Machine Learning Models for Seismic Energy Forecasting and Spatial Correlation Analysis for the Himalayan and Indo-Burmese Regions

This study investigates seismic energy forecasting and interregional correlation analysis using advanced analytical techniques to enhance understanding of seismic behavior. The research compiles a comprehensive catalog of processed earthquake records from 1900 to 2024, ensuring completeness and homogeneity. By using valid empirical relationships, the study converts earthquake magnitudes into seismic energy, resulting in an annual seismic energy time series that reflects significant nonlinearity and nonstationary in energy release patterns. To analyze these complex behaviors, the ensemble empirical mode decomposition (EEMD) technique is applied, which effectively breaks down the seismic energy time series into intrinsic mode functions (IMFs) that capture various oscillatory modes embedded in the data. The study emphasizes the interregional correlations among the IMFs, focusing on four distinct interregional regions within the study area. This approach allows for a region-specific breakdown of seismic energy patterns, revealing how seismic behavior varies across different regions. Notably, the research identifies significant correlations among the IMF3 across regions, suggesting potential tectonic interactions and stress propagation patterns. The findings indicate that temporal shifts in IMF3 can reveal strong visual correlations among regions, despite geographical separations, thereby contributing to a more profound understanding of regional seismic dynamics. In addition, this study uses various models, including multilayer perceptron (MLP), long short-term memory (LTSM), Informer, and temporal fusion transformer (TFT), to predict seismic energy for 2025. The models were rigorously trained using IMFs to optimize performance across different regions.