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

Reliable streamflow prediction is important for rational water resource planning. However, the strong nonlinearity and uncertainty of streamflow changes make accurate prediction challenging. Moreover, conventional streamflow prediction uses single-resolution data and provides deterministic prediction without uncertainty estimation, which leads to one-sidedness in data information extraction and risks in water resource decision-making. To improve streamflow prediction, this study proposes a probability prediction model integrating multi-resolution data for short-term streamflow prediction. In the proposed model, singular spectrum analysis (SSA) is utilized to process multi-resolution streamflow data to remove hidden noise. Then, support vector regression (SVR) is used for modelling, and grid search (GS) and cross-validation (CV) methods are employed to determine the optimal parameters of SVR. Finally, Gaussian process regression (GPR) is used for nonlinear fusion and probabilistic prediction. To verify the effectiveness of the proposed model, streamflow data from the Pingchuan bridge and Gaoya station with two different resolutions are collected, and several relevant models and indices are used for comparative analysis and comprehensive evaluation. The experimental results show that the proposed model significantly outperforms the relevant compared models, indicating that the proposed model effectively reduces the influence of interference signals on the modelling and fully utilizes the feature information provided by different resolution data to improve streamflow prediction. The results also confirm the superiority of integrating multi-resolution data over using single-resolution data in improving streamflow prediction. Moreover, the proposed model provides reliable uncertainty estimation in addition to providing accurate point prediction, which is helpful for water resource scheduling and decision-making. Therefore, the proposed model is recommended as a reliable method for streamflow prediction.