Real-time Flood Forecasting Research Articles

Classified real-time flood forecasting by coupling fuzzy clustering and neural network

This paper presented a new classified real-time flood forecasting framework by integrating a fuzzy clustering model and neural network with a conceptual hydrological model. A fuzzy clustering model was used to classify historical floods in terms of flood peak and runoff depth, and the conceptual hydrological model was calibrated for each class of floods. A back-propagation (BP) neural network was trained by using real-time rainfall data and outputs from the fuzzy clustering model. BP neural network provided a rapid on-line classification for real-time flood events. Based on the on-line classification, an appropriate parameter set of hydrological model was automatically chosen to produce real-time flood forecasting. Different parameter sets was continuously used in the flood forecasting process because of the changes of real-time rainfall data and on-line classification results. The proposed methodology was applied to a large catchment in Liaoning province, China. Results show that the classified framework provided a more accurate prediction than the traditional non-classified method. Furthermore, the effects of different index weights in fuzzy clustering were also discussed.

A conceptual and neural network model for real-time flood forecasting of the Tiber River in Rome

Rome is at risk from flooding when extreme events with a return period of about 200 years occur. For this reason, an accurate real-time flood forecasting system may be a useful non-structural countermeasure. Two different approaches are considered to develop a real-time forecasting system capable of predicting hourly water levels at Ripetta stream gauging station in Rome. The first is an adaptive, conceptual model (TFF model), which consists of a rainfall-runoff model that simulates the contribution of 41 ungauged sub-basins (covering approximately 30% of the catchment area) of the Tiber River and a hydraulic model to route the flood through the hydrographic network. The rainfall-runoff model is calibrated online during each flood event at every time step via an adaptive procedure while the flood routing model parameters were calibrated offline and held constant during the forecast. The second approach used is a data-driven one through the application of an artificial neural network (TNN model). Feedforward networks trained with backpropagation and Bayesian regularization were developed using a continuous historical dataset. Both models were used to forecast the most recent significant floods that occurred in Rome (November 2005 and December 2008) with lead times of 12 and 18 h. The results show good performance using both models when compared with observations for a series of absolute and relative performance measures as well as a visual inspection of the hydrographs. At present both models are suitable for real-time forecasting and the power of an integrated approach is still to be investigated.

Real-time forecast of the 2005 and 2007 summer severe floods in the Huaihe River Basin of China

Summary We have developed a one-way coupled hydro-meteorological modeling system consisting of the mesoscale atmospheric model MC2 (Canadian Mesoscale Compressible Community), the Chinese Xinanjiang hydrological model for runoff generation, a flow routing model, and a module for acquiring real-time gauge precipitation. The system had been successfully tested in a hindcast mode using a total of 18 meteorological cases from 1998 and 2003 in the Huaihe River Basin (HRB; 270,000 km 2 ) of China, and has been used to generate daily precipitation and flood forecasts in real-time for the 2005, 2006 and 2007 flooding season over the Wangjiaba sub-basin (30,500 km 2 ), part of the HRB. We run MC2 daily to produce a 96-h precipitation forecast, and then use the combined gauge-model precipitation to drive the hydrological model off-line to forecast the hydrograph at the Wangjiaba Station that is at the outlet of the Wangjiaba sub-basin. We examine the daily forecasts for the two most severe flood events encountered in the past three flooding seasons. The two events occurred in July 4–15, 2005 and June 30–July 25, 2007, which necessitated the use of several flood spillway and flood detention areas along the mainstream of the Huaihe River. A total of 19 daily 96-h precipitation forecasts from the two events are examined. The 19 daily forecasts with different lead times compare reasonably well with observations, although the skill as measured by the MC2 relative error and the MC2 forecast success rate is uneven over a 4-day forecast period. MC2 can better forecast the 96-h accumulation compared to 24-h amounts. We also analyze 10 daily hydrograph forecasts from the two events. The flood peak of the two events at the Wangjiaba Station is predicted well in both timing and intensity with a lead time beyond four days, although the quality of our daily hydrograph forecasts as measured by the relative percentage error of the forecast peak discharge and the Nash–Sutcliffe coefficient is not uniform over different forecast days. It is clear that the skill of the MC2 precipitation has the largest effect on the predicted hydrographs, and the accuracy of daily hydrograph forecasts can be improved substantially using the up-to-date gauge precipitation to complement the MC2 precipitation for driving the hydrological model in real-time flood forecast. Our results demonstrate the applicability and the value of using mesoscale model precipitation for real-time flood forecast over the Wangjiaba sub-basin, which can provide a long lead time of heavy precipitation and subsequent flooding for authorities in operational flood management decision making. The results also illustrate the potential of applying the coupled hydro-meteorological modeling system for real-time flood forecast over other regions.

Improving flood forecasting in Bangladesh using an artificial neural network

A river stage neural network model has been developed to study and predict the water level of Dhaka city. A total of five stations located at the border area of Bangladesh on the Ganges, Brahmaputra and Meghna rivers are selected as input nodes and Dhaka on the Buriganga river is the output node for the neural network. This model is trained with river stage data for a period of 1998 to 2004 and validated with data from 2005 to 2007. The river stage of Dhaka has been predicted for up to ten days with very high accuracy. Values of R2, root mean square and mean absolute error are found ranging from 0.537 to 0.968, 0.607 m to 0.206 m and 0.475 m to 0.154 m, respectively, during training and validation of the model. The results of this study can be useful for real-time flood forecasting by reducing computational time, improving water resources management and reducing the unnecessary cost of field data collection.

Uncertainty in flood estimation

The objective of this contribution is to form a clear picture of uncertainties we encounter in flood estimation, including both real-time flood forecasting and simulation for flood risk estimation. In simulation, we prefer the thesis of equifinality to obtain global optima. Many models producing acceptable simulations can be considered as multiple working hypotheses about the system process representations. Some of those hypotheses might later be confirmed or rejected, given additional data. In GLUE (Generalized Likelihood Uncertainty Estimation) the parameter sets are sampled randomly from physically reasonable ranges, often using uniform sampling where there is no strong information about prior expectations of parameter values. The parameter sets are then used to generate different realizations of the model outputs, which are then evaluated using some criteria (measures of likelihood) to provide a weight associated with each parameter set. Likelihood here is used in a much broader sense than in statistical inference. If some limits of effective observation error can be specified prior to running any simulations, models predicting outside of those limits can then be rejected as non-behavioural. Thus, any model evaluation of this type needs to take account of the multiple sources of model error more explicitly. This, however, is difficult for realistic cases. The procedure for the GLUE methodology is illustrated in examples. Usability for practical problems is suggested and future development is outlined.

Radar-based flood forecasting for river catchments

Weather radars provide high resolution in space and time and thus deliver distributed rainfall information. They therefore have high potential for exploitation in real-time flood forecasting. However, several aspects limit the accuracy of such data in flood forecasting systems. A major aspect is associated with the error included in the radar data and the limited lead time. To assess real-time flood forecasts and improve their reliable lead times with radar rainfall data, this paper presents a framework for a radar-based flood forecasting model integrating radar rainfall data, geographical information systems, the hydrological model HEC—HMS and a short-term precipitation nowcasting system. The benefit of the radar-based flood forecast is analysed and the limits of predictability due to the meteorological aspects of the rainfall forecast as well as the hydrological model are investigated. The incorporation of the radar-based rainfall forecast demonstrates good potential for successful hydrological modelling and prediction.

Integration of Hydrologic Gray Model with Global Search Method for Real-Time Flood Forecasting

This paper presents a hydrologic gray model integrated with a global search method to improve the accuracy of real-time flood forecasting for two watersheds. The model’s applicability is evaluated by comparing the runoff forecasts to the observed values. The model’s accuracy is compared with the accuracy of two base models that employ multiple regression equations and the model capability is verified in real situations. The model parameters are corrected by combining the gray system parameters. The fifth-order differential equation is adopted to represent the characteristics of the study watersheds. The statistical values between the observed values and the runoff forecasts in calibration and validation indicate that the simulations are in close agreement with the observations. The model provides more consistent and satisfactory runoff forecasts than the multiple regression models across all flow ranges; the accuracy of the runoff forecasts varies according to hydrograph stages and lead times. These resul...

Real-time flood forecasting of the Tiber river in Rome

The aim of present work is to investigate the real applicability of real time model of flood forecasting in Tiber river in Rome in order to assess if the lead time is compatible with civil protection actions.Three different models were applied: a simple regression model based on the results obtained by means of a Monte Carlo analysis; a linear model; a quasi-physically based model performing rainfall-runoff analysis. These models were applied to forecast historical floods.It is confirmed that the regression and the linear regression models seem more appropriate for flood forecasting in Rome. Finally, by applying the regression model, the optimal threshold discharges for warnings systems are identified by means of a Bayesian decisional model.

저류함수를 이용한 일단위 장기유출모의 모형 구축

강우-유출 및 저수지운영 모형의 최종 개발 목적은 가용한 관측자료를 이용하여 실무자가 하천의 유량예측 및 저수지운영 업무를 쉽고 정확하게 수행할 수 있도록 분석도구를 제공하는데 있다. 본 연구에서는 물관리 실무에서 홍수량 예측업무에 활용중인 단기 강우-유출 저류함수 모형을 확장하여 장기유출에 모의에 기여도가 높은 유효우량 및 침투량 산정기법을 개선하여 장기 유출분석이 가능하도록 저류함수기반의 장기 유출모형을 구축하였다. 개발된 모형의 적용가능성을 검토하기 위하여 낙동강 유역내 3개 댐유역(안동,합천,밀양)을 대상으로 연간 유출모의를 실시 하였으며, 현재 물관리 실무에서 장기유출모형으로 활용중인 SSARR모형과의 성능을 비교하였다. 수문곡선 비교 결과, SSARR 모형이 상대적으로 재현성 높은 결과를 보여주었지만, 본 연구에서 개발한 모형은 관측 가용한 자료만으로도 비교적 신뢰성 있는 장기유출모의가 가능하다는 점에서 향후 실무에서 유용하게 활용할 수 있을 것으로 사료된다. The purpose of developing a rainfall-runoff and reservoir model is to provide an analysis tool for hydrological engineers in order to forecast discharge of rivers and to accomplish reservoir operations easily and accurately. In this study, based on the short-term rainfall-runoff storage function model which has gained popularity for real time flood forecast in practical water management affairs, a long-term runoff model was developed for the improvement of the calculation method of effective rainfall and percolation at the infiltration area. Annual discharge was simulated for three dam watersheds(Andong, Hapcheon, Milyang) in Nakdong River basin to analyze the accuracy of the developed model and compare it to SSARR model, which is used as the long-term runoff model in current practical water management affairs. As the result of the comparison of hydrographs, SSARR model showed relatively better results. However, it is possible for the developed model to simulate reliable long-term runoff using relatively little available data and is useful for hydrological engineers in practical affairs.

Real-time flood forecast and flood alert map over the Huaihe River Basin in China using a coupled hydro-meteorological modeling system

A coupled hydro-meteorological modeling system is established for real-time flood forecast and flood alert over the Huaihe River Basin in China. The system consists of the mesoscale atmospheric model MC2 (Canadian Mesoscale Compressible Community) that is one-way coupled to the Chinese Xinanjiang distributed hydrological model, a grid-based flow routing model, and a module for acquiring real-time gauge precipitation. The system had been successfully tested in a hindcast mode using 1998 and 2003 flood cases in the basin, and has been running daily in a real-time mode for the summers of 2005 and 2006 over the Wangjiaba sub-basin of the Huaihe River Basin. The MC2 precipitation combined with gauge values is used to drive the Xinanjiang model for hydrograph prediction and production of flood alert map. The performance of the system is illustrated through an examination of real-time flood forecasts for the severe flood case of July 4–15, 2005 over the sub-basin, which was the first and largest flood event encountered to date. The 96-h forecasts of MC2 precipitation are first evaluated using observations from 41 rain gauges over the sub-basin. The forecast hydrograph is then validated with observations at the Wangjiaba outlet of the sub-basin. MC2 precipitation generally compares well with gauge values. The flood peak was predicted well in both timing and intensity in the 96-hour forecast using the combined gauge-MC2 precipitation. The real-time flood alert map can spatially display the propagation of forecast floods over the sub-basin. Our forecast hydrograph was used as operational guidance by the Bureau of Hydrograph, Ministry of Water Resources. Such guidance has been proven very useful for the Office of State Flood Control and Drought Relief Headquarters in operational decision making for flood management. The encouraging results demonstrate the potential of using mesoscale atmospheric model precipitation for real-time flood forecast, which can result in a longer lead time compared to traditional methods.

Verification of operational Quantitative Discharge Forecast (QDF) for a regional warning system – the AMPHORE case studies in the upper Po River

Abstract. In recent years, the interest in the prediction and prevention of natural hazards related to hydrometeorological events has grown due to the increased frequency of extreme rainstorms. Several research projects have been developed to test hydrometeorological models for real-time flood forecasting. However, flood forecasting systems are still not widespread in operational context. Real-world examples are mainly dedicated to the use of flood routing model, best suited for large river basins. For small basins, it is necessary to take advantage of the lag time between the onset of a rainstorm and the beginning of the hydrograph rise, with the use of rainfall-runoff transformation models. Nevertheless, when the lag time is very short, a rainfall predictor is required, as a result, meteorological models are often coupled with hydrological simulation. While this chaining allows floods to be forecasted on small catchments with response times ranging from 6 to 12 h it, however, causes new problems for the reliability of Quantitative Precipitation Forecasts (QPF) and also creates additional accuracy problems for space and time scales. The aim of this work is to evaluate the degree to which uncertain QPF affects the reliability of the whole hydro-meteorological alert system for small catchments. For this purpose, a distributed hydrological model (FEST-WB) was developed and analysed in operational setting experiments, i.e. the hydrological model was forced with rain observation until the time of forecast and with the QPF for the successive period, as is usual in real-time procedures. Analysis focuses on the AMPHORE case studies in Piemonte in November 2002.

The use of meteorological analogues to account for LAM QPF uncertainty

Abstract. Flood predictions based on quantitative precipitation forecasts (QPFs) provided by deterministic models do not account for the uncertainty in the outcomes. A probabilistic approach to QPF, one which accounts for the variability of phenomena and the uncertainty associated with a hydrological forecast, seems to be indispensable to obtain different future flow scenarios for improved flood management. A new approach based on a search for analogues, that is past situations similar to the current one under investigation in terms of different meteorological fields over Western Europe and East Atlantic, has been developed to determine an ensemble of hourly quantitative precipitation forecasts for the Reno river basin, a medium-sized catchment in northern Italy. A statistical analysis, performed over a hydro-meteorological archive of ECMWF analyses at 12:00 UTC relative to the autumn seasons ranging from 1990 to 2000 and the corresponding precipitation measurements recorded by the raingauges spread over the catchment of interest, has underlined that the combination of geopotential at 500 hPa and vertical velocity at 700 hPa provides a better estimation of precipitation. The analogue-based ensemble prediction has to be considered not alternative but complementary to the deterministic QPF provided by a numerical model, even when employed jointly to improve real-time flood forecasting. In the present study, the analogue-based QPFs and the precipitation forecast provided by the Limited Area Model LAMBO have been used as different input to the distributed rainfall-runoff model TOPKAPI, thus generating, respectively, an ensemble of discharge forecasts, which provides a confidence interval for the predicted streamflow, and a deterministic discharge forecast taken as an error-affected "measurement" of the future flow, which does not convey any quantification of the forecast uncertainty. To make more informative the hydrological prediction, the ensemble spread could be regarded as a measure of the uncertainty of the deterministic forecast.

Do Nash values have value?

How Do We Communicate Model Performance? The process of model performance evaluation is of primary importance, not only in the model development and calibration process, but also when communicating the results to other researchers and to stakeholders. The basic ‘rule’ is that every modelling result should be put into context, for example, by indicating the model performance using appropriate indicators, and by highlighting potential sources of uncertainty, and this practice has found its entry into the large majority of papers and conference presentations. While the question of how to communicate the performance of a model to potential end-users is currently receiving increasing interest (e.g. Pappenberger and Beven, 2006), we–as well as many other colleagues–observe regularly that researchers take much less care when communicating model performance amongst ourselves. We seem to assume that we are speaking about familiar performance concepts and that they have comparable significance for various types of model applications and case studies. In doing so, we do not pay sufficient attention to making clear what the values represented by our performance measures really mean. Even concepts as simple as the bias between an observed and a simulated time series need to be put into proper context: whereas a 10% bias in simulation of simulated discharge may be unacceptable in a climate change impact assessment, it may be of less concern in the context of real-time flood forecasting. While some performance measures can have an absolute meaning, such as the common measure of linear correlation, the vast majority of performance measures, and in particular quadratic-error-based measures, can only be properly interpreted when viewed in the context of a reference value (..)

APPLICATION OF GREY MODEL AND ARTIFICIAL NEURAL NETWORKS TO FLOOD FORECASTING

: The main focus of this study was to compare the Grey model and several artificial neural network (ANN) models for real time flood forecasting, including a comparison of the models for various lead times (ranging from one to six hours). For hydrological applications, the Grey model has the advantage that it can easily be used in forecasting without assuming that forecast storm events exhibit the same stochastic characteristics as the storm events themselves. The major advantage of an ANN in rainfall-runoff modeling is that there is no requirement for any prior assumptions regarding the processes involved. The Grey model and three ANN models were applied to a 2,509 km2 watershed in the Republic of Korea to compare the results for real time flood forecasting with from one to six hours of lead time. The fifth-order Grey model and the ANN models with the optimal network architectures, represented by ANN1004 (34 input nodes, 21 hidden nodes, and 1 output node), ANN1010 (40 input nodes, 25 hidden nodes, and 1 output node), and ANN1004T (14 input nodes, 21 hidden nodes, and 1 output node), were adopted to evaluate the effects of time lags and differences between area mean and point rainfall. The Grey model and the ANN models, which provided reliable forecasts with one to six hours of lead time, were calibrated and their datasets validated. The results showed that the Grey model and the ANN1010 model achieved the highest level of performance in forecasting runoff for one to six lead hours. The ANN model architectures (ANN1004 and ANN1010) that used point rainfall data performed better than the model that used mean rainfall data (ANN1004T) in the real time forecasting. The selected models thus appear to be a useful tool for flood forecasting in Korea.

Application of parallel computing to stochastic parameter estimation in environmental models

Parameter estimation or model calibration is a common problem in many areas of process modeling, both in on-line applications such as real-time flood forecasting, and in off-line applications such as the modeling of reaction kinetics and phase equilibrium. The goal is to determine values of model parameters that provide the best fit to measured data, generally based on some type of least-squares or maximum likelihood criterion. Usually, this requires the solution of a non-linear and frequently non-convex optimization problem. In this paper we describe a user-friendly, computationally efficient parallel implementation of the Shuffled Complex Evolution Metropolis (SCEM-UA) global optimization algorithm for stochastic estimation of parameters in environmental models. Our parallel implementation takes better advantage of the computational power of a distributed computer system. Three case studies of increasing complexity demonstrate that parallel parameter estimation results in a considerable time savings when compared with traditional sequential optimization runs. The proposed method therefore provides an ideal means to solve complex optimization problems.

Operational hydro-meteorological warning and real-time flood forecasting: the Piemonte Region case study

Abstract. The development and implementation of a real-time flood forecasting system in the context of the Piemonte Region's hydro-meteorological operational alert procedure is described. The area of interest is the Upper Po River basin (north-west Italy) of approximately 37000 km2 and its river network of about 3000 km and three big lakes. FloodWatch, a GIS-based decision support system for real-time flood forecasting, has been developed and used operationally at the Piemonte Region's Room for the Situation of Natural Hazards in Torino, Italy, since January 2000. The system is linked directly to the telemetric gauges system, uses daily quantitative precipitation and temperature forecasts issued by the Regional Meteorological Service and automatically supplies operational forecasts of water-level and discharge at about 30 locations for up to 48 hours. Strengths and limits of the system and its link with operational flood alert and management are discussed. The case study presented is the October 2000 flood event, when the north-west of Italy experienced one of the largest floods on record. Results highlight how the uncertainty linked to the use of meteorological forecasts greatly influences the quality of the hydrological forecasts. The proposed alert procedure, based on coded risk levels, can help effectively in facing forecast uncertainties.

Coupling meteorological and hydrological models for flood forecasting

Abstract. This paper deals with the problem of analysing the coupling of meteorological meso-scale quantitative precipitation forecasts with distributed rainfall-runoff models to extend the forecasting horizon. Traditionally, semi-distributed rainfall-runoff models have been used for real time flood forecasting. More recently, increased computer capabilities allow the utilisation of distributed hydrological models with mesh sizes from tenths of metres to a few kilometres. On the other hand, meteorological models, providing the quantitative precipitation forecast, tend to produce average values on meshes ranging from slightly less than 10 to 200 kilometres. Therefore, to improve the quality of flood forecasts, the effects of coupling the meteorological and the hydrological models at different scales were analysed. A distributed hydrological model (TOPKAPI) was developed and calibrated using a 1x1 km mesh for the case of the river Po closed at Ponte Spessa (catchment area c. 37000 km2). The model was then coupled with several other European meteorological models ranging from the Limited Area Models (provided by DMI and DWD) with resolutions from 0.0625° * 0.0625°, to the ECMWF ensemble predictions with a resolution of 1.85° * 1.85°. Interesting results, describing the coupled model behaviour, are available for a meteorological extreme event in Northern Italy (Nov. 1994). The results demonstrate the poor reliability of the quantitative precipitation forecasts produced by meteorological models presently available; this is not resolved using the Ensemble Forecasting technique, when compared with results obtainable with measured rainfall.

STORM<sup>3</sup>: a new flood forecast management and monitoring system in accordance with the recent Italian national directive

Abstract. The effectiveness of alert systems for civil protection purposes, defined as the ability to minimize the level of risk in a region subjected to an imminent flood event, strongly depends on availability and exploitability of information. It also depends on technical expertise and the ability to easily manage the civil protection actions through the organization into standardized procedures. Hydro-geologic and hydraulic risk estimation, based on the combination of different technical issues (in this case meteorological, hydro-geological, hydraulic matters), but also socio-economic ones, requires the integration between quasi-static and time-varying information within the same operative platform. Beside the real-time data exchange, a Decision Support System must provide tools which enable knowledge sharing among the civil protection centres. Moreover, due to the amount and heterogeneity of information, quality procedures become necessary to handle all forecasting and monitoring routines within operative centres, according to the latest national directive. In Italy procedures on the civil protection matter have been condensed into the Prime Minister's Directive (27 February 2004. STORM3, an innovative management and monitoring System for real-time flood forecasting and warning, takes in the Directive, supporting the operator step by step within the different phases of civil protection activities.

Fuzzy computing based rainfall-runoff model for real time flood forecasting

This paper analyses the skills of fuzzy computing based rainfall–runoff model in real time flood forecasting. The potential of fuzzy computing has been demonstrated by developing a model for forecasting the river flow of Narmada basin in India. This work has demonstrated that fuzzy models can take advantage of their capability to simulate the unknown relationships between a set of relevant hydrological data such as rainfall and river flow. Many combinations of input variables were presented to the model with varying structures as a sensitivity study to verify the conclusions about the coherence between precipitation, upstream runoff and total watershed runoff. The most appropriate set of input variables was determined, and the study suggests that the river flow of Narmada behaves more like an autoregressive process. As the precipitation is weighted only a little by the model, the last time-steps of measured runoff are dominating the forecast. Thus a forecast based on expected rainfall becomes very inaccurate. Although good results for one-step-ahead forecasts are received, the accuracy deteriorates as the lead time increases. Using the one-step-ahead forecast model recursively to predict flows at higher lead time, however, produces better results as opposed to different independent fuzzy models to forecast flows at various lead times. Copyright © 2004 John Wiley & Sons, Ltd.

Coupling between weather radar rainfall data and a distributed hydrological model for real-time flood forecasting / Couplage de données pluviométriques issues de radars météorologiques et d’un modèle hydrologique distribué pour la prévision de crue en temps réel

After the destructive flood in 1998, the Chinese government planned to build national weather radar networks and to use radar data for real-time flood forecasting. Hence, coupling of weather radar rainfall data and a hydrological (Xinanjiang) model became an important issue. The present study reports on experience in such coupling at the Shiguanhe watershed. After having corrected the radar reflectivity and the attenuation data, the weather radar rainfall was estimated and then corrected in real time using a Kalman filter. In general, the precipitation estimated from weather radar is reasonably accurate in most of the catchment investigated, after corrections as above. Compared to the results simulated by raingauge data, the simulations based on the weather radar data are of similar accuracy. Present research results show that rainfall estimated from the weather radar, the radar data correction method, the method of coupling, and the Xinanjiang model lend themselves well to application in operational real-time flood forecasting.