Magnetotelluric(MT)three-dimensional inversion has the advantages of simple data preprocessing, the model is close to actual situation, and the inversion result is more reliable and stable. It is one of the most advanced research topics and would take the place of the dominant two-dimensional inversion definitely. With the improvement of computing capability of computers and the breakthrough in inversion methods, great progress was made in MT three-dimensional inversion in recent years, from the theoretical research and test of this method at the beginning to the current application to practical data interpretation. For the great computation amount of MT three-dimensional inversion, current MT three-dimensional inversion algorithm programs are all implemented in parallel way and it is recommended to do three-dimensional inversion calculations on supercomputing system to make better use of computing resources and improve the inversion efficiency.

Different from the MT three-dimensional inversion algorithm programs which have basically realized the utility function, the practical application of MT three-dimensional inversion is still in an early stage. Users should be familiar with the use of multiple software and fulfill the function manually with the help of the software as follows: generating the files required for the inversion program, connecting to the supercomputer to upload data, inputting the command to perform the inversion, etc. The process of manually connecting and operating calculations is the most primitive cloud computing. All processes need to be done manually, which would cause not only heavy workload and the complicated operation, but also the problems for the long-term effective preservation and management of complex inversion data.

To conquer this, we develop independently a three-dimensional magnetotelluric inversion cloud computing system, toPeak, using Delphi language. This paper introduces some main features of toPeak. To begin with, system design and analysis are carried out in combination with the current situation and system structure and functions are realized. The main idea is to realize a set of cloud computing system platform based on server-client(C/S), on the basis of perfect inversion data management, integrate the most advanced three-dimensional magnetotelluric inversion algorithm program in the cloud, and connect through the Internet to realize all the system functions of three-dimensional magnetotelluric inversion. Then, the different parts of toPeak are introduced separately, including design structures and designs. The server is deployed on the supercomputer system(supercomputing)to receive the data for inversion tasks, configure and manage the storage of the inversion result data. Combined with the Internet connection, the server and the Internet together constitute a computing cloud. The client is deployed on the users’ windows operating system, including Windows visual data integration processing software and Internet operation middleware. The client is designed on the basis of object-oriented programming ideas, with data as the core, using data engineering objects to encapsulate and store all MT data, process and interpret the results, realize data processing inversion and other operations around this data project, and display the process and results of these processing and inversion in graphics using visualization technology. Internet operation middleware connects the client and server based on the SSH protocol to realize data processing and inversion, transmission and command sending and receiving. Furthermore, the whole work flow of inversion using toPeak and parts of procedure of it are shown. At last, some inversion results from toPeak are displayed. toPeak has realized the full functions require for implementing three-dimensional inversion and can grid, process and select, inverse and explain the data. It is a good tool for the practical use of three-dimensional inversion.

Many synthetic model studies suggested that the best way to obtain good 3D interpretation results is to distribute the MT sites at a 2D grid array with regular site spacing over the target area. However, MT 3D inversion was very difficult about 10 years ago. A lot of MT data were collected along one profile and then interpreted with 2D inversion. How to apply the state-of-the-art 3D inversion technique to interpret the accumulated mass MT profiles data is an important topic. Some studies on 3D inversion of measured MT profile data suggested that 2D inversions usually had higher resolution for the subsurface than 3D inversions. Meanwhile, they often made their interpretation based on 2D inversion results, and 3D inversion results were only used to evaluate whether the overall resistivity structures were correct. Some researchers thought that 3D inversions could not resolute the local structure well, while 2D inversion results could agree with the surface geologic features much well and interpret the geologic structures easily. But in the present paper, we find that the result of 3D inversion is better than that of 2D inversion in identifying the location of the two local faults, the Shade Fault(SDF)and the Yunongxi Fault(YNXF), and the deep structures.
In this paper, we first studied the electrical structure of SDF and YNXF based on a measured magnetotelluric(MT) profile data. Besides, from the point of identifying active faults, we compared the capacity of identifying deep existing faults between 2D inversion models and 3D models with different inversion parameters. The results show that both 2D and 3D inversion of the single-profile data could obtain reasonable and reliable electrical structures on a regional scale. Combining 2D and 3D models, and according to our present data, we find that both SDF and YNXF probably have cut completely the high resistivity layer in the upper crust and extended to the high conductivity layer in the middle crust. In terms of the deep geometry of the faults, at the profile's location, the SDF dips nearly vertically or dips southeast with high dip angle, and the YNXF dips southeast at depth. In addition, according to the results from our measured MT profile, we find that the 3D inversion of single-profile MT data has the capacity of identifying the location and deep geometry of local faults under present computing ability. Finally, this research suggests that appropriate cell size and reasonable smoothing parameters are important factors for the 3D inversion of single-profile MT data, more specifically, too coarse meshes or too large smoothing parameters on horizontal direction of 3D inversion may result in low resolution of 3D inversions that cannot identify the structure of faults. While, for vertical mesh size and data error thresholds, they have limited effect on identifying shallow tectonics as long as their changes are within a reasonable range. 3D inversion results also indicate that, to some extent, adding tippers to the 3D inversion of a MT profile can improve the model's constraint on the deep geometry of the outcropped faults.