Development Of Regional-scale Seismic Velocity And Attenuation Tomography Methods And Its Applications

Obtaining underground medium physical properties(e.g.,seismic velocity and Q values)can help us better understand the regional tectonic structure,fault zone structure and seismogenic zone structure.Seismic tomography is a very important technique for exploring the Earth's inner structure.Generally speaking,seismologists use P waves,S waves and surface waves generated by earthquakes or extracted from ambient noise to image underground structures.This dissertation mainly focuses on application of newly developed regional-scale seismic tomography methods in different areas and developing a new surface tomography algorithm based on deep lerning technique.It mainly includes the following four parts:(1)The spatiotemporal distribution of seismic velocity changes(SVCs)are of great significance for understanding physical state of subsurface medium and earthquake early warning.Previous studies have found that SVCs,including velocity changes caused by earthquakes and seasonal velocity changes caused by temperature changes,exist in different areas of southern California.However,the background SVCs in the entire region of Southern California are not clear.Are SVCs in steady state?Where are they located?Are SVCs local or global?Since ambient noise cross-correlation based technique and repeated earthquake coda wave interferometery method cannot provide accurate spatial information of SVCs,we performed a newly developed time-dependent tomography algorithm on 17-years(2000-2016)seismic P-wave data in Southern California,and obtained three-dimensional P wave velocity changes in each time period.We found that the temporal and spatial characteristics of velocity changes in Southern California are complex and suggest that multiple factors(earthquakes,temperature and rainfalls,etc.)jointly result in the complexity of velocity changes in time and space.The fluctuation amplitude of the relative average velocity changes at the depth of 7-13 km in the San Jacinto fault area is larger than that of the shallow depth of 4 km.We suggest that most of earthquakes at depths of 7-13 km result in the medium fractures open and closure at this depth range,while medium at shallow depth of 4 km remains relatively intact.(2)Newly developed double-difference time-dependent tomography method was used to reveal P-wave velocity changes caused by Borrego Springs Mw5.2 earthquake on June 10,2016.The fault could be damaged and then healed when the earthquake occurred on the fault,which could result in the medium velocity changes in the fault area.In order to understand whether there is stress loading in the fault area before the earthquake and whether the fault is healed after the earthquake,we applied time-dependent tomography algorithm to the seismic data before and after the earthquake and we obtained velocity changes of the foreshock period(105 days before the main shock),coseismic period(within 15 days after the main shock),post-earthquake period 1(15 days to 60 days after the main shock)and postseismic period 2(60 days to 105 days after the main shock).We found that during the foreshock period Vp around the main shock area had a slight increase;during the coseismic period Vp in the main shock area began to decrease;during the postseismic period 1,the amplitude of Vp decrease reached the maximum(-2%)and aftershocks are within the area with Vp decrease;during postseismic period 2,Vp in the aftershock zone began to increase.Previous studies found that Vp decreases during postseismic period due to the fault zone damage caused by the main shock and aftershocks.Vp in the shallow crust decreased during both posseismic periods 1 and 2,indicating that the shallow crust did not start to heal,while Vp in the deep began to increase in postseismic period 2,indicating that the fault began to heal in the deeper zone.(3)In order to better understand the physical state of Gofar transform fault,east Pacific rise,we performed 3-D body wave attenuation tomography using seismic data recorded by 16 ocean-bottom seismometers distributed on Gofar transform fault and obtained 3-D Qp and Qs distributions of the entire oceanic crust in the area,which are more sensitive to fractures,the existence of fluid and temperature variations compared to seismic velocity.The result shows that Qs distributions along the transform are segmented,which is consistence with Vs.The main shock area exhibits high Qp,high Qs,high Vp,high Vs,and low Vp/Vs,while the east side(foreshock area)of the main shock area presents low Vp,low Vs,high Vp/Vs,low Qp,low Qs,and low Qs/Qp(<1).Combined with the results of velocity and attenuation tomography,we interpret the east side of the main shock area(foreshock area)as a high fracture zone with rich fluids.(4)We developed a surface wave tomography algorithm based on deep learning technique and applied it to different regions,and obtained results comparable to traditional methods.With the rapid increase in the number of stations,people can extract massive dispersion curves from surface wave signals generated from different sources(e.g.earthquakes,artificial sources,and ambient noises).The traditional method performs repeated inversion for each dispersion curve,which is time-consuming and results in some artifacts in the final velocity model.We use the convolutional neural network to learn the corresponding relationship between one-dimensional velocity models and the corresponding dispersion curves.This method requires a large amount of velocity models and corresponding dispersion curves.The trained model can be applied to predict the corresponding velocity models with the massive dispersion data at one time.We applied new method to the surface wave data in continental China and the surface wave data in the southern California and the obtained results are better than that of traditional methods.