Long-term Hydrological Forecasts Research Articles

Two Strategies for Avoiding Overfitting Long-Term Forecasting Models: Downsampling Predictor Fields and Shrinking Coefficients

Long-term hydrological forecasting based on sea surface temperature (SST) fields faces the large p and small n problem, i.e., too many potential predictors and a limited number of samples. Considering the selection of predictors will also enhance the complexity of models and lead to overfitting, in this study, two strategies are used for building forecasting models for long-term streamflow forecasting. The first strategy is to downsample the SST field and optimize its spatial resolution; the second is to shrink model coefficients based on L1 regularization. We build models based on the downsampled SST fields with different spatial resolutions. It is found that the model based on a proper spatial resolution always performs better than the model based on the raw SST field. This result suggests that it is better to treat the spatial resolution of the predictor field as a hyperparameter, which is similar to hyperparameters for controlling the complexity of many machine learning models. For applying the second strategy, L1 norm regularization models, including (1) least absolute selection and shrinkage operator (LASSO), (2) relaxed LASSO, and (3) two-step approach of LASSO and ordinary least squares regression (LASSO+OLS) are explored. We have found that the relaxed LASSO model always performs better than the ordinary LASSO, indicating that relaxed LASSO is a better shrinkage approach. Furthermore, the skills of the presented models are compared with the stepwise regression, and the lower skill of the stepwise regression based on SST fields with a high spatial resolution suggests that one should not select predictors based on fields with high resolutions considering the limited number of samples.

Medium- and Long-Term Precipitation Forecasting Method Based on Data Augmentation and Machine Learning Algorithms

Accurate medium and long-term precipitation forecasting plays a vital role in disaster prevention and mitigation and rational allocation of water resources. In recent years, there are various methods for medium- and long-term precipitation forecasting based on machine learning algorithms. However, machine learning has a high demand for the size of sample data. Therefore, this article proposes a data augmentation algorithm based on the K-means clustering algorithm and synthetic minority oversampling technique (SMOTE), which can effectively enhance sample information. Besides, through constructing random forest (RF), extreme gradient boosting (XGB), recurrent neural network (RNN), and long short-term memory (LSTM) are, respectively, constructed as the models to forecast monthly grid precipitation of the Danjiangkou River Basin. This study aims to improve the accuracy of medium- and long-term precipitation forecasting. The main results are the following two aspects: 1) in most years, the anomaly correlation coefficient and Pg score of SMOTE-km-XGB and SMOTE-km-RF exceed that of XGB and RF; furthermore, compared with the other three methods, SMOTE-km-XGB method is more suitable for precipitation forecasting in the studied basin in this article; and 2) the forecasting results of two deep learning methods (RNN and LSTM) show that the sample data processed by the K-means clustering algorithm and SMOTE data augmentation algorithm have not achieved considerable results in deep learning. This study improves the accuracy of precipitation forecast by expanding and balancing the information of sample data, and provides a new research idea for improving the accuracy of medium- and long-term hydrological forecasting.

Implementation of the TOPKAPI model in South Africa: Initial results from the Liebenbergsvlei catchment

Flash floods and droughts are of major concern in Southern Africa. Hydrologists and engineers have to assist decision makers to address the issue of forecasting and monitoring extreme events. For these purposes, hydrological models are useful tools to: Identify the dominant hydrological processes which influence the water balance and result in conditions of extreme water excess and / or deficit Assist in generating both short- and long-term hydrological forecasts for use by water resource managers. In this study the physically-based and fully distributed hydrological TOPKAPI model (Liu and Todini, 2002), which has already been successfully applied in several countries in the world (Liu and Todini, 2002; Bartholomes and Todini, 2005; Liu et al., 2005; Martina et al., 2006), is applied in Africa for the first time. This paper contains the main theoretical and numerical components that have been integrated by the authors to model code and presents details of the application of the model in the Liebenbergsvlei catchment (4 625 km2) in South Africa. The physical basis of the equations, the fine-scale representation of the spatial catchment features, the parsimonious parameterisation linked to field / catchment information, the good computation time performance, the modularity of the processes, the ease of use and finally the good results obtained in modelling the river discharges of Liebenbergsvlei catchment, make the TOPKAPI model a promising tool for hydrological modelling of catchments in South Africa.

Hybrid Models Combining EMD/EEMD and ARIMA for Long-Term Streamflow Forecasting

Long-term streamflow forecast is of great significance for water resource application and management. However, accurate monthly streamflow forecasting is challenging due to its non-stationarity and uncertainty. Time series analysis methods have been proved to perform well in stationary time series forecasting, which can be derived from decomposition of the non-stationary sequence. As common decomposition methods in time domain, Empirical Mode Decomposition (EMD) and Ensemble Empirical Mode Decomposition (EEMD) are selected to decompose the components with different time-scale characteristics in the original hydrological time series in this study. The derived components are proved to be stationary by the stationarity test. Thus, Autoregressive Integrated Moving Average (ARIMA) model, a simple and effective time series analysis method, is used to forecast the components. A hybrid EMD/EEMD-ARIMA model is proposed in this study for long-term streamflow forecasting, which is applied to the upper stream of the Yellow River. The original daily streamflow time series of six years at the Tangnaihai station are firstly decomposed by EMD/EEMD into several stationary or simple non-stationary sub-series to explore detailed data information with different time scales. ARIMA models with appropriate parameters are then established for each subsequence to forecast the stream flow of the next year. Predicted ten-day and monthly stream flow is finally obtained combing the predictions of all the components. The EMD-ARIMA hybrid model performs best in forecasting high and moderate value of streamflow and fits best with the observation compared with EEMD-ARIMA and ARIMA models. The results not only verify the effectiveness of the proposed hybrid EMD/EEMD-ARIMA model in exploiting comprehensive information to improve the prediction but also indicate that the EMD-ARIMA model with end points disposal performs the best and can be used for long-term hydrological forecasting.

Verification of ECMWF System 4 for seasonal hydrological forecasting in a northern climate

Abstract. Hydropower production requires optimal dam and reservoir management to prevent flooding damage and avoid operation losses. In a northern climate, where spring freshet constitutes the main inflow volume, seasonal forecasts can help to establish a yearly strategy. Long-term hydrological forecasts often rely on past observations of streamflow or meteorological data. Another alternative is to use ensemble meteorological forecasts produced by climate models. In this paper, those produced by the ECMWF (European Centre for Medium-Range Forecast) System 4 are examined and bias is characterized. Bias correction, through the linear scaling method, improves the performance of the raw ensemble meteorological forecasts in terms of continuous ranked probability score (CRPS). Then, three seasonal ensemble hydrological forecasting systems are compared: (1) the climatology of simulated streamflow, (2) the ensemble hydrological forecasts based on climatology (ESP) and (3) the hydrological forecasts based on bias-corrected ensemble meteorological forecasts from System 4 (corr-DSP). Simulated streamflow computed using observed meteorological data is used as benchmark. Accounting for initial conditions is valuable even for long-term forecasts. ESP and corr-DSP both outperform the climatology of simulated streamflow for lead times from 1 to 5 months depending on the season and watershed. Integrating information about future meteorological conditions also improves monthly volume forecasts. For the 1-month lead time, a gain exists for almost all watersheds during winter, summer and fall. However, volume forecasts performance for spring varies from one watershed to another. For most of them, the performance is close to the performance of ESP. For longer lead times, the CRPS skill score is mostly in favour of ESP, even if for many watersheds, ESP and corr-DSP have comparable skill. Corr-DSP appears quite reliable but, in some cases, under-dispersion or bias is observed. A more complex bias-correction method should be further investigated to remedy this weakness and take more advantage of the ensemble forecasts produced by the climate model. Overall, in this study, bias-corrected ensemble meteorological forecasts appear to be an interesting source of information for hydrological forecasting for lead times up to 1 month. They could also complement ESP for longer lead times.

Estimation and Forecast of Components of Runoff in the River Basins of Central Asia

Various options for method of the over long-term hydrological forecasts of annual runoff for several river basins in Central Asia (Amu Darya, Syr Darya, Tarim, Ganges, Indus, Mekong, Brahmaputra, Yangtze, Huang He) are investigated. Annual air temperature, precipitation, and normalized difference of anomalies of these variables according to data of meteostations and for gridded values having 0.5 degree increments in latitude and longitude were tested as inputs. Informative capability of each predictor for calculating runoff in river basins of Central Asia was described. Expected air temperature as a predictor for long-term forecasts of river runoff was obtained by extrapolating the proper temporal series. Simulation, calculation and forecast of glacial runoff and evaporation were performed in the ablation zone of Central Asia glaciers. The input data for the model are: precipitation, average temperature and humidity during the summer, generalized areal, altitudinal and morphological parameters of the glaciation. In order to optimize the method of calculating of regime of glaciation, the populations of glaciers in the basins of the tributaries of the Amu Darya, Syr Darya, Indus, Ganges, Yellow, Yangtze, Brahmaputra and Tarim rivers were allocated to 3815 homogeneous groups of objects. Eight values of altitude and 23 gradations of glacier area are consistently applied as classifi cation criteria. For each group the average and weighted average morphometric characteristics of glaciers were determined. Data from 53 225 glaciers with a total area of 81144 km2 in the basins of Central Asia are used to demonstrate the capabilities of the method. Modelled values of water balance components (runoff, precipitation, and evaporation) were obtained for river basins of Vakhsh, Panj, Zarafshan, Naryn, Brahmaputra.

中长期水文预报的最优组合模型研究

本文建立了径流中长期预报的自回归、季节性自回归、门限自回归、最近邻抽样回归、人工神经网络、支持向量机等六种模型，并对这些模型结果进行综合，开展最优组合预报。以飞来峡水库为研究实例，选取平均绝对误差和均方误差作为评价指标，发现人工神经网络模型模拟精度较高；支持向量机模型模拟精度高、且具有最好的预报性能；最优组合预报模型综合各单一预报模型的优点，结果稳健、通用性强。 This paper applies six models, including the autoregressive model, the seasonal autoregressive model, the threshold autoregressive model, the nearest neighbor bootstrap regressive model, the artificial neural network model and the support vector machine model into the mid and long-term hydrological forecasting. Based on Feilaixia reservoir project, the results show that the artificial neural network model is able to time series very well. The support vector machine model has the powerful ability of not only the simulation but also the forecasting. The results of those models were combined by the optimal combined forecasting model. The mean absolute error and the mean square error are selected as the measurements. Relying on the merits of each single model, the results of the optimal combined forecasting model work very well and are very well in robustness.

Mechanism and forecasting methods for severe droughts and floods in Songhua River Basin in China

The influence of various factors, mechanisms, and principles affecting runoff are summarized as periodic law, random law, and basin-wide law. Periodic law is restricted by astronomical factors, random law is restricted by atmospheric circulation, and basin-wide law is restricted by underlying surface. The commensurability method was used to identify the almost period law, the wave method was applied to deducing the random law, and the precursor method was applied in order to forecast runoff magnitude for the current year. These three methods can be used to assess each other and to forecast runoff. The system can also be applied to forecasting wet years, normal years and dry years for a particular year as well as forecasting years when floods with similar characteristics of previous floods, can be expected. Based on hydrological climate data of Baishan (1933–2009) and Nierji (1886–2009) in the Songhua River Basin, the forecasting results for 2010 show that it was a wet year in the Baishan Reservoir, similar to the year of 1995; it was a secondary dry year in the Nierji Reservoir, similar to the year of 1980. The actual water inflow into the Baishan Reservoir was 1.178 × 1010 m3 in 2010, which was markedly higher than average inflows, ranking as the second highest in history since records began. The actual water inflow at the Nierji station in 2010 was 9.96 × 109 m3, which was lower than the average over a period of many years. These results indicate a preliminary conclusion that the methods proposed in this paper have been proved to be reasonable and reliable, which will encourage the application of the chief reporter release system for each basin. This system was also used to forecast inflows for 2011, indicating a secondary wet year for the Baishan Reservoir in 2011, similar to that experienced in 1991. A secondary wet year was also forecast for the Nierji station in 2011, similar to that experienced during 1983. According to the nature of influencing factors, mechanisms and forecasting methods and the service objects, mid- to long-term hydrological forecasting can be divided into two classes: mid- to long-term runoff forecasting, and severe floods and droughts forecasting. The former can be applied to quantitative forecasting of runoff, which has important applications for water release schedules. The latter, i.e., qualitative disaster forecasting, is important for flood control and drought relief. Practical methods for forecasting severe droughts and floods are discussed in this paper.

Singular Spectrum Analysis and ARIMA Hybrid Model for Annual Runoff Forecasting

High accuracy forecasting of medium and long-term hydrological runoff is beneficial to reservoir operation and management. A hybrid model is proposed for medium and long-term hydrological forecasting in this paper. The hybrid model consists of two methods, Singular Spectrum Analysis (SSA) and Auto Regressive Integrated Moving Average (ARIMA). In this model, the time series of annual runoff are first decomposed into several sub-series corresponding to some tendentious and periodic motions by using SSA and then each sub-series is predicted, respectively, through an appropriate ARIMA model, and lastly a correction procedure is conducted for the sum of the prediction results to ensure the superposed residual to be a pure random series. The annual runoff data of two reservoirs in China are analyzed as case studies. The results have been compared with the predictions made by ARIMA and Singular Spectrum Analysis-Linear Recurrent Formulae (SSA-LRF). It is shown that hybrid model has the best performance.

The Sources of Uncertainty in Forecasting Caspian Sea Level and Estimating the Inundation Risk of Coastal Areas

Stochastic variations in the climate and hydrological regime, both natural and anthropogenic, are the main cause of uncertainty in long-term hydrological forecasts and hence increase the estimated risk of economic activity in the coastal zone of internal seas. Some sources of uncertainty, which appear during the hydrological analysis, are considered with the purpose to assess this risk. Digital relief models were used to determine the morphological characteristics (as functions of the sea level) and assess their contribution to variations in the level regime. To take into account the sample uncertainty in the parameter estimates of stochastic models of the “impellent” processes, it is proposed to use the existing methodology of probabilistic-deterministic prediction of water level variations in a closed water body in combination with the Bayesian approach.

Long-range forecasts of River Po discharges based on predictable solar activity and a fuzzy neural network model / Prévisions à long terme des débits du Fleuve Pô basées sur l’activité solaire prévisible et sur un modèle de réseau de neurones flou

Is it possible to make seasonal and interannual forecasts of hydrological variables if one cannot predict next week’s rainfall? Contrary to common view, some scientists support the hypothesis that variations in mean global temperature and precipitation are controlled more by external forcing (solar variability and volcanic eruptions) than by increasing atmospheric concentration of greenhouse gases. Temperature and precipitation are connected with special phases of the 11-year sunspot cycle, which coincide with significant accumulation of energetic solar eruptions. Because of the possibility of identifying years with many solar eruptions, the attractive prospect emerges of the long-term hydrological forecasting based on cycles of solar activity. Starting from this assumption, an expert system was built based on a fuzzy neural network model for seasonal and interannual forecasting of the Po River discharge. It was found that indices of solar activity and of global circulation are sufficient to yield useful forecasts of hydrological variables.

HYDROLOGICAL ASPECTS OF THE WORLD WEATHER WATCH

ABSTRACT The World Weather Watch is a world weather system which is being implemented with a view to ensuring that modern scientific and technological advances are utilized to the maximum to improve our knowledge of atmospheric processes and that the benefits of the system are made available to all countries. The World Weather Watch, by its very nature, contains elements of direct interest and use to hydrology and water resources development. In particular, it will make available pictorial and numerical data that have been obtained by remote sensing techniques, covering large areas of the earth. The analysis and application of these data for hydrological and water resource purposes will permit greater accuracy, greater speed and better coordination of both short and long-term hydrological forecasts, in addition to the benefits that will accrue through their use for hydrological research purposes.