Under the guidance of the abovementioned basic prediction theory, the development of deep mineral resource prediction is focused on solving major scientific issues, such as the mechanism of 3D structural reconstruction of deep mineralization space, the positioning mechanism of deep mineralization, and the method of deep mineralization prediction. Research needs to be carried out from three aspects: geological prospecting models of metallogenic systems, multiscale 3D geological modeling, and 3D quantitative prediction (Fig. 1).
The mineralization system is a concept developed on the basis of systems science (Zhai, 1999; Li, 1996). The mineralization system is a product of the geological system process and is a specific part delineated based on economic needs and production technology. Wyborn et al. (1994) proposed that the original intent of the mineralization system was to carry out mineral resource prediction for elements of prediction. According to Wyborn's definition of the metallogenic system, this system includes all the elements that control the generation and preservation of the deposit and the genesis of the source-transport-storage-preservation process that formed the deposit. The metallogenic system is unique in geological history. A deposit and traces of preserved objects are the final products of the metallogenic system. Transforming the process of mineralization and its key factors (source, transport, storage, etc.) into predictable specific spatial elements of mineralization is the key to prediction and target delineation (Hagemann et al., 2016).
Studies of the geological prospecting model of the metallogenic system focus on summarizing the 3D structural characteristics of six metallogenic systems (on a geological basis); at the scale of metallogenic belts, guided by the theory of the metallogenic system, studies summarize the "source, transport, aggregation, and transformation of different types of mineralization". Studies on the preservation of 3D structural characteristics guided by the "trinity" prospecting prediction theory at the prospecting area scale summarize the 3D structures of "metallogenic geological bodies, metallogenic structures and structural planes, and metallogenic features" of different types of mineralization.
Research on multiscale 3D geological modeling focuses on breakthroughs in such work at two scales: the exploration area and the metallogenic belt (see Table 1). Multiscale layered and superimposed 3D geological modeling methods are key research areas. In an exploration area with detailed geological data and effective modeling depth, 3D modeling software is used in combination with an exploration profile and borehole data in a multivariate geological database, and a 3D geological model is established. In large-scale and small-scale areas (metallogenic belts) where geological data are scarce, the 3D spatial reconstruction of deep and marginal metallogenic geological anomalies can be conducted through geological knowledge reasoning and 3D gravity and magnetic inversion, and this work can be constrained by the shallow 3D solid model.
Modeling scale Prospecting area Metallogenic belt Elevation range According to drilling depth 3 000 m underground Data Large amounts of (1) drilling and laboratory data; (2) a prospecting line profile; small amounts of (1) geophysical data, (2) geochemical data, etc. Large amounts of (1) surface geological survey measurement data and (2) geophysical data; a small number of (1) measured profiles; (2) borehole and laboratory data. Modeling method (1) Contour splicing based on adjacent sections (Micromine software, Surpac software, etc.); (2) radial basis function (RBF) spatial interpolation based on sampling points of adjacent boreholes (Leapfrog software, etc.). (1) Modeling method based on human-machine interactive editing (GeoCAD and other software); (2) modeling method based on geological knowledge and spatial interpolation; (3) geophysical 3D inversion method. Modeling content Ore body, specific lithology, ore-forming indicators, fracture zones, ore-forming structural surface, underground 3D primary halo, 3D model of grade distribution, etc. Surface model, stratum model, fold model, fracture model, rock model, alteration zone model, 3D gravity and magnetic model, etc.
Table 1. Two scales of 3D geological modeling work
With the development of mineral resource evaluation theory, various mineralization prediction methods are also constantly improving. Mineralization prediction methods can be classified into two categories: knowledge-driven and data-driven (Carranza, 2009). Knowledge-driven methods assign various parameters based on the knowledge and experience of experts to integrate multiple pieces of information. The data-driven method is based on quantitative analysis of the correlation between predicted factors and known mineral points, and mineralization prediction is carried out according to the established mathematical model. The mathematical models commonly used for mineralization prediction include weight of evidence, logistic regression, artificial neural networks, support vector machines and random forests (Ouyang et al., 2019).
The delineation of the mineralization system based on the theory of mineralization systems and the deep learning method guarantees a high degree of accuracy both in theory and in data mining methods. For the final delineated range of the metallogenic system, combining specific field investigations and existing research on strata, rock masses, minerals and deposits, the prediction model of the metallogenic system can be perfected, and the reliability of the metallogenic system can be classified. A typical metallogenic system in the study area should be selected as the model area to explore the reflection of resource estimation parameters, such as the range, depth, and ore-bearing rate of the mineralized geological body in the area, to realize the "positioning" of the metallogenic system along with its "quantity" and "probability" (Xiao et al., 2010).