Weekly streamflow forecasting of Rhine river based on machine learning approaches

Abstract The Rhine River is a vital waterway in Europe, crucial for navigation, hydropower generation, and ecosystem health. Thus, accurately forecasting its streamflow is essential for effective water resource management. This study explored the utilization of several Machine Learning (ML) techniques including Multi-layer Perceptron (MLP), Support Vector Regression (SVR), K-Nearest Neighbor (KNN), and eXtreme Gradient Boosting (XGBoost), for forecasting weekly streamflow for the Rhine River. Meteorological data (e.g., precipitation, temperature, vapor pressure deficit) collected from meteorological stations situated on the main river course (i.e., Mannheim and Worms) spanning from 2013 to 2023 were used as predictors. Two scenarios were considered for predicting weekly streamflow according to the results of the best input combination. According to the results streamflow is most significantly predicted by precipitation, vapor pressure, and relative humidity, while average and maximum temperatures play a smaller role. Various quantitative and visually-oriented evaluation metrics were employed to validate and compare the performance of the proposed models. It found that the XGBoost outperformed than other algorithms in prediction of weekly streamflow in Rhine River. Overall, weekly streamflow forecasting for the Rhine River is crucial for effective water resource management, navigation, hydropower generation, flood control, and ecosystem health. By providing timely insights into flow variations and identifying the optimal predictors, weekly forecasts empower stakeholders to make informed decisions and ensure the Rhine’s continued sustainability.