Figure - available from: Water Resources Management
This content is subject to copyright. Terms and conditions apply.
Source publication
Predicting extreme events is one of the major goals of streamflow forecasting, but models that are reliable under such conditions are hard to come by. This stems in part from the fact that, in many cases, calibration is based on recorded time series that do not comprise extreme events. The problem is particularly relevant in the case of data-driven...
Similar publications
The heavy rainfall event that occurred on 5–6 July 2017 in Northern Kyushu, Japan, caused extensive flooding across several mountainous river basins and resulted in fatalities and extensive damage to infrastructure along those rivers. For the periods before and during the extreme event, there are no hydrological observations for many of the flooded...
Citations
... The central premise for a surrogate model to provide results consistent with the original model is to be able to approximate the relationship between input and output similar to the original model (Smith, 2013;Asher et al., 2015). However, the current surrogate models based on this relationship cannot provide a reliable prognosis for outliers (or extremes) beyond the training data space, although they generally have excellent predictive power for regions within the training data (Razavi et al., 2012b;Asher et al., 2015;Matos et al., 2017;Tran et al., 2020). This is because the surrogate model is adapted locally to a constrained number of training points (also referred to as design sites), and thus only sites close to the training space can be diagnosed (Bowden et al., 2012). ...
... For example, in the context of water resource problems, there will be a high probability of extreme events that, due to climate change, have not been experienced in the past (Prein et al., 2016;Bao et al., 2017;Bloschl et al., 2020). There is also the possibility that extreme events that deviate from recorded events will occur due to climate internal variability, even assuming that climate stationarity is maintained (Matos et al., 2017;Kim et al., 2018;Doi and Kim, 2020;2021). Therefore, a common solution for ensuring the predictive power in the entire data space is to expand the data range of the design site to cover all possible cases (Schöbi et al., 2017). ...
This study presents the strengths of polynomial chaos-kriging (PCK), a new surrogate model that merges polynomial chaos extension (PCE) and Gaussian process with kriging variance. This combination enabled streamflow prediction for extreme events that deviated significantly from the trained data space, and allowed for quantifying predictive uncertainty robustly and efficiently. The uncertainty quantification results to eight testing flood events through a modeling framework that applies generalized likelihood uncertainty estimation (GLUE) to surrogate models are as follows. (1) PCK outperformed PCE and ordinary kriging (OK) in mimicking predictive and sensitive behaviors of the original model with a smaller-sized training dataset. (2) Three surrogate models trained on the identical dataset exhibited equivalent predictability with the original model for six smaller events similar to their training data space. However, for two extreme events, which differed significantly from the training set, only PCK was found to accurately predict the hydrograph and flood peaks. (3) Since two types of acceptance thresholds, defined here as “accuracy-aimed” or “efficiency-aimed” threshold, have their own pros and cons, the type and size of the threshold should be determined depending on the availability of computational resources and the degree of accuracy needed. (4) A new “performance score” is proposed here to assess the overall performance of the surrogate models. This compensates for situations in which the performance of a surrogate model can be misjudged through individual indices of efficiency or accuracy in the process of uncertainty quantification.
... Despite the fact that the AI-based models are the suitable in simulating and forecasting streamflow time series, there are some problems with their application under the condition of highly nonlinear or chaos-based streamflow data. In last 5 years, the nonlinear AI models exhibited the outstanding results with a harmony between the forecasted and observed streamflow data (i.e., Huang et al., 2014;Chen et al., 2015;Meshgi et al., 2015;Zhang et al., 2015;Deo and Sahin, 2016;Kasiviswanathan et al., 2016;Yaseen et al., 2016a;Matos et al., 2018;Mehdizadeh and Sales, 2018;Guo et al., 2020;Niu et al., 2021). ...
Reliable and realistic streamflow forecasting is very important in hydrology, hydraulic, and water resources engineering as it can directly affect the dams operation and performance, groundwater recharge/exploitation, sediment conveyance capability of river, watershed management, etc. However, an accurate streamflow forecasting is not an easy task due to the high uncertainty associated with climate conditions and complexity of collecting and handling both spatial and non-spatial data. Therefore, hydrologists from all over the world have developed and adopted several types of data-driven techniques ranging from traditional stochastic time-series modeling to modern hybrid artificial intelligence models for future prediction of streamflow. In literature, studies dealing with streamflow forecasting used a variety of techniques having dissimilar concepts and characteristics, and streamflow datasets at different time scale such as daily, monthly, seasonal and yearly etc. This chapter first describes and classifies available data-driven techniques used in streamflow forecasting into suitable groups depending upon their characteristics. Then, growth of the salient data-driven models both single and hybrid such as time-series models, artificial neural network models, and other artificial intelligence models is discussed with their applications and comparisons as reported in studies on streamflow forecasting over time. Thereafter, current approaches used in the recent five-year streamflow-forecasting studies are briefly summarized. Also, challenges experienced by the researchers in applying data-driven techniques for streamflow forecasting are addressed. It is concluded that a vast scope exists for improving streamflow forecasts using emerging and modern tools and combining them with location-specific and in-depth knowledge of the physical processes occurring in the hydrologic system.
... Many researchers and scientists have tried to apply the ANN technology to solve flood forecast problems and have achieved meaningful results [17][18][19][20][21][22][23][24][25][26][27][28][29][30]. However, due to the high nonlinearity inherent in the flood forecast, the training and testing accuracies usually cannot be both satisfactory which indicate that the forecast capability and stability is not good enough. ...
Machine learning-based data-driven models have achieved great success since their invention. Nowadays the artificial neural network-based machine learning methods have made great progress than ever before, such as the deep learning and reinforcement learning, etc. In this study, we coupled the artificial neural network with the K-nearest neighbor method to propose a novel hybrid machine learning (HML) hydrological model for flood forecast purpose. The advantage of the proposed model over traditional neural network models is that it can predict discharge continuously without accuracy loss owe to its specially designed model structure. In order to overcome the local minimum issue of the traditional neural network training, a genetic algorithm and Levenberg-Marquardt-based multi-objective training method was also proposed. Real-world applications of the HML hydrological model indicated its satisfactory performance and reliable stability, which enlightened the possibility of further applications of the HML hydrological model in flood forecast problems.
... In recent years, with the improvement of hydrological data acquisition technology and the development of artificial intelligence computing, data-driven flood forecasting models have gained increasing attention [1][2][3][4][5][6][7][8]. The development of intelligent computing has gone through three important stages. ...
Sudden floods in the medium and small watershed by a sudden rainstorm and locally heavy rainfall often lead to flash floods. Therefore, it is of practical and theoretical significance to explore appropriate flood forecasting model for medium and small watersheds for flood control and disaster reduction in the loess region under the condition of underlying surface changes. This paper took the Gedong basin in the loess region of western Shanxi as the research area, analyzing the underlying surface and floods characteristics. The underlying surface change was divided into three periods (HSP1, HSP2, HSP3), and the floods were divided into three grades (great, moderate, small). The paper applied K Nearest Neighbor method and Fireworks Algorithm to improve the Extreme Learning Machine model (KNN-FWA-ELM) and proposed KNN-FWA-ELM hybrid flood forecasting model, which was further applied to flood forecasting of different underlying surface conditions and flood grades. Results demonstrated that KNN-FWA-ELM model had better simulation performance and higher simulation accuracy than the ELM model for flood forecasting, and the qualified rate was 17.39% higher than the ELM model. KNN-FWA-ELM model was superior to the ELM model in three periods and the simulation performance of three flood grades, and the simulation performance of KNN-FWA-ELM model was better in HSP1 stage floods and great floods.
This work proposes a new statistical modelling approach to forecast the hourly river level at a gauging station, under potential flood risk situations and over a medium-term prediction horizon (around three days). For that aim we introduce a new model, the switching regression model with ARMA errors, which takes into account the serial correlation structure of the hourly level series, and the changing time delay between them. A whole modelling approach is developed, including a two-step estimation, which improves the medium-term prediction performance of the model, and uncertainty measures of the predictions. The proposed model not only provides predictions for longer periods than other statistical models, but also helps to understand the physics of the river, by characterizing the relationship between the river level in a gauging station and its influential factors. This approach is applied to forecast the Ebro River level at Zaragoza (Spain), using as input the series at Tudela. The approach has shown to be useful and the resulting model provides satisfactory hourly predictions, which can be fast and easily updated, together with their confidence intervals. The fitted model outperforms the predictions from other statistical and numerical models, specially in long prediction horizons.
