Research On 3D Joint Inversion And Application Of Gravity And Magnetotelluric Data

Geophysical inversion is an important method to process and interpret geophysical data.It can reconstruct the subsurface distribution of petrophysical properties and geological structures from geophysical data and improve the ability of deep mining,and geological disaster monitoring and detection.To this end,achieving the purpose of providing technical support to ensure energy supply for national economic development and to protect the safety of people’s lives and property.As typical exploration methods,gravity and magnetotelluric(MT)methods are widely used since their low cost,high efficiency and wide measurement range.Gravity method has higher lateral resolution and is widely used in energy and mineral exploration.By contrast,MT method has higher vertical detection ability and is widely used to explore deep structures.However,due to the gravity signal decreasing exponentially with increasing depth,the gravity inversion has low vertical resolution and is not sensitive to deep structures.At the same time,due to the sparse distribution of MT observations in most cases,the MT inversion is usually insensitive to shallow structures and has low lateral resolution.Therefore,independent inversion of gravity or MT cannot achieve models with a common structure in most cases due to different data having different sensitivity and inversion results having different resolutions.Joint inversion can make full use of the technical advantages of various geophysical exploration methods,realize information fusion and complementarity,reduce the multisolution of inversion,and obtain stable inversion results.For inversion problems,how to realize high-precision and high-efficiency geophysical inversion is the basis to make the inversion techniques practical.The traditional geophysical inversions took structured grids.Compared with structured grids,unstructured grids have the advantage of fitting the complex terrain and underground geological structures and being able to refine grids locally as needed.Therefore,the inversion based on unstructured grids plays an important role in improving the inversion accuracy.Moreover,with emerging of artificial intelligence,the efficient inversion of geophysical data based on machine learning plays an active role in improving the practical ability of geophysical inversion.Based on these backgrounds,this paper firstly studies the three-dimensional(3D)independent inversion of gravity and MT data respectively and realizes the 3D inversion of gravity based on unstructured grids,3D inversion of gravity based on machine learning,3D inversion of MT data based on unstructured vector finite-element method.Furthermore,based on these individual inversions,the 3D joint inversion of gravity and MT data based on unstructured grids is studied and realized.For the 3D inversion of gravity data,the traditional regularization inversion and the modern machine learning inversion are studied respectively in this paper.First,the 3D gravity data inversion based on unstructured grids is completed.In this process,the "skin effect" problem of gravity inversion is solved by using the depth weighting function based on the sensitivity matrix,and the model constraint is realized by adapting the roughness constraint that is suitable for unstructured grids.Apart from that,based on the existing machine learning inversion methods,this paper configures a distinctive solution for solving inversion problems and establishes a new machine learning network – Dec Net that has the advantages of small memory requirements,low computing cost,high inversion accuracy,and reconstructing underground density anomalies efficiently.Theoretical model tests and field data application verify the effectiveness of the above-mentioned gravity inversion methods.For 3D MT inversion,the objective function based on the unstructured vector finiteelement is constructed,and the derivative of the data fitting term is calculated by the implicit solution method.Then,the L-BFGS method is used to optimize the objective function.The effectiveness of the proposed method is verified by testing on models with flat and fluctuant terrain.Based on individual inversions,in order to realize joint inversion based on unstructured grids,this paper proposes a local Pearson joint constraint based on unstructured meshes.The virtual background grid is used to calculate the Pearson correlation coefficient between inverted model parameters in the local control area,and the correlation between model parameters is maximized by minimizing the local Pearson joint constraint.This method avoids calculating the model gradient of unstructured grids and is suitable for joint inversion on unstructured grids.Based on this,the step model and SEG model tests are used to verify the effectiveness of the local Pearson joint constraint method.Finally,the factors affecting joint inversion results are discussed in detail.To test the practicability of the above inversion methods,the gravity inversion towards the Olympic Dam mining area,the MT inversion towards the Longgang volcanic area,and the joint inversion of gravity and MT towards the Yellowstone super volcanic area are respectively implemented.For Olympic Dam,the lateral distribution of the high-density anomaly discovered by gravity inversion is basically consistent with the distribution of the mine with iron content larger than 5% provided by the geological information.This demonstrates that gravity inversion has high lateral resolution,and technical advantages for shallow mineral exploration.The MT inversion of the Longgang volcanic area reveals the 3D electrical structures from mantle to the upper crust.This demonstrates that MT inversion has high vertical exploration depth and is good at reconstructing deep structures.The joint inversion of gravity and MT in the Yellowstone volcanic area has obtained density and resistivity models share a common structure.This demonstrates that joint inversion can comprehensively utilize the technical advantages of gravity and MT,and obtain high-precision joint inverted results.Apart from that,for Longgang volcanic area,the Hunjiang Falt and Yalv Falt correspond well to the interfaces of the high and low resistivity.By utilizing the prior geochemical and geophysical information as supporting data,it is inferred that Longgang volcanic area is a single volcanic,and its upper magma is generated by the migration of magma from the mantle to the shallow crust along these faults and has not experienced the crustal magma chamber migration stage.However,the nearby Changbai Mountain volcano is a typical bimodal volcanic system.This volcanic evolution process deserves further study and discussion.Then,for Yellowstone volcanic area,the joint inversion results show that there are two underground lowdensity,low-resistivity anomalies.According to the geological and geophysical information,it is inferred that the shallower anomaly is the partially melted rhyolite magma chamber and the deeper anomaly is the partially melted basalt magma chamber.The connection of the two inverted magma chambers presented by density and resistivity models solves the controversial problem of whether these two magma chambers are interlinked in seismic and geodynamic studies.In addition,since the geophysical research about the magma migration system between mantle and crustal is still not finished,the joint inversion results also provide a reference for further research.In conclusion,this paper systematically realizes single inversions and the joint inversion of gravity and MT data by theoretical research and practical applications.Then,demonstrates the advantages of gravity inversion in mineral resource exploration,MT inversion in deep structure reconstruction,and joint inversion in comprehensive utilizing the information of gravity and MT data and achieving high-precision inversion results.Particularly,the unstructured joint inversion based on local Pearson joint constraint avoids calculating the gradient of unstructured grids and provides a technical reference for promoting the joint inversion of geophysical datasets based on unstructured grids.The inversion of Longgang volcano and Yellowstone volcalno is informative for the study of magma migration and evolution,and plays an vital role in monitoring and warning magmatic and seismic activities.