Abstracts International Symposium on Innovations in Flood Forecasting Systems

Transcription

1 Abstracts International Symposium on Innovations in Flood Forecasting Systems

2 Book of Abstracts March 2011 International Symposium on Innovations in Flood Forecasting Systems Flanders Hydraulics Research Antwerp, Belgium 1

3 Waterbouwkundig Laboratorium Flanders Hydraulics Research Leen Boeckx Berchemlei Antwerp - Belgium leen.boeckx@mow.vlaanderen.be

4 International Symposium on Innovations in Flood Forecasting SystIntroduction In the past decennia several flood forecasting centres have been established in Europe to continuously evaluate the risk of flooding. These flood forecasting centres often are responsible for keeping hydrological and hydraulic models up to date and operational. These models provide forecasted waterlevels and discharges on a regular base. The systems used, differ for every flood-forecasting centre. Therefore, to celebrate the 10th anniversary of its forecasting centre (HIC), Flanders Hydraulics Research has chosen to organise the International Symposium on Innovations in Flood Forecasting Systems, which aims to unite European technical experts in Flood Forecasting, providing them with the opportunity to share experiences and to present innovative solutions they have implemented. 2

5 Table of contents Table of contents... 3 Improvements in Traffic related Water level Forecasts in Germany... 4 A data based approach to quantify and visualise uncertainty in flood forecasting Application of data assimilation techniques in the operational flood forecasting system for Dutch rivers... 6 The flood forecast model LARSIM: application experiences and evaluation of operational runoff forecasts in the Moselle basin... 7 Model predictive real time flood control... 9 SNOW 4 An Operational Model for Estimating Precipitation Supply A parallelization approach for multi model real time flood forecasting of the Po river in Italy

6 Improvements in Traffic-related Water-level Forecasts in Germany Author: Dennis Meissner Federal Institute of Hydrology Germany, Am Meinzer Tor 1, Koblenz, Germany Abstract The presentation starts with a brief overview of the forecasting landscape in Germany focusing on the Federal Waterway Rhine River. The different responsibilities of the federal states and the federal government originating from the federalistic regime are explained as well as the aims and tasks of the Federal Institute of Hydrology (BfG) related to water-level forecasting. Since 1996 the BfG has published traffic-related water-level forecasts for the free flowing section of the River Rhine. As in 2008, the forecast period could be extended from two to four days with a view to improving the efficiency and reliability of the Rhine as an environmentally friendly waterway, while ensuring a sound and stable forecast quality. The current forecasting system operated by BfG is descripted with regard to the management and processing of input data, the use of hydrological and hydraulic models as well as the publication of forecast results. The forecast quality achieved is presented as well. Furthermore the speech will focus on the improvements made over the last years with respect to the communication of forecast uncertainty, which is realized as a continuous online-comparison of measured and predicted water levels so far, and with regard to the coordination of forecast activities between BfG and the Flood Forecasting Centers for the River Rhine. Last but not least the main research activities pursued by BfG in order to achieve and improve sustainable forecast quality are presented. 4

7 A data-based approach to quantify and visualise uncertainty in flood forecasting. Author: Niels Van Steenbergen Flanders Hydraulics Research/ Katholieke Universiteit Leuven, Berchemlei 115, 2140 Antwerp, Belgium Abstract In the last decades the Flanders region in Belgium has suffered from different floods. Besides structural measures, the Flemish Government has set up flood forecasting systems to be used as early warning systems. These flood forecasting systems make use of hydrological and hydrodynamic models and input time series (measured rainfall, predicted rainfall, evapotranspiration, water levels and discharges). The uncertainty of these models and time series, certainly the predicted rainfall, is high and not always known. Therefore the result of flood forecasting systems is also uncertain. To estimate this uncertainty the Katholieke Universiteit Leuven together with Flanders Hydraulics Research of the Flemish government have set up a method to calculate and visualise this uncertainty and at the same time the exceedance probability of alert and alarm levels. Hereby detailed probabilistic information can be given to the water managers in order to take accurate measures to reduce the impact of floods, but also to prevent false alarms. The method consists of an uncertainty analysis on historical simulation results and observations. The model residuals (difference between model result and the observations at river gauging stations) have been analysed statistically and it has been found that these residuals are not normally distributed. Therefore a non parametric technique was selected. Because the residuals are correlated with the value of the simulated water level and the time horizon, the residuals are split up into discrete value classes, based on the simulated water levels and into classes of different time horizons. For each combination of value class and time horizon, different percentile values of the residuals are calculated. These percentile values are stored in a so called three dimensional error matrix. Based on a 3D interpolation in the error matrix a bias correction is executed and confidence intervals on simulation results are calculated and visualised. By interpolating between the confidence intervals the exceedance probability of a certain alert or alarm level is calculated and visualised. This method is implemented in a software program, that is connected to the database of the forecasting system. Hereby it is possible to automatically update the error matrix, based on new simulations. Also a mapping method for the inundation probability will be presented. 5

8 Application of data assimilation techniques in the operational flood forecasting system for Dutch rivers Author: Matthijs Lemans Deltares, Rotterdamseweg 185, 2629 HD Delft (Postbus MH Delft), the Netherlands Abstract In many operational flood forecasting systems, the amount of data is significant. Observed data from different sources is automatically feeded into to the system, together with radar, satellite and/or numerical weather prediction data. The data often has to be validated and transformed in order to run different kinds of models, like rainfall-runoff and hydraulical models. The DELFT-FEWS system, designed by Deltares, is a widely used toolbox for developping forecasting systems. It is a data management system with a highly modular structure, with an open interface to models and data. This allows the operational application of several data assimilation techniques. These techniques aim for improving the forecast as much as possible by making use of the available observed data in the system. Examples of data assimilation techniques are ARMA error correction modules, ensemble Kalman filtering for state updating and Quantile Regression, based on predictive uncertainity. The presentation will show an example of a DELFT-FEWS application for the Dutch rivers Rhine and Meuse, where several data assimilation techniques are implemented, improving the forecasts significantly. 6

9 The flood forecast model LARSIM: application experiences and evaluation of operational runoff forecasts in the Moselle basin Authors: Kai Gerlinger[1], Norbert Demuth[2] [1] HYDRON Engineering Consultants, Karlsruhe, Germany, Haid und Neu Str. 7, D-76131, Karlsruhe, Germany [2] State Environment Agency Rhineland-Palatinate, Mainz, Germany Abstract Flood forecasting and warning is a prerequisite for successful mitigation of flood damage. The water balance model LARSIM links different weather forecasts, meteorological point measurements and radar data, the state of the river catchment, river discharges and water levels to provide the relevant flood forecast information for realtime decision-making on flood prevention. LARSIM simulates the terrestrial part of the water cycle, including interception, snow accumulation and melt, evapotranspiration, soil-water movement, runoff generation, runoff concentration, and river routing. In addition, retention ponds, reservoirs, and lakes as well as water withdrawal and water addition are accounted for. The meteorological forcing variables (precipitation, air temperature, etc.) are internally interpolated from point measurements to any simulation point within the model. Also external interpolation methods based on different geostatistical techniques are applied. Fall-back procedures in case of missing meteorological or hydrological input data are implemented, to assure forecast functionality even during times of reduced data availability. The parameterization of the model is commonly based on readily available data, such as digital elevation maps, digital maps of land cover and soil classification, digital river networks and geometries along with additional information about retention ponds and reservoirs. LARSIM has been successfully used with different spatial and temporal resolutions to simulate the terrestrial water cycle of various catchments in different parts of the world. For real time discharge forecasting LARSIM commonly runs on an hourly time step. Beside operational forecast the model has also been used to predict the impact of climate change on the terrestrial water cycle and to analyze the effect of land use changes on storm flow. The model is continuously applied for operational flood and low-flow forecasting by the flood forecast centres of the state authorities in southern Germany (federal states of Baden-Wuerttemberg, Rhineland-Palatinate, Hesse and Bavaria), in western Austria (Vorarlberg) and eastern France (Alsace, Lorraine). The operational LARSIM models cover amongst others most of the tributaries to the Rhine basin (up to the gauge of Koblenz (approx km²)) and to the German Danube Basin (up to gauge of Passau (approx km²). The models have a high spatial resolution (subarea structure based on 1km²-grids or on hydrologic subareas (< 5km²)). For visualisation and communication of the forecast results different user interfaces as well as programs for plot generation and internet publication are used by the authorities. For operational use, LARSIM has been amended by a variety of automated, process-oriented optimization and data assimilation techniques to adjust the input and output variables as well as the state variables. In brief, hydrology is always simulated for at least two days before the point of forecast (present point in time), using measured data for model forcing. The simulation results are internally compared with measured discharges. In the case of discrepancies, the model is automatically optimized to better match the measurements. This optimization helps to improve the quality of flood forecasts as well as low-flow forecasts. On one hand LARSIM is used for flood forecasting at gages which seeks to provide accurate information about expected water levels and discharges up to several hours ahead. On the other hand LARSIM is also used for early flood warning to inform water authorities and the public up to several days before the occurrence of flood events. Because of the long forecast time and the associated uncertainties of the precipitation forecasts the early flood warning is only a rough estimation of a possible upcoming flood. Therefore early flood warning maps are provided and published which predict the flood risk not for a single gage but for administrative districts. These maps are based on a statistical evaluation of the forecasted discharges. 7

10 The application experiences with LARSIM show the ability of the model to simulate and forecast different flood situations (convective and stratiform precipitation as well as snow melt) quite well. Simulation and forecast quality strongly depend on the available data. Regional differences in the model reliability are often due to uncertainties in the rating curves but can also be the result of a simplified representation of complex hydrologic basin properties (like karst or undefined gravel flood plains). To better assess the uncertainty of the model the flood forecasting services in the Moselle basin decided to evaluate/verify the uncertainties of their operational runoff forecasts. Defined uniform criteria for different lead times and hydrological situations were defined. These criteria were integrated in an evaluation tool to calculate automatically the different scores. The verification will be carried out for about 60 gauges and at least daily forecasts of the last 13 years. The defined verification criteria will be presented. In general, uncertainty is inherent in the forecasting processes and there is growing recognition by flood forecasters and managers that estimation of uncertainties is an important aspect in decision making. LARSIM is currently extended for the evaluation and presentation of flood forecasting uncertainty based on ensemble modeling. Several challenges remain including the presentation of this uncertain information in a consistent and easily understood manner for decision makers and to convert this uncertainty to a corresponding risk. 8

11 Model predictive real-time flood control Authors: Mauricio Villazon, Po-Kuan Chiang, Patrick Willems Katholieke Universiteit Leuven, Hydraulics Division, Kasteelpark Arenberg 40, BE-3001 Leuven Abstract Real-time flood control applications require fast hydraulic models, because of the large number of iterations required in optimization procedures. Simplified conceptual hydraulic models can meet this need, but have to be calibrated based on simulation results with more detailed full hydrodynamic river models. To support the conceptual model building and calibration process, a technique has been developed, and tested for two cases in Flanders. In one of these cases, the conceptual model also has been applied in support of the real-time control of hydraulic regulation structures and flood storage reservoirs. The model simplification is reached by lumping the processes in space. Water levels and discharges are simulated, not every 50 meters as the full hydrodynamic model does, but only at the relevant locations. These are the locations up- and downstream of the hydraulic regulation structures, to be considered by the real-time controller, as well as the locations along the river network where potential flooding is induced. Advanced conceptual modelling procedures have been considered based on separation of static and dynamic storage along river reaches. Also a semi-automatic procedure has been tested based on the identification and calibration of dynamic non-linear transfer functions. These functions have been advanced to account for the flood effects on the routing and by a floodplain model combined with an overflow submodel that allows modelling the flow between the river and the floodplains. The (manual and/or semi-automatic) procedure for conceptual model identification, calibration and validation has been tested for two case-studies in Flanders (rivers Dender and Demer). In both cases, detailed full hydrodynamic models including floodplains were available. They were implemented in the InfoWorks-RS (IWRS) and MIKE11 softwares, by Flanders Hydraulics Research and K.U.Leuven (for MIKE11) and by the Flemish Environment Agency and Soresma/IMDC (for IWRS). The conceptual model building, calibration and validation have been done based on the simulation results for both historical and synthetic flood events. Next to the accuracy of the model results during these events (in terms of agreement with the results of the more detailed full hydrodynamic model), also the robustness and stability of the conceptual model has been evaluated. In one of the two cases, the river Demer case, the conceptual model is currently being applied in studies of realtime regulation of flood control reservoirs. In a research project for the Flemish Environment Agency and in cooperation with the Electrotechnical Department of K.U.Leuven, the technique of Model Predictive Control (MPC) has been tested. This technique allows most optimal regulation of the hydraulic structures and the reservoirs storage (in real time) in order to minimize the flood risk, given the reservoir storage capacity available and taking into the results of flood forecasting. In an ongoing doctoral research study at K.U.Leuven, the use of Genetic Algorithms (GA) is being tested in support of the optimization problem involved in the real-time control. Based on the results so far, real-time control is found a powerful technique to regulate flood control reservoirs in a far more efficient way. The technique allows event-specific optimization of the regulation, consideration of combined regulation strategies (regulation objectives for different types of variables, i.e. river and reservoir levels, and at different locations) and regulation priorities. It also allows minimizing the global flood cost (or maximize the global benefit of the flood regulation) taking into account the (potentially complex) up- and downstream interactions in a river basin, the combined effect of different reservoirs, and considering the sum of all flood consequences (including economical, social and ecological impacts) in an entire river basin. 9

12 International Symposium on Innovations in Flood Forecasting Systems SNOW 4 - An Operational Model for Estimating Precipitation Supply Author: Uwe Böhm Reick Schneider Deutscher Wetterdienst, Lindenberger Weg 24, Berlin, Germany uwe.boehm@dwd.de Abstract SNOW 4 is a grid-based model for simulating snow cover formation and depletion. The core of the model consists of a set of physical equations which describe the snow cover energy and mass balance. Based on the snow surface energy balance, the melting heat is calculated as a result of the radiation balance and heat fluxes between atmosphere, soil and snow cover. Depending on the available melting heat, melting of snow or freezing of liquid water within the snow layer takes place. Retention, aging and regeneration are taken into consideration. Regionalised observations are used both to define the initial state for a 30-hour analysis and force the model. Hourly measurements of air temperature, water vapour pressure, wind speed, global radiation and precipitation are interpolated to the model grid. For a forecast period of up to 3 days, SNOW 4 obtains the required input data from the operational products of the COSMO-EU weather forecast model. The size of a grid box is approximately 1km2, and the model area covers a surface of 750x1000km2 centred over Germany (including most parts of the catchment areas of the rivers Rhine, Moselle, Danube and Elbe). The internal time step is set to 1 hour. Once a day, the compliance between model and regionalized snow cover data is assessed. If discrepancies exceed certain thresholds, the model must be adjusted. The model computes hourly snow cover water equivalents and precipitation supply from snow melt and rain falling into the snow layer (Fig. 1). The model simulations are updated every six hours based on the most recent observations and weather forecasts. The model has been preoperationally tested during the winter 2009/2010. A comprehensive evaluation focussing on water equivalent gave evidence for a good overall performance and an added value compared to the driving COSMO-EU model. For the first time, SNOW 4 is used fully operationally between October 2010 and June 2011, supplying users with input data for flood warning and short-term forecasting systems. SNOW 4 products have so far been in use at the German Federal Institute for Hydrology, the flood forecasting authorities of ten German Länder and the Government of the Austrian Federal Land of Vorarlberg (Fig.2). Figure 1: Output snapshot of SNOW 4 Figure 2: User supply regions of SNOW 4 10

13 A parallelization approach for multi-model real time flood forecasting of the Po river in Italy. Authors: A. Agnetti [2], L. Casicci[1], S. Pecora[2], F.Tonelli [2], M. Vergnani[1] [1] AIPO Interregional Agency for the Po River, [2] ARPA-EMR, Environmental Agency of Emilia Romagna Region, Via Garibaldi 75, Parma, Italy Abstract The Po basin is of great economic, industrial and environmental importance in Italy. It extends over an area of about km2 and includes six regions: Lombardia, Piemonte, Liguria, Emilia-Romagna, Veneto, Valle d Aosta and the autonomous province of Trento. Though not exceptionally large by european standards, the Po basin displays considerable variety and complexity in relation to population density, productive enterprises and water utilization. The lower part of the basin is prone to hydrogeological disasters related to high-impact weather events which can cause severe flooding by the Po river and its tributaries. In order to accomplish the 2007/60 European flood directive requirements, an innovative flood forecasting system, based on deterministic and ensemble techniques, is developed to face hydrogeological risk situations and to define alert strategies on the most important interregional Italian basin. The system was developed under an agreement among the Italian Department of Civil Protection, the Po river basin Autority, the Interregional Agency for the Po river, Region Emilia- Romagna, Region Lombardia, Region Piemonte, Region Valle d Aosta and Region Veneto. The Po river basin is covered by a dense real time observational network, with rain-gauges, thermometers and hydrometers, as well as a radar network which ensures a complete spatial coverage of Northern Italy. The flood forecasting system is fed by data coming from the monitoring network as well as by the meteorological forecasts of the atmospheric deterministic model COSMO-LAMI and of the limited-area ensemble prediction system COSMO- LEPS, both based on the non-hydrostatic limitedarea model COSMO. At the moment, the forecasting system is composed by three different hydrological and hydraulic chains, based on the hydrological models NAM, HEC-HMS and TOPKAPI and on the hydraulic models Mike11, HEC-RAS, Sobek and PAB. Each chain consists of a hydrological model simulating the response of the basins and a river routing model for flood wave propagation. The forecasts are performed in a deterministic and probabilistic framework in which the three operational hydrological and hydraulic chains are fed by observations, COSMO LAMI and LEPS. The resulting forecasts are then post-processed with a 11

14 statistical approach to make the results more readable and useful to the end user. The real time ensemble forecasting system produce a new forecast every 3 or 4 hours with a lead time of 72 or 120 hours, depending on the hydrological/hydraulic chain. This big amount of information regarding the meteorological uncertainty of the prediction system is not the only important plus of the system; also, the multi-model approach helps to take into account the hydrological/hydraulic model and calibration uncertainty. These models are encapsulated in the system, which enables to connect the different modules and to handle them via an open interface. The core software of the system, named Flood Early Warning System (FEWS), incorporates a comprehensive library of general data handling and quality control utilities. The philosophy of the system is to provide an open system to allow any kind of external modules (for example third parties forecasting models) to be used. The FEWS system provides a state of the art flood forecasting and warning system. A central management system called Master Controller collects all external data source in a single central data base and send those data to the Forecasting Shell Servers that are in charge of running the forecasting chains at scheduled intervals. Model results are then imported again in the central data base and then disseminated to the FEWS clients (forecasters). Results can be also distributed through the internet (html pages) or in other forms (fax, and SMS). The system is also linked with a third-parties open source software named CONDOR for grid computing (parallelization) of the ensemble forecasting processes. Condor is a specialized workload management system for compute-intensive jobs and provides a job queueing mechanism, scheduling policy, priority scheme, resource monitoring, and resource management. Users submit their serial or parallel jobs to Condor, which places them into a queue, chooses when and where to run the jobs based upon a policy, carefully monitors their progress, and ultimately informs the user upon completion. 12