Applying Compressed Sensing,Ambient Noise Processing,and Machine Learning To Earthquake Data In Sichuan And Yunnan

Sichuan and Yunnan provinces,located on the east side of the collison zone between the Indian plate and the Eurasian plate.It is one of the areas with the most active seismicities and serve earthquake hazards in Chinese Mainland.It is a key area for seismological researches.We apply compressed sensing,ambient seismic noise,and machine learning in earthquake monitoring and data processing using the Sichuan and Yunnan seismic records.In data processing,we utilize machine learning method to select reliable earthquake data for analysis.As the volume of the data is growing,it is very inefficient to manually select.With the development of the machine learing,we want to make the machine 'learn' our picking data skills to help select reliable earthquake data for analysis,instead manually picking.We apply ResNet18,a very efficient method in computer vison recognition area,to select reliable data for PGA and Ds analysis,and the accuary reaches 95%and 98%respectively.Our seismogram discriminator works well in selecting proper analysis data in Sichuan-Yunnan earthquake dataset.Normally,earthquake monitoring relys on travel time table and syntheticseismic waveforms with known underground velocity models.When an earthquake occurred,we compared the travel time information and the waveforms from the new coming earthquake with the ones from travel time table and synthetic seismic waveforms.We choose the source parameters from the one in the trave time table and synthetic waveforms which has the smallest misfit and most matching waveforms with the new coming earthquake to be the source information.These methods largely depend on the accuracy of the underground velocity model.We propose a novel method to obtain the travel time table and seismic waveforms of the study area without using underground velocity models.Applying compressed sensing,we contruct the travel time table and the seismic waveforms through analyzing the historic earthquake data,which largely minimize the influence from the underground velocity models.We apply the constructed travel time table and seismic waveforms by compressed sensing to the earthquake search engine for earthquake monitoring.Earthquake search engine,utilizing fast web search technology for seismology,is a new method which can obtain earthquake source parameter rapidly after an earthquake occurrs.We apply the compressed sensing and earthquake search engine in Jinggu,Yunnan for earthquake monitoring.Seismic velocity is an important geophysical property.Usually,when we solve the forward modeling problems,the seismic velocity is a constant value.Dynamic and static stress changes,fluid migration,fault zone damage and healing,and nonlinear response of near-surface material to strong ground motion have all implicated in causing changes to the measured seismic velocities.It is very difficult to measure the seismic velocity in the Earth directly.We measure the seismic velocity changes using ambient seismic noise.We obtain the empiral Green's functions between two seismic stations by cross-correlating the ambient seismic noise from two stations.By analyzing the time differences between the coda of the empirical Green's functions,we obtain the seismic velocity changes.We mainly study the seismic velocity changes in Longmenshan faults before and after Lushan earthquake occurred.The results indicate that after Lushan earthquake,the seismic velocity in the northern section of the Longmenshan faults increases.Combined with b-value and stress analysis,we think that the seismic velocity increase is caused by the increased stress fields.