Streamflow Forecasting Research Articles

Modelling Stage-discharge Relationships for Hydrological Stations in the Da River Basin using First-kind Chebyshev Polynomial Approximation

Observed discharge time series is essential data for hydrological analysis, streamflow forecasting, as well as water resources planning and management. Estimation of continuous discharge time series is generally based on the curves (or empirical models) that can properly simulate the relationships between water levels (stages) and concurrent water discharges observed at a particular hydrological station. For stations with complex hydraulic conditions, in addition to gauged stages at the discharge gauging section, another variable such as stage measured at an auxiliary station (to calculate the slope) or rate of change in stage is added to the estimation process. This paper presents the results of applying the first-kind Chebyshev polynomial approximation method in constructing rating curves for three hydrological stations in the Da river basin, each having different basin characteristics and various influencing factors. The analysis results using five evaluation indices (MAE, σ, Pbias, KGE, and MAPE) show that the Chebyshev polynomials are highly effective in modelling both simple and complex stage-discharge relationships. The Chebyshev polynomial approximation method is suggested to be used for establishing stage-discharge relationships in hydrometry.

Глубокие нейронные сети архитектуры трансформер в задачах гидрологических прогнозов

The theoretical analysis of modern neural network models used for processing time series is carried out. At the same time, special attention is paid to the architecture of building deep machine learning algorithms. The advantages of the neural network model Temporal Fusion Transformer (TFT), which is selected as the base for modeling the process of spring flood formation, are shown. The possibilities of using the TFT model for a long-term forecast of maximum water levels with a lead time of 60 and 90 days for several points of the Iset river (the basin of the Tobol river system) are numerically analyzed in detail. The daily time series for 27 years (1991-2017) for eight hydrometeorological characteristics were used as the initial information for training the model (dependent (trainable) sample). The results of forecasts for an independent sample (2018-2022), as well as operational forecast data for 2023 are presented. Several directions of development of neural network modeling for long-term and short-term forecasts of streamflow are identified. Keywords: hydrological long-term forecasts, flooding, neural networks, discharge, water level, water regime, artificial intelligence, deep machine learning, Temporal Fusion Transformer

Post-processing quantitative precipitation forecasts using the seasonally coherent calibration model

ABSTRACT Skilful precipitation ensemble forecasts are necessary to produce trustworthy hydrologic predictions. Raw quantitative precipitation forecasts (QPFs) from the numerical weather prediction (NWP) models are known to be error-prone. In this study, sub-basin averaged deterministic QPFs with five-day lead times from the European Centre for Medium-Range Weather Forecasts (ECMWF) are post-processed through the Seasonally Coherent Calibration (SCC) model for the Narmada and Godavari River basins of India. The SCC model incorporates seasonal climatology from long observations into forecasts and produces calibrated forecasts based on a joint probability model. The SCC model results are compared with the post-processed forecasts from the state-of-the-art Quantile Mapping (QM) method. The results suggest that the probabilistic ensemble forecasts generated from the SCC model have improved skill throughout five-day lead times. Further, the application of SCC-calibrated precipitation forecasts is demonstrated using the Soil & Water Assessment Tool (SWAT) to generate streamflow forecasts.

Investigating technology opportunities toward improved Colorado water monitoring: Insights from key informant interviews and stakeholder surveys

Prolonged drought conditions in the western United States have prompted an interest in improved water resource management. In the past two decades, Colorado has experienced decreasing precipitation and increasing temperatures, causing persistent meteorological and agricultural drought. In times of limited water resources, more precise water accounting and water conservation is often required. We investigated technological and monitoring gaps in Colorado water resource management using a sequential explanatory mix-method approach, including key informant interviews and a survey study. Key informant interviews gave insight into water monitoring concerns and informed a state-wide survey on monitoring gaps and water resource challenges. Qualitative and quantitative analysis of interview transcripts and survey results found that the most critical monitoring gaps for water managers in Colorado are: (1) streamflow forecasting and improved understanding of Colorado snowpack, (2) expanding groundwater and soil moisture monitoring, (3) wildfire impacts on watershed health, (4) improved accuracy and reliability of data sources, and (5) addressing these gaps while maintaining an emphasis on community collaboration. Challenges for Colorado managers varied by stakeholder basin and sector but included changes in hydrologic systems due to the effects of extended drought, climate change, and related anthropogenic impacts and negative impacts of intensifying wildfire seasons on water quality and watershed health.

Application of multilayer perceptron (MLP) method for streamflow forecasting (case study: Upper Citarum River, Indonesia)

Flood forecasting is a critical component of flood early warning. The discharge that occurs is one of the parameters that can be used as a reference for predicting flooding. Various discharge forecasting models based on physically based models or data-driven models have been developed. One of the flood forecasting methods that can be considered for forecasting discharge on watersheds with limited physical data is the Artificial Neural Network (ANN). Furthermore, the ANN method allows the analysis process to be completed in less time and with fewer resources. One of the ANN models employed in this work is the multilayer perceptron (MLP). The MLP model was developed in this study to predict streamflow at the Citarum river, particularly at the Dayeuhkolot hydrological station at 2, 4, 6, 8, 10, 12, and 24 hours ahead. Two data input scenarios were used in the modeling scene. First, input data in the form of station rainfall data and discharge data. The second is regional rainfall and discharge data. Before predicting the discharge in the coming hours, the hyperparameters model is optimized using the GridSearchCV method. The model’s performance is assessed using the RMSE, NSE, and R2 values. The MLP method produced satisfactory results for both scenarios when predicting discharge in less than 4 hours with the NSE and R2 value higher than 0.9. Scenario 2 input data produces a slightly better prediction model than scenario 1. Based on NSE and R2 values, discharge prediction with a prediction time of more than 6 hours produces less accurate results.

Effect of the quality of streamflow forecasts on the operation of cascade hydropower stations using stochastic optimization models

Effect of the quality of streamflow forecasts on the operation of cascade hydropower stations using stochastic optimization models

Seasonal catchment memory of high mountain rivers in the Tibetan Plateau

Rivers originating in the Tibetan Plateau are crucial to the population in Asia. However, research about quantifying seasonal catchment memory of these rivers is still limited. Here, we propose a model able to accurately estimate terrestrial water storage change (TWSC), and characterize catchment memory processes and durations using the memory curve and the influence/domination time, respectively. By investigating eight representative basins of the region, we find that the seasonal catchment memory in precipitation-dominated basins is mainly controlled by precipitation, and that in non-precipitation-dominated basins is strongly influenced by temperature. We further uncover that in precipitation-dominated basins, longer influence time corresponds to longer domination time, with the influence/domination time of approximately six/four months during monsoon season. In addition, the long-term catchment memory is observed in non-precipitation-dominated basins. Quantifying catchment memory can identify efficient lead times for seasonal streamflow forecasts and water resource management.

Long-term probabilistic streamflow forecast model with “inputs–structure–parameters” hierarchical optimization framework

Accurate, reliable, and stable streamflow forecasts are essential for risk assessment and decision making in water resources management. Owing to the limited value of deterministic forecasting, probabilistic forecasting incorporating probability and confidence intervals is achieved by optimizing the predictors and forecast equation while identifying forecast error features. The combination of deterministic forecasts and error identification can facilitate water resources management. This study improved the traditional forecast model’s inputs selection, structure designation, and error parameters calibration to propose a long-term probabilistic streamflow forecast model with hierarchical optimization of the predictors, forecast equation, and errors characteristics. The model develops information entropy theory to screen the driving predictor set, the long short-term memory (LSTM) model to conduct deterministic forecasts, and the generalized autoregressive conditional heteroskedasticity (GARCH) model to identify time-varying errors. The proposed model, used in some case studies to forecast the monthly streamflow of two lakes, Hongze and Luoma lakes in China, was evaluated using the multidimensional index method considering “accuracy–reliability–stability” performance. The results revealed the following: (1) Predictor screening based on information entropy investigates the statistical characteristics of predictors, thereby improving the reliability of streamflow forecasts. (2) The LSTM model exploits the response between the driving predictors and streamflow, whereas GARCH model identifies the time-varying features of forecast errors effectively, which reduces the probability of forecast failure and increases the accuracy and stability of probabilistic forecasts. (3) Under current meteorological observation conditions and forecasting capability, the proposed forecasts can extend the maximum forecast lead time from 1 month to 3 months. (4) The proposed model improves the accuracy (root mean square error: 6.7%–34.8%), reliability (Brier score: 15.3%–27.9%), and stability (mistaken distance: 36.4–52.6%) in comparison with the benchmark of the two case studies, indicating that “inputs–structure–parameters” hierarchical optimization provides effective forecast information for water resources management.

Wildfire and climate change amplify knowledge gaps linking mountain source-water systems and agricultural water supply in the western United States

Agricultural production in the western United States relies on water supplies from mountain source-water systems that are sensitive to impacts from wildfire and a changing climate. The resultant challenges to water supply forecasting directly impact agricultural producers and irrigation managers who rely on snowmelt and streamflow forecasts for crop selection and irrigation scheduling. To date, much research has focused on source-water system processes and agricultural production separately, but in this short communication we highlight a substantial need for new research connecting these disparate systems to improve forecasting accuracy. We identify key knowledge and data gaps regarding the functioning of source watersheds and their contributions to agricultural water resources with associated uncertainties in the context of wildfire and changing climate. In doing so, we encourage researchers, resource managers, and agricultural producers to consider the interdependency of water supply source and sink relationships through improved observations, monitoring, and modeling to ensure sustainable food production in the western US.

Performance evaluation of artificial neural network model in hybrids with various preprocessors for river streamflow forecasting

Abstract Accurate forecasting of hydrological processes and sustainable management of water resources is inevitable, especially for flood control and water resource shortage crisis in low-water areas with an arid and semi-arid climate, which is a limitation for residents and various structures. The present study uses different data preprocessing techniques to deal with complex data and extract hidden features from the stream time series. In the next step, the decomposed time series were used, as input data, to the artificial neural network (ANN) model for streamflow modeling and forecasting. The preprocessors employed included discrete wavelet transform (DWT), empirical mode decomposition (EMD), complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN), successive variational mode decomposition (SVMD), and multi-filter of the smoothing (MFS). These preprocessors were used in hybrid with the ANN model to forecast the daily streamflow. In general, the results showed that the optimal performance of hybrid models has two basic steps. The first step is choosing a suitable approach to utilizing the input data to the model. The second step is to use the appropriate preprocessor. Overall, the results show that the MFS-ANN model in short-term forecasting and the SVMD-ANN model in long-term forecasting performed better than other hybrid models.

Streamflow prediction using a hybrid methodology based on variational mode decomposition (VMD) and machine learning approaches

The optimal management of water resources depends on accurate and reliable streamflow prediction. Therefore, researchers have become interested in the development of hybrid approaches in recent years to enhance the performance of modeling techniques for predicting hydrological variables. In this study, hybrid models based on variational mode decomposition (VMD) and machine learning models such as random forest (RF) and K-star algorithm (KS) were developed to improve the accuracy of streamflow forecasting. The monthly data obtained between 1956 and 2017 at the Iranian Bibijan Abad station on the Zohreh River were used for this purpose. The streamflow data were initially decomposed into intrinsic modes functions (IMFs) using the VMD approach up to level eight to develop the hybrid models. The following step models the IMFs obtained by the VMD approach using the RF and KS methods. The ensemble forecasting result is then accomplished by adding the IMFs’ forecasting outputs. Other hybrid models, such as EDM-RF, EMD-KS, CEEMD-RF, and CEEMD-KS, were also developed in this research in order to assess the performance of VMD-RF and VMD-KS hybrid models. The findings demonstrated that data preprocessing enhanced standalone models’ performance, and those hybrid models developed based on VMD performed best in terms of increasing the accuracy of monthly streamflow predictions. The VMD-RF model is proposed as a superior method based on root mean square error (RMSE = 13.79), mean absolute error (MAE = 8.35), and Kling–Gupta (KGE = 0.89) indices.

Comparative evaluation of LSTM, CNN, and ConvLSTM for hourly short-term streamflow forecasting using deep learning approaches

This study investigates the effectiveness of three deep learning methods, Long Short-Term Memory (LSTM), Convolutional Neural Network (CNN), and Convolutional Long Short-Term Memory (ConvLSTM), in short-term streamflow forecasting in the Kelantan and Muda River basins in Malaysia. The ability to forecast streamflow in the short-term is crucial for water resource management to prevent disasters such as floods. Hourly streamflow and rainfall data from the two river basins were used to evaluate the accuracy of the methods at one-hour, three-hour, and six-hour prediction intervals. The results showed that all three deep learning methods performed with high accuracy in predicting streamflow. The study also found that LSTM performed better in small basin with well spatial distributed rainfall stations, while CNN and ConvLSTM were more effective in moderate to high streamflow, and large river basin. Moreover, early prediction intervals had better accuracy compared to later ones. The findings of this study demonstrate the potential of using deep learning methods for short-term streamflow forecasting in water resource management. The study highlights the importance of accurate short-term streamflow forecasting and emphasizes the potential of using deep learning algorithms to achieve this goal.

A comparative study of data-driven models for runoff, sediment, and nitrate forecasting

Effective prediction of qualitative and quantitative indicators for runoff is quite essential in water resources planning and management. However, although several data-driven and model-driven forecasting approaches have been employed in the literature for streamflow forecasting, to our knowledge, the literature lacks a comprehensive comparison of well-known data-driven and model-driven forecasting techniques for runoff evaluation in terms of quality and quantity. This study filled this knowledge gap by comparing the accuracy of runoff, sediment, and nitrate forecasting using four robust data-driven techniques: artificial neural network (ANN), long short-term memory (LSTM), wavelet artificial neural network (WANN), and wavelet long short-term memory (WLSTM) models. These comparisons were performed in two main tiers: (1) Comparing the machine learning algorithms' results with the model-driven approach; In order to simulate the runoff, sediment, and nitrate loads, the Soil and Water Assessment Tool (SWAT) model was employed, and (2) Comparing the machine learning algorithms with each other; The wavelet function was utilized in the ANN and LSTM algorithms. These comparisons were assessed based on the substantial statistical indices of coefficient of determination (R-Squared), Nash-Sutcliff efficiency coefficient (NSE), mean absolute error (MAE), and root mean square error (RMSE). Finally, to prove the applicability and efficiency of the proposed novel framework, it was successfully applied to Eagle Creek Watershed (ECW), Indiana, U.S. Results demonstrated that the data-driven algorithms significantly outperformed the model-driven models for both the calibration/training and validation/testing phases. Furthermore, it was found that the coupled ANN and LSTM models with wavelet function led to more accurate results than those without this function.

The Increasing Role of Seasonal Rainfall in Western U.S. Summer Streamflow

AbstractSummer streamflow variations strongly affect water supply reliability and ecological functioning of western U.S. (WUS) streams. Traditional snow‐based forecasts of summer streamflow are becoming less accurate with warming‐induced reductions in winter snow accumulation. This reflects a rising importance of competing runoff‐generating processes in controlling summer streamflow variations, primarily an increasing role of rainfall in contrast to snowmelt. Here, based on a snowmelt‐rainfall tracking algorithm applied to two hydrological models, we show that cool‐season rainfall provides an important volumetric contribution to summer streamflow for many WUS streams in the current climate, and this contribution will increase under climate warming, especially in years with warm snow droughts and abnormally dry summers. We also show that seasonal rainfall (warm‐/cool‐seasons) dominates the variability of summer streamflow across ∼70% area of WUS. We show that an increasing warm‐season rainfall contribution to summer streamflow (largely replacing snowmelt) results in reduced summer streamflow predictability.

Hydropower System Operation and the Quality of Short-Term Hydrologic Ensemble Forecasts

Improving the operational effectiveness of hydropower systems is becoming more relevant considering the shift to renewable energy sources and the rising social, financial, and environmental costs associated with the construction of new hydropower facilities. This challenge is interdisciplinary as it involves technical, managerial, and institutional aspects. This paper focuses on a technical aspect, more specifically on the relationship between the quality of short-term ensemble streamflow forecasts and the energy produced by a hydropower system. A secondary objective is to measure the contribution of both the meteorological and structural uncertainties on the energy output. To achieve this, a numerical experiment comprising multiple sets of hydrologic ensemble forecasts of different quality and a suite of reservoir optimization models is developed for a case study in Canada (the hydropower system of the Gatineau River basin). These ensemble forecasts are processed by the short-term reservoir operation model in rolling-horizon mode over a planning period of 6 years. Each day, the short-term optimization model seeks to maximize the energy output over the 14-day forecast lead time considering the expected future value of the system derived from a midterm optimization model. The relationship between hydropower generation and common statistical scores characterizing the ensemble forecasts indicates that although there is a link between the quality of the forecasts and the energy production, it is not a one-to-one causal relationship. Our results also show that the diversity of hydrological models is beneficial to the production of energy, indicating that the diversity of model structures compensates the deficiencies of individuals models and adds value to the forecast.

Short-term forecasts of streamflow in the UK based on a novel hybrid artificial intelligence algorithm

In recent years, the growing impact of climate change on surface water bodies has made the analysis and forecasting of streamflow rates essential for proper planning and management of water resources. This study proposes a novel ensemble (or hybrid) model, based on the combination of a Deep Learning algorithm, the Nonlinear AutoRegressive network with eXogenous inputs, and two Machine Learning algorithms, Multilayer Perceptron and Random Forest, for the short-term streamflow forecasting, considering precipitation as the only exogenous input and a forecast horizon up to 7 days. A large regional study was performed, considering 18 watercourses throughout the United Kingdom, characterized by different catchment areas and flow regimes. In particular, the predictions obtained with the ensemble Machine Learning-Deep Learning model were compared with the ones achieved with simpler models based on an ensemble of both Machine Learning algorithms and on the only Deep Learning algorithm. The hybrid Machine Learning-Deep Learning model outperformed the simpler models, with values of R2 above 0.9 for several watercourses, with the greatest discrepancies for small basins, where high and non-uniform rainfall throughout the year makes the streamflow rate forecasting a challenging task. Furthermore, the hybrid Machine Learning-Deep Learning model has been shown to be less affected by reductions in performance as the forecasting horizon increases compared to the simpler models, leading to reliable predictions even for 7-day forecasts.

Advancing Medium-Range Streamflow Forecasting for Large Hydropower Reservoirs in Brazil by Means of Continental-Scale Hydrological Modeling

Streamflow forecasts from continental to global scale hydrological models have gained attention, but their performance against operational forecasts at local to regional scales must be evaluated. This study assesses the skill of medium-range, weekly streamflow forecasts for 147 large Brazilian hydropower plants (HPPs) and compares their performance with forecasts issued operationally by the National Electric System Operator (ONS). A continental-scale hydrological model was forced with ECMWF medium-range forecasts, and outputs were corrected using quantile mapping (QM) and autoregressive model approaches. By using both corrections, the percentage of HPPs with skillful forecasts against climatology and persistence for 1–7 days ahead increased substantially for low to moderate (9% to 56%) and high (72% to 94%) flows, while using only the QM correction allowed positive skill mainly for low to moderate flows and for 8–15 days ahead (29% to 64%). Compared with the ONS, the corrected continental-scale forecasts issued for the first week exhibited equal or better performance in 60% of the HPPs, especially for the North and Southeast subsystems, the DJF and MAM months, and for HPPs with less installed capacity. The findings suggest that using simple corrections on streamflow forecasts issued by continental-scale models can result in competitive forecasts even for regional-scale applications.

Seasonal Streamflow Forecast in the Tocantins River Basin, Brazil: An Evaluation of ECMWF-SEAS5 with Multiple Conceptual Hydrological Models

The assessment of seasonal streamflow forecasting is essential for appropriate water resource management. A suitable seasonal forecasting system requires the evaluation of both numerical weather prediction (NWP) and hydrological models to represent the atmospheric and hydrological processes and conditions in a specific region. In this paper, we evaluated the ECMWF-SEAS5 precipitation product with four hydrological models to represent seasonal streamflow forecasts performed at hydropower plants in the Legal Amazon region. The adopted models included GR4J, HYMOD, HBV, and SMAP, which were calibrated on a daily scale for the period from 2014 to 2019 and validated for the period from 2005 to 2013. The seasonal streamflow forecasts were obtained for the period from 2017 to 2019 by considering a daily scale streamflow simulation comprising an ensemble with 51 members of forecasts, starting on the first day of every month up to 7 months ahead. For each forecast, the corresponding monthly streamflow time series was estimated. A post-processing procedure based on the adjustment of an autoregressive model for the residuals was applied to correct the bias of seasonal streamflow forecasts. Hence, for the calibration and validation period, the results show that the HBV model provides better results to represent the hydrological conditions at each hydropower plant, presenting NSE and NSElog values greater than 0.8 and 0.9, respectively, during the calibration stage. However, the SMAP model achieves a better performance with NSE values of up to 0.5 for the raw forecasts. In addition, the bias correction displayed a significant improvement in the forecasts for all hydrological models, specifically for the representation of streamflow during dry periods, significantly reducing the variability of the residuals.

Streamflow forecasting for the Hunza river basin using ANN, RNN, and ANFIS models

Abstract Streamflow forecasting is essential for planning, designing, and managing watershed systems. This research study investigates the use of artificial neural networks (ANN), recurrent neural networks (RNN), and adaptive neuro-fuzzy inference systems (ANFIS) for monthly streamflow forecasting in the Hunza River Basin of Pakistan. Different models were developed using precipitation, temperature, and discharge data. Two statistical performance indicators, i.e., root mean square error (RMSE) and coefficient of determination (R2), were used to assess the performance of machine learning techniques. Based on these performance indicators, the ANN model predicts monthly streamflow more accurately than the RNN and ANFIS models. To assess the performance of the ANN model, three architectures were used, namely 2-1-1, 2-2-1, and 2-3-1. The ANN architecture with a 2-3-1 configuration had higher R2 values of 0.9522 and 0.96998 for the training and testing phases, respectively. For each RNN architecture, three transfer functions were used, namely Tan-sig, Log-sig, and Purelin. The architecture with a 2-1-1 configuration based on tan-sig transfer function performed well in terms of R2 values, which were 0.7838 and 0.8439 for the training and testing phases, respectively. For the ANFIS model, the R2 values were 0.7023 and 0.7538 for both the training and testing phases, respectively. Overall, the findings suggest that the ANN model with a 2-3-1 architecture is the most effective for predicting monthly streamflow in the Hunza River Basin. This research can be helpful for planning, designing, and managing watershed systems, particularly in regions where streamflow forecasting is crucial for effective water resource management.

Medium Term Streamflow Prediction Based on Bayesian Model Averaging Using Multiple Machine Learning Models

Medium-term hydrological streamflow forecasting can guide water dispatching departments to arrange the discharge and output plan of hydropower stations in advance, which is of great significance for improving the utilization of hydropower energy and has been a research hotspot in the field of hydrology. However, the distribution of water resources is uneven in time and space. It is important to predict streamflow in advance for the rational use of water resources. In this study, a Bayesian model average integrated prediction method is proposed, which combines artificial intelligence algorithms, including long-and short-term memory neural network (LSTM), gate recurrent unit neural network (GRU), recurrent neural network (RNN), back propagation (BP) neural network, multiple linear regression (MLR), random forest regression (RFR), AdaBoost regression (ABR) and support vector regression (SVR). In particular, the simulated annealing (SA) algorithm is used to optimize the hyperparameters of the model. The practical application of the proposed model in the ten-day scale inflow prediction of the Three Gorges Reservoir shows that the proposed model has good prediction performance; the Nash–Sutcliffe efficiency NSE is 0.876, and the correlation coefficient r is 0.936, which proves the accuracy of the model.