Seismic Event Detection and First Arrival Picking Using Continuous Wavelet Transform and Machine Learning Techniques

Seismic data interpretation plays a crucial role in comprehending subsurface structures, particularly in industries such as oil and gas exploration, earthquake monitoring, and geophysical research. Conventional techniques for detecting seismic events and identifying first arrivals often encounter challenges when dealing with complex data patterns and noise, resulting in inaccurate outcomes. This study introduces an innovative strategy that combines continuous wavelet transform with machine learning techniques to increase the accuracy and effectiveness of seismic event detection and first arrival identification. The continuous wavelet transform provides robust time–frequency localization, which improves feature extraction from seismic data, whereas machine learning algorithms like K-means clustering automate data segmentation and pattern recognition. The proposed approach is evaluated on both synthetic and actual seismic datasets. The findings demonstrate that the combined continuous wavelet transform and machine learning technique significantly enhances the accuracy of first arrival picks, even in noisy environments, surpassing traditional approaches. In particular, the algorithm maintained a significant accuracy level (82%) when tested with synthetic datasets that included as much as 30% random noise, and impressive accuracy of up to 92% for real datasets, involving Aqaba and Yilmaz shot gathers, highlighting its robustness and reliability. This development can improve seismic interpretation practices, facilitating higher-resolution subsurface imaging and more dependable geological modeling.