The role of matching pursuit algorithm and multi-scale daily rainfall data obtained from decomposition in runoff prediction

Abstract

Study Region: Huai River Basin, China Study Focus: Runoff data preprocessing and strongly correlated influencing factors are vital to predicting runoff by machine learning methods due to the impact of human activities and changing climate. To address the issue of noise interference in the runoff components post-decomposition. This study introduces a novel denoising technique that integrates the matching pursuit algorithm with permutation entropy. Furthermore, to enhance the correlation of the input factors, multi-scale daily rainfall data obtained through decomposition is utilized as input to the model for the first time. The aim of preprocessing the original runoff is to diminish noise interference and reduce the complexity of the subsequences. This method effectively increases the prediction accuracy. New hydrological insights for the region: The matching pursuit algorithm considerably decreases the noise interference and the complexity of the runoff. The effectiveness of the proposed method is shown through comparison with typical denoising methods. Meanwhile, multi-scale daily rainfall data acquired through rainfall data decomposition have stronger correlation with runoff components. Through data preprocessing methods and strongly correlated factors, despite the significant fluctuations and severe seasonal variations observed at the three selected test hydrological stations in the Huaihe River basin, the proposed model still achieves accurate prediction results. This offers new viewpoints on forecasting runoff in basins characterized by significant flow variations.

Keywords

• Matching pursuit (MP)
• Multi-scale daily rainfall data
• Denoising
• Fully stepwise variational mode decomposition (FSVMD)
• Permutation entropy (PE)
• Runoff prediction