Streamflow Prediction Model Research Articles

Enhancing streamflow prediction in the Wujiang River basin: a two-stage decomposition approach with deep learning integration

ABSTRACT Global climate change and human activities have profoundly impacted the geological and hydrological processes in watersheds, increasing the challenges in streamflow prediction. In this study, we propose a streamflow prediction model based on deep learning and dual-mode decomposition specifically tailored for the Wujiang River basin, a significant tributary on the southern bank of the upper Yangtze River. This model effectively addresses the prediction error issue caused by high-frequency components through dual-mode decomposition of time-series data. The results demonstrate that employing the coupled complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN)–variational mode decomposition (VMD)–temporal convolutional network (TCN)–long short-term memory network (LSTM) model reduces prediction errors by at least 35% compared to single decomposition models. Furthermore, compared to individual TCN or LSTM models, the TCN–LSTM coupled model exhibits greater stability and higher prediction accuracy during training, with reductions in mean absolute error and root mean square error by 43.13 and 24.57%, respectively. This model holds promising prospects for application and can provide crucial insights for water resource management.

Prediction of streamflow in Chalakudy River Basin, Kerala, by integrating teleconnection patterns of large-scale atmospheric circulations

ABSTRACT In this study, the relationship between large-scale climatic drivers and streamflow of the Chalakudy River Basin, Kerala, was analysed using methods such as bivariate wavelet coherence (BWC), multiple wavelet coherence (MWC), and partial wavelet coherence (PWC) analysis. The four prominent global climate indices chosen are the Indian Ocean Dipole (IOD), North Atlantic Oscillation (NAO), El Niño Southern Oscillation (ENSO), and Pacific Decadal Oscillation (PDO), along with other local climate drivers. The BWC analysis showed that streamflow and rainfall had a very strong in-phase relationship, whereas the maximum temperature and average temperature showed an anti-phase relationship. In the case of global climate drivers, ENSO has a significant impact on the streamflow of the Chalakudy River Basin. The average wavelet coherence (AWC), which quantifies the teleconnections, confirms the observation with a high coherency of 0.75 between streamflow and rainfall and 0.51 for streamflow and ENSO. Streamflow prediction models were developed using random forest (RF) and artificial neural network (ANN) techniques by considering the influence of significant global and local climatic drivers. It was observed that the RF model performed slightly better than the ANN model, with R = 0.875, NSE = 0.766, RMSE = 23.468, and RSR = 0.524.

Application of SA-Conv1D-BiGRU model for streamflow prediction in southern Ethiopia

ABSTRACT Streamflow prediction offers crucial information for managing water resources, flood control, and hydropower generation. Yet, reliable streamflow prediction is challenging due to the complexity and nonlinearity of the rainfall-runoff relationship. This study investigated the comparative performance of the newly integrated self-attention-based deep learning (DL) model, SA-Conv1D-BiGRU with Conv1D-LSTM, and bidirectional long short-term memory (Bi-LSTM) models for streamflow prediction under different time-series conditions, and a range of variable input combinations based on flood events. All datasets passed quality control procedures, and the time lag for generating input series was established through Pearson correlation analysis. 80% of the data was used for training, whereas 20% was used to evaluate the model's performance. The performance of the models was evaluated using three metrics: mean absolute error (MAE), root mean square error (RMSE), and correlation coefficient (R2). The findings reveal the excellent potential of DL models for streamflow prediction, with the SA-Conv1D-BiGRU model outperforming other models under different time-series characteristics. Despite the complexity, the Conv1D-LSTM models did not outperform the Bi-LSTM model. In conclusion, the results are condensed into themes of model variability and time-series characteristics. Consequently, different architectures in DL models had a greater influence on streamflow prediction accuracy than input time lags and time-series features.

Streamflow prediction model for agriculture dominated tropical watershed using machine learning and hierarchical predictor selection algorithms

Study regionRana watershed, located in the mid-Mahanadi River basin in the state of Odisha, India. Study focusThis study attempted to develop a generalizable machine learning (ML)-based streamflow prediction model implementing prediction selection algorithms to the physiographic characteristics, and hydro-meteorological data collected for Rana Watershed. New hydrological insightsThe pertinent predictors identified were land use/ land cover (LULC), one and two-day lagged rainfall, one-day lagged PET, and one-day lagged streamflow and its categorized flow regime. The random forest algorithm, which outperformed the other five algorithms evaluated, was trained using identified predictors to develop a streamflow prediction model called “stRFlow”. The mean absolute error, root mean squared error, coefficient of determination, and Nash-Sutcliffe efficiency during training were 0.753 m3/s, 3.584 m3/s, 0.973, and 0.972 and testing were 2.829 m3/s, 10.503 m3/s, 0.855, and 0.851, respectively. The Kling-Gupta efficiency was found to be 0.96 and 0.92 during training and testing, respectively. There was an enhancement to model proficiency with the addition of LULC to temporal predictors. Moreover, the partial auto-correlation factor for the streamflow and examining the categorization of specific lagged flow regimes enhanced the predictive capacities of “stRFlow”. Results depict the potential of stRFlow and the framework in streamflow modeling in similar hydroclimatic regions with applicability for practical and reliable streamflow predictions globally.

Spatio-temporal deep learning model for accurate streamflow prediction with multi-source data fusion

Accurately predicting streamflow and early flood warning are important but also very challenging, due to the complexity and stochastic nature of the runoff process. By describing the process nature, this study proposes a spatio-temporal deep learning model, which integrates multiple sources of information including Hydrology-related data from the Global Land Data Assimilation System (GLDAS), hydro-meteorological and streamflow data, to better capture the complexity of the hydrological processes. The Maximum Information Coefficient (MIC) is utilized to reduce data dimensionality and assess the relationship between streamflow and other variables. The Completely Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN) and Variational Mode Decomposition (VMD) methods are combined to extract the most significant features in the predictors. The Time Convolutional Network (TCN) and Gated Recurrent Unit (GRU) are also combined for prediction, enhancing the model's robustness and balancing short-term and long-term forecasting effects. The integration of predictive sub-models with Random Forest (RF) improves overall performance. The framework is applied to 11 hydrological stations in the upper, middle, and lower reaches of the Jialing River mainstream basin in China. According to performance measures, our approach outperforms other baseline models. Flood, probabilistic, and lead time predictions of streamflow make the model more widely applicable and robust. In order to increase the interpretability, the SHAP tool is used in this study to analyze the contribution of each selected influencing variable to the long-term trend of streamflow.

Interpretable and explainable hybrid model for daily streamflow prediction based on multi-factor drivers.

Streamflow time series data typically exhibit nonlinear and nonstationary characteristics that complicate precise estimation. Recently, multifactorial machine learning (ML) models have been developed to enhance the performance of streamflow predictions. However, the lack of interpretability within these ML models raises concerns about their inner workings and reliability. This paper introduces an innovative hybrid architecture, the TCN-LSTM-Multihead-Attention model, which combines two layers of temporal convolutional networks (TCN) followed by one layer of long short-term memory (LSTM) units, integrated with a Multihead-Attention mechanism for predicting streamflow with streamflow causation-driven prediction samples (RCDP), employing local and global interpretability studies through Shapley values and partial dependency analysis. The find_peaks method was used to identify peak flow events in the test dataset, validating the model's generality and uncovering the physical causative patterns of streamflow. The results show that (1) compared to the LSTM model with the same hyperparameter settings, the proposed TCN-LSTM-Multihead-Attention hybrid model increased the R2 by 52.9%, 2.5%, 43.1%, and 10.7% respectively at four stations in the test set predictions using RCDP samples. Moreover, comparing the prediction results of the hybrid model under different samples in Hengshan station, the R2 for RCDP increased by 5.06% and 1.22% compared to streamflow autoregressive prediction samples (RAP) and meteorological-soil volumetric water content coupled autoregressive prediction samples (MCSAP) respectively. (2) Historical streamflow data from the preceding 3days predominantly influences predictions due to strong autocorrelation, with flow quantity (Q) typically emerging as the most significant feature alongside precipitation (P), surface soil moisture (SSM), and adjacent station flow data. (3) During periods of low and normal flow, historical data remains the most crucial factor; however, during flood periods, the roles of upstream inflow and precipitation become significantly more pronounced. This model facilitates the identification and quantification of various hydrodynamic impacts on flow predictions, including upstream flood propagation, precipitation, and soil moisture conditions. It also elucidates the model's nonlinear relationships and threshold responses, thereby enhancing the interpretability and reliability of streamflow predictions.

Investigation of climate change impacts on daily streamflow extremes in Eastern Black Sea Basin, Turkey

This study was undertaken to evaluate how future uncertain climate-related hydrological responses and low and highflow frequencies affect local hydrology up to the 2100, depending on different Global Circulation Models (GCM) and different Concentration Scenarios (CS). To this end, daily total precipitation and daily mean temperature data were used for RCP4.5 and RCP8.5 CSs by employing GCMs (GFDL-ESM2M, HadGEM2-ES and MPI-ESM-MR). In the study, the years 1971–2000 were chosen as the baseline period. Future streamflow predictions were modeled by machine learning. The models were calibrated with baseline period data and future streamflow values were predicted by using the best Streamflow Prediction Model during the 2055s (2041–2070) and 2085s (2071–2100) periods. Flow Duration Curves from the streamflow predictions were obtained and by using them, discharges corresponding to 95% probability of exceedance for lowflow and 5% probability of exceedance for highflow were calculated. According to the findings, it is predicted that there will be decrease of approximately 21% in lowflow discharges and approximately 30% in highflow discharges. This prediction has suggested water resources managers to implement mitigation measures to climate change in Eastern Black Sea Basin and most probably in Türkiye, especially in the context of continuity of aquatic life and sustainable water supply. It is expected that the paper will take a very important place in estimating and evaluating the expected changes in basin hydrology in the coming years, especially in the highflows and lowflows, which are the key concepts in hydroelectric power generation and in flood management.

Assessing Objective Functions in Streamflow Prediction Model Training Based on the Naïve Method

Reliable streamflow forecasting is a determining factor for water resource planning and flood control. To better understand the strengths and weaknesses of newly proposed methods in streamflow forecasting and facilitate comparisons of different research results, we test a simple, universal, and efficient benchmark method, namely, the naïve method, for short-term streamflow prediction. Using the naïve method, we assess the streamflow forecasting performance of the long short-term memory models trained with different objective functions, including mean squared error (MSE), root mean squared error (RMSE), Nash–Sutcliffe efficiency (NSE), Kling–Gupta efficiency (KGE), and mean absolute error (MAE). The experiments over 273 watersheds show that the naïve method attains good forecasting performance (NSE > 0.5) in 88%, 65%, and 52% of watersheds at lead times of 1 day, 2 days, and 3 days, respectively. Through benchmarking by the naïve method, we find that the LSTM models trained with squared-error-based objective functions, i.e., MSE, RMSE, NSE, and KGE, perform poorly in low flow forecasting. This is because they are more influenced by training samples with high flows than by those with low flows during the model training process. For comprehensive short-term streamflow modeling without special demand orientation, we recommend the application of MAE instead of a squared-error-based metric as the objective function. In addition, it is also feasible to perform logarithmic transformation on the streamflow data. This work underscores the critical importance of appropriately selecting the objective functions for model training/calibration, shedding light on how to effectively evaluate the performance of streamflow forecast models.

A Novel Stochastic Tree Model for Daily Streamflow Prediction Based on A Noise Suppression Hybridization Algorithm and Efficient Uncertainty Quantification

A Novel Stochastic Tree Model for Daily Streamflow Prediction Based on A Noise Suppression Hybridization Algorithm and Efficient Uncertainty Quantification

Alternate pathway for regional flood frequency analysis in data-sparse region

Accurately analyzing flood frequency is crucial for developing effective flood management strategies and designing flood protection infrastructure, but the complex and nonlinear nature of the hydrological system poses significant challenges. The use of advanced statistical techniques, hydrological modeling, and computational tools has facilitated the implementation of several regional flood frequency analysis (RFFA) methods, including machine learning (ML) approaches. However, such regional methods, being statistical and data-intensive with limited hydrologic inferences, can be challenging for regions with limited data. In this study, we evaluated an alternate approach or path for RFFA in two contrasting regions: a data-sparse region in India and a data-dense region in the USA. The alternate approach involves using deep learning (DL)-based model for streamflow prediction, followed by at-site flood frequency analysis to estimate flood quantiles. The proposed alternate approach was compared with the classical RFFA method, which utilized the Random Forest (RF) and eXtreme Gradient Boosting (XGBoost) algorithms to establish the intricate relationship between different flood quantiles and predictor variables related to physio-meteorological conditions. The findings showed that the alternate approach outperformed the classical RFFA method in the data-sparse region of India, reducing the mean absolute error (MAE) and root mean squared error (RMSE) by approximately 50 %, and achieving a higher coefficient of determination (R2 = 0.85 to 0.96). In the data-dense region, the classical and alternate approaches produced comparable results. However, the alternate approach has the benefit of flexibility and offers a complete time series of daily flow at the ungauged watersheds, enabling the estimation of other flow attributes or magnitudes for any return period without needing a separate classical RFFA model. The study demonstrates that the alternate approach can produce reliable flood quantile estimates in regions with limited data, offering a promising solution for managing floods in these areas.

A deep learning-based hybrid approach for multi-time-ahead streamflow prediction in an arid region of Northwest China

Abstract Accurate streamflow prediction is crucial for effective water resource management. However, reliable prediction remains a considerable challenge because of the highly complex, non-stationary, and non-linear processes that contribute to streamflow at various spatial and temporal scales. In this study, we utilized a convolutional neural network (CNN)–Transformer–long short-term memory (LSTM) (CTL) model for streamflow prediction, which replaced the embedding layer with a CNN layer to extract partial hidden features, and added an LSTM layer to extract correlations on a temporal scale. The CTL model incorporated Transformer's ability to extract global information, CNN's ability to extract hidden features, and LSTM's ability to capture temporal correlations. To validate its effectiveness, we applied it for streamflow prediction in the Shule River basin in northwest China across 1-, 3-, and 6-month horizons and compared its performance with Transformer, CNN, LSTM, CNN–Transformer, and Transformer–LSTM. The results demonstrated that CTL outperformed all other models in terms of predictive accuracy with Nash–Sutcliffe coefficient (NSE) values of 0.964, 0.912, and 0.856 for 1-, 3-, 6-month ahead prediction. The best results among the five comparative models were 0.908, 0.824, and 0.778, respectively. This indicated that CTL is an outstanding alternative technique for streamflow prediction where surface data are limited.

Post-Processing Ensemble Precipitation Forecasts and Their Applications in Summer Streamflow Prediction over a Mountain River Basin

Ensemble precipitation forecasts (EPFs) can help to extend lead times and provide reliable probabilistic forecasts, which have been widely applied for streamflow predictions by driving hydrological models. Nonetheless, inherent biases and under-dispersion in EPFs require post-processing for accurate application. It is imperative to explore the skillful lead time of post-processed EPFs for summer streamflow predictions, particularly in mountainous regions. In this study, four popular EPFs, i.e., the CMA, ECMWF, JMA, and NCEP, were post-processed by two state of art methods, i.e., the Bayesian model averaging (BMA) and generator-based post-processing (GPP) methods. These refined forecasts were subsequently integrated with the Xin’anjiang (XAJ) model for summer streamflow prediction. The performances of precipitation forecasts and streamflow predictions were comprehensively evaluated before and after post-processing. The results reveal that raw EPFs frequently deviate from ensemble mean forecasts, particularly underestimating torrential rain. There are also clear underestimations of uncertainty in their probabilistic forecasts. Among the four EPFs, the ECMWF outperforms its peers, delivering skillful precipitation forecasts for 1–7 lead days and streamflow predictions for 1–4 lead days. The effectiveness of post-processing methods varies, yet both GPP and BMA address the under-dispersion of EPFs effectively. The GPP method, recommended as the superior method, can effectively improve both deterministic and probabilistic forecasting accuracy. Moreover, the ECMWF post-processed by GPP extends the effective lead time to seven days and reduces the underestimation of peak flows. The findings of this study underscore the potential benefits of adeptly post-processed EPFs, providing a reference for streamflow prediction over mountain river basins.

A Probability Model for Short-Term Streamflow Prediction Based on Multi-Resolution Data

A Probability Model for Short-Term Streamflow Prediction Based on Multi-Resolution Data

A Hybrid Model Combining the Cama-Flood Model and Deep Learning Methods for Streamflow Prediction

A Hybrid Model Combining the Cama-Flood Model and Deep Learning Methods for Streamflow Prediction

Cascaded-ANFIS to simulate nonlinear rainfall–runoff relationship

Hydrologic models require atmospheric, dynamic and static models to simulate river flow from catchments. Thus the accuracy of hydrologic modelling highly depends on the data quality. Therefore, simulation is always challenging in data-scarcity environments. In addition, physical flow measurements are infeasible in the Spatiotemporal domain, and soft computing techniques are helpful in river flow simulation in data-scarcity environments. In this research paper, an efficient and accurate Cascaded-ANFIS-based model for rainfall–runoff was proposed and evaluated using five case studies in three countries: Japan, Vietnam, and Sri Lanka. The investigation focused on predicting streamflow by the influence of past data, with each river’s dataset examined to determine the best configuration of past rainfalls affecting streamflow volume. The proposed algorithm was compared against six state-of-the-art regression algorithms. The results showed that it outperformed the other algorithms in every case study except the Kalu River dataset, with zero bias calculated. The developed R-R model can be considered a generic model for streamflow prediction in data-scarcity environments, with excellent acceptability of simulated river flows against measured river flows observed across different geographic and climatic regions.

Using Adaptive Chaotic Grey Wolf Optimization for the daily streamflow prediction

Due to the importance of water resources management that requires the optimal model for streamflow prediction, the study of water flow prediction is of great importance. Hence, the present paper aims to introduce an innovative model that can predict streamflow with the highest accuracy. Accordingly, a multilayer perceptron neural network whose optimum neuron number is specified using the evaluation metrics is considered for simulation. The dataset used for this purpose consists of 80% experimental and 20% numerical data. Accordingly, the MLP network used in this research is optimized using various optimizers, namely Particle Swarm Optimization, Genetic Algorithm, Grey Wolf Optimization, Chaotic Grey Wolf Optimization, Advanced Grey Wolf Optimization, and Adaptive Chaotic Grey Wolf Optimization. Considering the statistical results, Adaptive Chaotic Grey Wolf Optimization is introduced as the leading optimizer. The obtained results highlight that Adaptive Chaotic Grey Wolf Optimization used for optimizing the training process of the MLP network with 15 neurons can predict the daily streamflow with the best standard error, mean square error, mean absolute percent error, mean absolute error, and normalized mean square error of about 5.3126, 2.3049, 0.6684, 1.1038, and 0.4483, respectively.

Advanced Machine Learning Techniques to Improve Hydrological Prediction: A Comparative Analysis of Streamflow Prediction Models

The management of water resources depends heavily on hydrological prediction, and advances in machine learning (ML) present prospects for improving predictive modelling capabilities. This study investigates the use of a variety of widely used machine learning algorithms, such as CatBoost, ElasticNet, k-Nearest Neighbors (KNN), Lasso, Light Gradient Boosting Machine Regressor (LGBM), Linear Regression (LR), Multilayer Perceptron (MLP), Random Forest (RF), Ridge, Stochastic Gradient Descent (SGD), and the Extreme Gradient Boosting Regression Model (XGBoost), to predict the river inflow of the Garudeshwar watershed, a key element in planning for flood control and water supply. The substantial engineering feature used in the study, which incorporates temporal lag and contextual data based on Indian seasons, leads it distinctiveness. The study concludes that the CatBoost method demonstrated remarkable performance across various metrics, including Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and R-squared (R2) values, for both training and testing datasets. This was accomplished by an in-depth investigation and model comparison. In contrast to CatBoost, XGBoost and LGBM demonstrated a higher percentage of data points with prediction errors exceeding 35% for moderate inflow numbers above 10,000. CatBoost established itself as a reliable method for hydrological time-series modelling, easily managing both categorical and continuous variables, and thereby greatly enhancing prediction accuracy. The results of this study highlight the value and promise of widely used machine learning algorithms in hydrology and offer valuable insights for academics and industry professionals.

Streamflow prediction using machine learning models in selected rivers of Southern India

ABSTRACT The need for adequate data on the spatial and temporal variability of freshwater resources is a significant challenge to the water managers of the world in water resource planning and management. The problems will be acute in the coming years because of the increase in frequency and intensity of hydrologic extremes due to climate change. Therefore, streamflow prediction has become an important area of research because of its importance in flood mitigation, reservoir operation, and water resource management. In this paper, we have tested four Machine Learning models (ML models): Support Vector Machines (SVM), Random Forest (RF), Long Short-Term Memory (LSTM), and Multivariate Adaptive Regression Splines (MARS) for streamflow prediction at daily and monthly time scales in three rivers draining in the different climatic and geological settings. The SVM, RF, LSTM, and MARS models have been trained and tested in the Suvarna, Aghanashini, and Kunderu River Basins in peninsular India. Model intercomparison was made to identify the best suitable model for streamflow prediction. The RF outperforms other models for daily streamflow, and MARS outperforms other models for monthly streamflow prediction in the Suvarna river with Nash-Sutcliffe efficiency (NSE) values of 0.676 and 0.924, respectively. SVM (NSE = 0.741) and RF (NSE = 0.826) are found to be the best models for daily and monthly streamflow prediction in the Aghanashini river. MARS outperformed other models in the case of high, severe, and extreme flow simulation with NSE values of 0.481, 0.374, and 0.455, respectively, in the Aghanashini river. Other hydrological variables (groundwater level data, antecedent soil moisture, potential evapotranspiration data) and a better spatial resolution of rainfall data can be used to develop more accurate machine-learning models for streamflow predictions.

Predicting daily streamflow with a novel multi-regime switching ARIMA-MS-GARCH model

Study regionWeihe River Basin of China Study focusIn recent decades, changing environments destroyed the natural structure of streamflow, making accurate streamflow prediction challenging. This study develops a multi-regime Markov-switching Generalized Autoregressive Conditional Heteroskedasticity (MS-GARCH) model to predict daily streamflow time series with structural breaks (SB), named ARIMA (Autoregressive Integrated Moving Average)-MS-GARCH model. Consequently, the multi-regime ARIMA-MS-GARCH model is compared with other classical single-regime ARIMA-GARCH models to evaluate whether and to what extent it improves streamflow prediction accuracy. New hydrology insights for the regionThere exist structural breaks in the daily streamflow time series, and the number of breakpoints at each station varies. The GARCH model ignores the description of volatility aggregation in the daily streamflow time series (α<β). The ARIMA-MS-GARCH model (R2 and NSE in the range of 0.682–0.984 and 0.582–0.935, respectively) outperforms the ARIMA-GARCH model (R2 and NSE in the range of 0.558–0.935 and 0.077–0.721, respectively) for daily streamflow prediction, and the ARIMA-GARCH model seriously overestimates the peak values. The MS-GARCH model based on the Student's t distribution is more suitable for daily streamflow prediction than that based on the normal distribution, in which MAE and RE reduce by 23.90%− 52.28% and 23.46%− 54.67%, respectively, and R2 and NSE increase by 5.09% − 15.54% and 1.63% − 60.65%.

A novel workflow for streamflow prediction in the presence of missing gauge observations

AbstractStreamflow predictions are vital for detecting flood and drought events. Such predictions are even more critical to Sub-Saharan African regions that are vulnerable to the increasing frequency and intensity of such events. These regions are sparsely gaged, with few available gaging stations that are often plagued with missing data due to various causes, such as harsh environmental conditions and constrained operational resources. This work presents a novel workflow for predicting streamflow in the presence of missing gage observations. We leverage bias correction of the Group on Earth Observations Global Water and Sustainability Initiative ECMWF streamflow service (GESS) forecasts for missing data imputation and predict future streamflow using the state-of-the-art temporal fusion transformers (TFTs) at 10 river gaging stations in the Benin Republic. We show by simulating missingness in a testing period that GESS forecasts have a significant bias that results in poor imputation performance over the 10 Beninese stations. Our findings suggest that overall bias correction by Elastic Net and Gaussian Process regression achieves superior performance relative to traditional imputation by established methods. We also show that the TFT yields high predictive skill and further provides explanations for predictions through the weights of its attention mechanism. The findings of this work provide a basis for integrating Global streamflow prediction model data and the state-of-the-art machine learning models into operational early-warning decision-making systems in resource-constrained countries vulnerable to drought and flooding due to extreme weather events.