Multi-scale 3D modeling by machine learning and numerical simulation for potential magma conduit in the Kalatongke district, Xinjiang, China

• Machine learning technology and numerical simulation technology were jointly used to solve hot scientific issues. • Multi-source geological, geophysical, and engineering data were integrated to build multi-scale three-dimensional models. • A three-dimensional magmatic conduit model was built and six high potential targets were delineated in the Kalatongke district. The Kalatongke district in the Central Asian Orogenic Belt has the largest magmatic Cu–Ni sulfide deposit in Xinjiang, China. After more than 40 years of continuous multi-scale scientific research on known deposits in the district, useful results have been achieved, and large volumes of geological data, geophysical data and engineering data have been accumulated. However, there are still many hot scientific issues in the Kalatongke district, such as the location of magma conduit, magma flow feature, the relationship between carbonaceous rock and mineralization, the stress characteristics during mineralization, and the stratification characteristics of intermediate–basic rocks. We used machine learning technology and numerical simulation technology to tackle the above-mentioned scientific problems from a district- and deposit-scale mineral system perspective. The specific district-scale research steps were as follows: (1) unsupervised K-means method was used to cluster data on four petrophysical properties (density, magnetic susceptibility, resistivity and seismic wave propagation velocity) to obtain a district-scale 3D geological body model, which was applied to reveal the ore-bearing rock and its surrounding rock; (2) faults were interpreted using gravity data, magnetic data, electromagnetic data and seismic data, and a 3D regional fault model was established to show the location of potential magma conduit; and (3) a stress field model and a pore pressure model reflecting fluid characteristics were obtained by numerical simulation technology, so that the district-scale stress feature in 3D space can be integrated with magma conduit. The specific deposit-scale research steps were as follows: (1) a secondary fault model was built by 3D modeling based on geological–geophysical data in order to indicate potential secondary magma conduit; and (2) based on 239 borehole data, a deposit-scale rock model was classified using neural network technology to show the spatial distribution feature of carbonaceous rock and orebody; and (3) the ratio of Cu to Ni geochemical data was used to obtain a Cu–Ni ratio model, which was used to determine the direction of magma flow. Based on these studies of district-scale and deposit-scale ore-forming system, the features of tension and extrusion stress were combined with the melt–fluid flow direction deduced from the Cu–Ni ratio model, and a 3D magmatic conduit model in the Kalatongke district was built. According to this magmatic conduit model and the metallogenic geological characteristics of the area, three A-rank and three B-rank targets were delineated, which can further guide mineral exploration in the district. The research methods and techniques described in this paper are useful for investigating potential magma conduits of magmatic Cu–Ni sulfide deposits elsewhere.