Seismic Source Parameters Inversion And Tectonic Motion Analysis Based On Deep Learning And InSAR Technology

Earthquakes that occur around the world every year are often accompanied by very serious consequences,which will not only cause huge economic losses,but also seriously threaten the safety of human habitation.Robust and accurate source parameters,mainly referring to the geometric parameters of faults and slip distribution,are of great significance for earthquake disaster response and assessment,earthquake mechanism and medium and long-term forecasting.Different from the short-term large-scale crustal movement of earthquakes,tectonic movement usually refers to the slow crustal movement under the action of the earth’s internal dynamics.The study of tectonic mo-vement is of great significance to the formation of geological landforms,earthquake mechanism,engineering construction,natural resources and the long-term development of human beings.Synthetic Aperture Radar(InSAR)using interferometric technology can measure the earth’s surface deformation with submillimeter accuracy,and has the characteristics of high spatial resolution,high precision,wide coverage,continuous monitoring,all-day,all-weather,etc.In order to fully exploit the potential of InSAR technology in the field of source parameters and tectonic movement,In this paper,the inversion method of seismic source parameters and tectonic movement analysis based on InSAR technology are studied.The main research contents and innovations include:(1)Aiming at the problem that the large amount of coseismic InSAR deformation data cannot be manually inverted by traditional nonlinear algorithms,an overall inversion framework based on deep learning and InSAR data for automatic real-time inversion of seismic source parameters is proposed.In the inversion framework,the source parameter inversion model based on InSAR data and Res Net(ESPI-Res Net)can perform fault type classification and fault geometric parameter inversion.This paper generates a simulated coseismic interferogram dataset based on the existing seismic and geological data,trains and validates the source parameter inversion model on the simulated dataset,and compares and selects the optimal inversion network structure.The simulation test set results show that the fault classification accuracy is 99.6%,and the fault geometric parameter inversion results have a low RMSE.Based on the trained source parameter inversion model,this paper inverts the fault types and geometric parameters of the2020 Mw 6.3 Yutian earthquake,2017 Mw 6.5 Jiuzhaigou earthquake and 2016 Mw 5.9 Menyuan earthquake.The results are in good agreement with the existing source parameter data.(2)The inversion process of fault coseismic slip distribution based on InSAR data is realized,and the coseismic slip distribution of the 2016 Mw 5.9 Menyuan earthquake is obtained by inversion of the fault geometric parameters obtained from the source parameter inversion model and the Sentinel-1 ascending and descending orbit InSAR data.The inversion results show that the maximum slip of the Menyuan earthquake is0.76 m,which is concentrated near the center of the fault,about 9.4 km underground,and the fault slip rapidly attenuates to 0.3 m in the range of 3-4 km.Comparing the simulated observation of fault coseismic slip distribution with the actual observation by InSAR,the root mean square value is 4.36 mm,and the residual error of 93.7% of the observation points is less than ±1.5 cm.(3)Aiming at the unknown or controversial problems of the fault slip rate and strain distribution and the vague regional correlation still existing in the analysis of the tectonic movement of the Qinghai-Tibet Plateau.A section of about 1200 km long and 250 km wide spanning the western Qinghai-Tibet Plateau from the Altun fault to the southern margin of the Himalayas was used as the experimental area while Sentinel-1 data from 2015 to 2021 and GPS data are ueed as the data source.In this paper,the tectonic movement of the experimental area is analyzed,and the main conclusions and contributions are as follows:Using the SBAS-InSAR technology to invert the large-scale average linear InSAR radial deformation rate map in the experimental area,the deformation rate of96% points is distributed in the range of-20mm/yr～5mm/yr.Based on the deformation rate map,the slip rates of the main left-lateral strike-slip faults in the experimental area are relatively low,which are Riganpei Co(0.4±0.3 mm/yr);Altyn Tagh(5.1±0.3 mm/yr);Bairebu Co(1.1±0.2km));Yanghu(5.5±0.2 mm/yr).Due to the influence of horizontal tectonic movement signals,permafrost freeze-thaw cycle signals and noise,the single-track InSAR deformation rate map is not sensitive to the vertical tectonic movement trend of the Qinghai-Tibet Plateau.Combined with GPS and InSAR data,this paper inverts the horizontal velocity field and regional strain field of the experimental area.The east-west velocity field reveals that in the experimental region of 35-36°N,with the increase of latitude,the eastward movement rate gradually decreases,and roughly decreases to 0 in the Anyimaqen-Kunlun suture zone,and the higher latitude region has a trends in the Western movement.The regional strain field shows that the strain is concentrated near the Altyn Tagh fault,with a strain rate of 4-5 μstrain/yr,while the Yanghu-Changliangshan-Riganpei strikeslip fault in the western Tibetan Plateau has a lower strain rate of about 0-1.5μstrain/yr /yr.The sudden strength difference of the lithosphere at the boundary of the Qinghai-Tibet Plateau may be the main reason for the concentrated distribution of strain on the large-scale faults at the boundary of the Qinghai-Tibet Plateau.