Development of wavelet-based machine learning models for predicting long-term rainfall from sunspots and ENSO

Abstract

Abstract The variations in rainfall and its spatial and temporal distribution in wet and dry seasons have increased substantially globally owing to the effect of climate change. These disparities can lead to droughts and severe water shortages, as exemplified by the unprecedented drought in Taiwan in 2021, which is considered the worst in 50 years. From a broader perspective, the overall climate and water resources on Earth are influenced by factors, such as the El Niño phenomenon and solar activity. Accordingly, this study examines the relationship between rainfall and planetary- or large-scale influencing factors, such as sunspots and the El Niño-Southern Oscillation. Additionally, rainfall patterns under various conditions are predicted using machine learning models combined with wavelet analysis. These models use 60-years historical data to build models, and the Bayesian network model exhibited the best overall prediction accuracy (85.7%), with sunspots emerging as the most influential factor. The novel findings of this study strongly confirmed that the relationship between sunspot and local rainfall patterns can serve as a valuable reference for water resources management and planning by relevant organizations.

Keywords

• Planetary scale
• Sunspots
• Rainfall pattern
• Machine learning
• Water resources
• Bayesian network