Shear Wave Structure Of Typical Regions In NE China From Joint Inversion Of Receiver Function And Ambient Noise Dispersion

Seismic observation provides us an effective way to detect the inner structure of the earth.Through the comprehensive analysis of a large number of seismic observation data,it is possible to infer the potential hazards of dangerous areas such as volcanoes,and provide a dynamic evidence for the regional tectonic evolution.Northeastern China is located at the junction of the Eurasian Plate and Pacific Plate,it includes many typical tectonic units like Changbai volcano,Songliao Basin and Great Xing'an Range.The deep structure and formation mechanism of these tectonic zions have been studied by many previous researches,but it is still contriversial now.Are there magma chambers in the crust of the Changbai volcano? Where the magma come from? What is the relation between Songliao Basin and regional tectonic activities? Why there are significant differences in the deep structure of Songliao Basin and its surrounding areas? Those issues need a further research.In this work,we use abundent seismic data to analyse and discuss above issues.We finally present new 3-D S-wave velocity models of these typical tectonic zions by conducting joint inversion of the receiver function and the ambient noise dispersion.This paper introduces the basic principles and data processing flow of the receiver function and ambient noise method.We present a brief introdunction to the extraction method of receiver function and ambient noise dispersion curve,and list the simulated formulas of the receiver function and dispersion curve in layered medium.We also introduce the linear inversion method and nonlinear trans-dimensional Bayesian methods for receiver function and the dispersion curve.We study the detailed 3-D crustal and upper mantle structure beneath the active Changbai intraplate volcano in NE China by using a trans-dimensional Bayesian inversion of teleseismic receiver functions and Rayleigh-wave group velocity dispersions from ambient noise.More than 12,000 teleseismic receiver functions recorded at 78 seismic stations and 1573 group velocity dispersions,are adopted in this study.Receiver function H-? stacking measurements reveal a thick crust(~40 km)with a high Vp/Vs ratio(~1.8)beneath the Changbai volcano.Our joint inversion results show a heterogeneous crustal structure in the study region.A low-velocity body at depths of 8-15 km is visible directly beneath the Changbai volcano,which has a lateral extent of ~100km in the north-south direction and may reflect a large magma chamber in the mid-crust.Our results also reveal a 5~10km depressed Moho and a low-velocity anomaly in the uppermost mantle beneath the Changbai volcano.These features may indicate an upwelling channel of the asthenospheric material with a high mafic composition,and the mafic intrusion attaches to the bottom of the crust and thus deepens the Moho beneath the volcano.Our results support the notion that the Changbai volcanism is caused by hot and wet mantle upwelling associated with subduction-driven corner flow in the big mantle wedge above the stagnant Pacific slab in the mantle transition zone.Huge labor cost for receiver-function picking prevents its rapid development,it is thus necessary to develop rapid and accurate approaches to auto-process the mass seismic data.We present deep learning methods of CNN and RNN to auto-pick receiver functions.A total of 20 years' receiver functions from MDJ and BJT stations,are adopted to test our methods.We use those auto-picking receiver functions to estimate the crustal thickness,average Vp/Vs and azimuth-anisotropy of crust in order to make a comparative analysis with those results computed from manually selected receiver functions.Several groups' tests show that our methods are valid to auto-pick the receiver functions.Our receiver functions auto-picking methods is useful to build a reference model for permanent stations,so as to automativally select the receiver functions.On the basis of above supervised deep learning methods and characteristics of the seismic data,a semi-supervised deep learning SGAN method is proposed to auto-pick the receiver functions of portable stations.For this approach,although only a little labeled receiver functions are included in training dataset,it still obtains a high accuracy model for auto-picking.The SGAN method is applied to NECESSArray network located in NE China.Crust and upper mantle seismic structure around the Songliao Basin is inverted by those auto-picking receiver functions and Rayleigh wave dispersions.Our presented 3-D shear wave velocity model shows the thick sedimentary layers beneath the Songliao Basin and the Erlian Basin.The middle and lower crust beneath the Songliao Basin appears as high-V anomaly,while the lower and middle crust in Great Xing'an Range shows a characteristic of low-V anomaly.We thus infer that the Songliao Basin has a different formation mechanism with Erlian Basin.Low-V anomaly in the mid-crust of Great Xing'an Range indicates a hot melt material beneath volcanoes.For Jingpohu volcano,the upper crust reveals a high-V anomaly,whereas the lower crust and the upper mantle show low-V anomalies,with a depressed Moho,which reflects that it is related to the hot material upwelling from asthenosphere.Our research suggests that the deep structure of the Songliao Basin and Great Xing'an Range,may be influenced by the different tectonic activities.