Machine Learning Utilization for Waveform Denoising and Hypocenter Determination of Microseismic Event at Volcanic Geothermal Field

Abstract Microseismic monitoring is essential for evaluating reservoir within volcanic geothermal field. Nevertheless, numerous waveforms with low Signal to Noise Ratio (SNR) often leads to the inaccurate picking of event arrival time and subsequent hypocenter determination. This study aims to employ machine learning for denoising to improve the SNR and automate the arrival time picking. To address this objective, we utilize python packages called DeepDenoiser and EQTransformer. Additionally, the hypocenter prediction was performed using the GaMMA (Gaussian Mixture Model Associator) function. Our methodology is applied to a dataset comprising 43 days seismic recordings from 11 to 26 dynamic station which is moved after weeks of recording resulting 156 station location at the vicinity volcanic geothermal field. Results indicate significant noise reduction achieved through the machine learning-based denoising process. Furthermore, the number of hypocenters increased, as we obtained 103 events, which is 30% higher than with ordinary processing. In conclusion, the utilization of machine learning techniques shows promise in advancing microseismic data analysis in volcanic geothermal environments. Further evaluation and refinement of these methods could lead to improved monitoring and understanding of reservoir dynamics in such settings.