium. • Results of hydrological and river routing models can be compared. EFFS NAS-1 Partners
IMWM-PL GGI-LT
Vituki Plc
SHMU-SK Slovak UT-SK NIMH-RO
NIMH-BG
Uni Ljubljana-SI EFFS Project Organigramm
WP 1 Co-ordinator & interface with EC WL-Delft Hydraulics
WP 2 Meteorological data WP 3 Hydrological data WP 4: Building a collection & supply collection geographical DB DWD GRDC UNI Bristol
WP 5 Parameter and forecast WP 6 Downscaling and validating uncertainty estimation meteorological forecasts Lancaster DMI
WP 7 Development of the WP 8 Application, validation & European-scale FF system comparison of watershed models JRC Uni Bologna
WP 9 Set-up & testing of WP 10 Dissemination of prototype system forecasts & warnings RIZA SHMI EFFS structure 1. Global model 2. HIRLAM Weather forecast 3. EPS up to 10 days ahead 4. HIRLAM Mini-EPS
conversion from GRIB to P,T,…. NEFIS format Hydrological Models Soil, land-use, meteo and hydro data from hydrological networks model parameters save in DB
River routers 1. Sobek 2. Muskingam-Cunge EFFS open- save in DB architecture platform Operational testing Inundation modelling 1. Overtopping? 2. Defence structure postprocessing collapse?
Warning save in DB Numerical Weather forecasts
• ECMWF supplies medium range meteorological data from their deterministic model and from their Ensemble Prediction System (EPS).
• DWD and DMI supply deterministic high resolution short range forecasts from their High Resolution Local Area Model (HIRLAM).
• HIRLAM is initialised through Global ECMWF model output
• The high-resolutions forecasts make it possible to estimate the impact of horizontal resolution of the atmospheric predictions on water level forecasts.
• In addition, DMI is experimenting with mini ensembles with the aim to investigate uncertainties in the precipitation forecasts. High Resolution Local Area Model nested domains
Courtesy Danish Meteorological Institute Precipitation hindcast
Courtesy Deutscher Wetterdienst Hydrological measuring networks in flood forecasting
• The operation of a river basin-scale flood forecasting system, and in particular of a continental-scale system, as EFFS, requires regular update through measured data.
• Which data are required?
1. Precipitation 2. Temperature 3. Synoptic weather data 4. Discharge data at critical locations 5. Water level data at critical locations Measuremement network for the Rhine Basin
Stream flow gauges DWD meteorological stations Data acquisition in EFFS •The following scheme shows the data-stream from measuring networks to the EFFS is performed. •The data acquisition was led by the GRDC in Work-Package 3.
Courtesy GRDC Utilisation of measured data in the flood forecasting system • The regularly update measurement form the hydrological networks are used for the following purpose: • In now-casting (extremely short lead-time, max 24 hrs) directly measured precipitation data form gauges and radar are used to drive models. • in EFFS (mdium-range lead-time) hydrological and hydraulic models are updated (data assimilation) through data measured over the 2 weeks precedent the begin of the forecast.
Hattingen / Ruhr - • The data are transferred from the 1993 flood event loggers over FTP into the data- base of the flood forecasting system. Hydrological modelling
• Different hydrological models are used within the EFFS system: 1. HBV (SHMI) semi-distributed model the entire Rhine Basin (calibrated at the BFG, Koblenz) representing the Rhine Basin up to Switzerland through 134 sub-basins. 2. LISFLOOD (JRC) raster-based model for simulation of continental river basins at 5 sqkm resolution. Used in particular for Rhine, Odra, Danube. 3. TopKapi (Uni Bologna) for simulation of the Po river basin
N.B.: The system is however conceived as “open “ allowing any other model to be incorporated through an appropriate model-adapter. LISFLOOD model
Courtesy Joint Research Centre River routing
• Lateral inflows into the main river system are calculated by the hydrological model (e.g. HBV).
• The water is subsequently routed along the main river system (e.g. Rhine).
• Within the EFFS river routing is performed with the WL | Delft Hydraulics Saint-Venant model SOBEK.
• Discharge Q and water level height H at critical sections are calculated.
• Effects of engineering structures such as weirs, locks and bridges are included in the schematisation. River routing
L o b ith [86 2 .2 ] Ree s [ 8 3 7 .4 ] 17 16 L i pp e [81 4 .50] We s e l [ 8 1 4 . 0 ] E m s c h e r [7 9 7 .80] 15 R uhr o r t [7 8 0 .8 ] Mo d e l B R u h r [78 0 .10] 14 Dü sse ld or f [ 7 4 4 .2 ]
E rft [7 3 6 .5 5] 13 W u ppe r [70 3 .3 0] Col o gne [68 8 . 0 ] 12 Si eg [ 6 5 9 .4 ] B o n n [ 6 5 4 .8 ]
A h r [6 2 9 .4 5] 11 W ie d [ 6 1 0 .2 0 ] S a y nba c h [59 9 .9 0 ] Mod e l A An de rna c h [ 6 1 3 .8 ] 9 10 Mo s e ll e [ 5 92 .3 ] N e t te [60 8 .70] L a h n [ 5 8 5 .7 ] So bek m o d e l M o s e l le [ 5 92.3] 8 8 Lo b it h [8 6 2 .2 ] W is p e r [ 5 4 0 .30] Ka u b [ 5 4 6 .2 ] 7 M a in z [ 4 9 8 .3 ] Ree s [ 8 3 7 .4 ] 7 17 Nahe [ 5 2 9 .1] Ma i n Mai n [ 4 96 . 6 ] 16 L i pp e [ 8 1 4 .5 0] Na h e [4 9 6 .6 ] We s e l [ 8 1 4 . 0 ] [52 9 .1 0] Se lz 5 E m s c he r [79 7 .80] [5 1 8 .8 0] 6 M o d a u [47 3 .70] 15 R u h r o r t [7 8 0 .8 ] Wo r m s [ 4 4 3 . 4 ] W e s c hn i tz [ 4 5 4 .70] Ru hr [78 0 .1 0 ] 4 4 14 Nec k a r [ 4 2 8 .2 ] D ü s s el d o rf [ 7 4 4 .2 ] 3 Ne c k a r [ 4 2 8 .2 ] Sp ey er b a c h [ 4 0 0 .2 0 ] 2 E r ft [7 3 6 .5 5 ] 13 W u ppe r [ 7 0 3 .30] Qu ei c h [ 3 8 4 .8 0 ] 1 Co lo g n e [68 8 . 0 ] 1 Pf in z [ 3 8 0 .70 ] Al b [ 3 7 1 .2 0 ] 12 Si eg [ 6 5 9 .4 ] Ma x a u [ 3 62 .3 ] B o n n [ 6 5 4 .8 ]
Ah r [6 2 9 .4 5 ] 11 An de r n a c h [ 6 1 3 .8]
48 h o u r fo r e c a st WS D Floodplain inundation modelling
• Once water overtops a dyke and invades lowlands (e.g. polders), 2-D flood-wave propagation modelling is needed to forecast the extent of the inundation.
• Inundation modelling within EFFS can be performed by two different models:
1. The 2D inundation of the University of Bristol and JRC (LISFLOOD-FP. 2. The Delft 2D inundation model. Inundation Modeling River Severn (UK) Raw LiDAR data (6 x 3 km domain)
Topography map
Vegetation height map
Inundation modelling results
Courtesy University of Bristol Uncertainty
• Flood forecasting cannot be separated form the problem of the uncertainty inherent to the input, model structure and parameter values. • Any flood forecasting system is made up by a cascade of models, usually : Meteo Hydrology Routing Inundation
• Each model contains parameters which lie within a range, i.e. their exact value is not known. • In addition the inputs (e.g. rainfall) as well as the initial conditions are affected by uncertainty. • Subsequently the predictions lead to a bandwidth of forecasted values, which may over- or underestimate the actual water levels. Model integration: prototype FEWS
• The modeling steps needed in a flood forecasting system are integrated through an integration platform (shell-tool). • The shell-tool is designed as open and re-usable software system. • The system is adaptable to different hydrological and hydraulic models through model-adapters. • The platform is linked to a data-base system. • Results of various model runs are saved and retrieved subsequently from the DB. • Within the shell-tool model runs, analysis of results and postprocessing are facilitated. The Open-Architecture Platform Testing of the prototype
• The EFFS is being tested within the frame of the project in a semi-operational fashion by the Dutch institute RIZA on the river Rhine. • Several hundreds of forecasts have been carried out though direct access to DWD weather forecast data and subsequent running of the model cascades. • The predicted water levels are compared directly with the ones measured at the German-Dutch border flow measuring station. • Results of these test-runs will be presented later on in the conference. Flood Warning & dissemination
• Once a high water is forecast by the system, decision need to be made if evacuation is necessary. • False alarms can be as damaging as not issuing a warning at the right moment. • Decisions stay ultimately with the forecaster and need use of historical records for verification and uncertainty reduction. • A systematic approach to address this issue is matter of ongoing research. Summary
• The outlined system components will be addressed by the project partners through dedicated sessions during this conference. • Results and applications of various project applications will be presented. • Systematic quantification and handling of the inherent uncertainties on forecast results are a complex issue which needs particular attention from a research point of view. Contributions
To this presentation have contributed the following people :
• Keith Beven • Jaap Kwadijk • Paul Bates • Ad de Roo • Erdmann Heise • Kai Sattler • Tony Hollingsworth • Eric Sprokkereef • Bo Holst • Ezio Todini • Michael Hils
The conference organisation would like to wish you a pleasant stay and is looking forward to fruitful discussion!
Thank you very much for your attention.