Mineral exploration follows an information workflow that includes data acquisition, information extraction, and decision-making using all information. The practice of mineral discoveries has evolved from data-driven approaches to information-driven workflows. In this paper, we examine how research efforts have evolved across these three stages and argue that advances have focused primarily on information extraction, while data acquisition and decision-making have received comparatively less attention. We suggest that future research should adopt an information-centered perspective that considers the flow of information across the entire workflow rather than optimising individual stages in isolation. From this perspective, we discuss information-efficient acquisition approaches, such as ergodic sampling, and the growing role of artificial intelligence (AI) in integrating diverse geoscientific datasets to support more informed decisions. We argue that balancing research across all three stages will improve the efficiency and effectiveness of mineral exploration.
Introduction
Modern mineral exploration is based on information — gathering information in the field through data collections and extracting information from data for decision-making. Each stage of the exploration, ranging from reconnaissance, to target identification for discovery, to deposit delineation upon a discovery, has its unique needs for information as well as the requisite tools to gather or extract the information. The geophysics component of mineral exploration is no exception.
Much of the geophysical technology development in mineral exploration over the last three decades has focused on advancing instrumentation and the inversion of data to obtain physical property models. In particular, tremendous advances have been made in the inversion of geophysical data since early 1990. This line of R&D has essentially focused on information extraction from available data. In contrast, the way geophysical data are acquired has not changed significantly. Similarly, the way geophysical information is used, or interpreted, has also not changed significantly. These contrasts are unequivocally reflected in the disparities in the amounts of publications produced in applied geophysics journals. Figure 1 shows the numbers of publications on the data acquisition, inversion, and integrated interpretation in the SEG Digital Library. Published works on the inversion outnumber the other two categories by more than an order of magnitude.
Thus, advances in different geophysical technologies for information gathering, extraction, and utilisation are highly uneven. While the middle part on the information extraction has received the greatest attention and experienced the most advances, the front and back ends have not fared as well. To state this situation differently, we now have much better means to extract the information from available data, but we may not have the best means to gather the data for more information, nor do we have the advanced means to fully leverage the extracted information in the context of the underlying geology for decision making that leads to mineral deposit discoveries.
In this article, we address the front and back ends of the information stream in mineral exploration. We discuss the latest technological development in the strategies and methodologies for efficient geophysical data acquisition and in automated integration of multiphysical and geological information for decision-making. Specifically, we present the efficient data acquisition using ergodic sampling (Zhang and Li, 2023, 2026) at the front end and the machine learning-assisted information integration at the back end (e.g., Melo and Li, 2021; McAliley and Li, 2023). The former increases the information bound in the data, which feeds into the inversion algorithms developed in the last three or four decades, and the latter maximises the joint value of extracted information for imaging geology and facilitating discoveries. There have been many advances made by the community of researchers and practitioners. For brevity, we use primarily our group’s work to highlight the recent advances and road ahead.
Two key research directions in the mineral exploration geophysics
We present two developments here to highlight the new technologies available and the need for much more R&D in these directions. The first is the newly developed efficient geophysical data acquisition using ergodic sampling and the second is the machine learning integration of information.
Ergodic sampling for efficient airborne data acquisition
Geophysical data are collected as discrete samples by necessity and then interpolated or reconstructed to form a grid coverage of the data area. Typical examples are gravity or magnetic data. Among these, airborne magnetic surveys are flown exclusively along parallel lines spaced equally apart. Such equal-spaced line designs are rooted in the Nyquist sampling strategy. This type of survey strategy may be sound and necessary when evaluated using Fourier transform-based signal reconstruction. However, compressive sensing theory (Donoho, 2006) has shown that this survey strategy grossly oversamples the data unnecessarily, leading to unwarranted cost in time and expenditure. The increased time and cost bleed the exploration effort and reduces the discovery rate.
The newly developed ergodic sampling theory (Zhang and Li, 2023, 2026; Zhang et al., 2025) enables the acquisition of nearly the same information but uses only a fraction of the survey lines spaced irregularly in an optimised pattern. This strategy leverages the compressive sensing theory but addresses the practical aspect of data sampling among the three components of compressive sensing, namely, the sampling, transform, and reconstruction. The key is that ergodic survey design deploys the smaller number of lines in an optimised irregular pattern to achieve nearly the same information sampling ability (ISA) as the dense lines. We note that compressive sensing in its original form relies on random sampling, which is a form of irregular sampling, whereas ergodic sampling is a form of optimised non-random, irregular sampling. The quality of the ergodic sampling pattern is quantified by four different information sampling ability (ISA) criteria, which are sampling interval distribution, angle distribution, sample density distribution, and spectral resolution function (SRF) (Zhang and Li, 2023). The design of an ergodic sampling begins with the standard dense line pattern and performs an optimisation to find the ergodic pattern that matches the four ISA of ergodic pattern with those of the standard dense lines. The relationship between ISA and sample number is nonlinear. Therefore, we can acquire a limited number of data but maximise the information through ergodic survey design.
Figure 2 demonstrates the effectiveness of ergodic sampling through a field trial consisting of two independent airborne surveys. The ergodic pattern in this example uses only 50% of lines and the resultant data grid of total-field magnetic anomaly is virtually the same as that from the dense lines (Zhang et al., 2025).
This example and published work demonstrate that it is feasible and highly desirable to perform partial physical sampling in the field during the data acquisition and obtain the equivalent of traditional data set through post-acquisition computational reconstruction. We refer to this combination of partial physical sampling and numerical reconstruction as the computational geophysical acquisition. It will be able to substantially reduce the time and expenditure required for data acquisition and thereby expedite the exploration workflow and speed up discoveries.
Machine learning integration of information
There have been tremendous advances in the inversion of geophysical data over the last three decades. Virtually every single geophysical data type can now be inverted to produce physical property models to image the subsurface. The technology has matured to the degree that the services are being consolidated in the industry. Yet, it can be argued that the rate of discovery of world-class mineral deposits has not been commensurate with the development of these technologies. We believe that part of the reason could be the insufficient integration of geophysical information in the context of underlying geology. Such integration consists of two major components, namely, the integration of information from multiple geophysical data sets and the integration of geophysical information with geology and geochemistry.
There have been significant research and progress in multiphysics integration. Examples of advances in this area include joint inversion of geophysical and petrophysical data (Sun and Li, 2016), geology differentiation (Melo and Li, 2021), petrophysically guided inversion (Astic and Oldenburg, 2019). An important direction in mineral exploration R&D is also how to explore large regions, winnow the sterile areas, and then identify prospective areas using geophysics in a low-cost and automated manner. Figure 3 shows an example of unsupervised machine learning integration of multiple geophysical inversions to image a mineralisation zone. The method can be applied broadly in the green field exploration.
Meanwhile, integration of geophysical information with geology and geochemistry requires advanced methodology that can work with qualitative and conceptual geological and geochemical data that are implicit in nature. Whereas classical inversions are mostly ineffective with such information, ML/AI-assisted methods can excel with such data and lead to a new generation of algorithms.
Such algorithms and methods are needed to enable effective information integration and do so in a more time-efficient and objective manner. For example, conditional variational autoencoder (CVAE) can serve to capture conceptual information (McAliley and Li, 2023) and the vast amount of information from historical projects and interpretations (Hirsch and Li, 2025). Figure 5 illustrates a type of such geological conceptual information and a CVAE that can capture and use such information either in the inversion or in general information integration. Much remains to be researched and developed. There lies one category of challenges and opportunities in the mineral exploration of the future.
Concluding remarks
We have presented a brief review of historical development, current state of the art, and R&D directions in the immediate and near future in mineral exploration geophysics. Adopting an information-centric perspective, we have identified two areas of research in the mineral exploration that lag behind in technological development, namely, data acquisition to maximise the available information for the given expenditure and the integration of multiple sources of information to image geology for decision-making and deposit discoveries. We have highlighted here the recent advances in these two areas and discussed the potential road ahead.
Acknowledgements
We would like to thank Damian Arnold and Tom Davis for the invitation to write this article. The conceptualisation and work presented here have been developed primarily within the Geo-Multiphysics Research Consortium (GMRC) supported by sponsors from mineral and energy exploration industries.
References
- Astic, T. and Oldenburg, D.W. [2019] A framework for petrophysically and geologically guided geophysical inversion using a dynamic Gaussian mixture model prior. Geophysical Journal International, 219(3), 1989-2012.
- Donoho, D.L. [2006] Compressed sensing. IEEE Transactions on Information Theory, 52(4), 1289-1306.
- Hirsch, S. and Li, Y. [2025] Development of conditional variational autoencoder for electromagnetic inversion. IMAGE 2025, Expanded Abstract.
- McAliley, W.A., and Li, Y. [2023] Stochastic Inversion of Geophysical Data by a Conditional Variational Autoencoder. Geophysics, 89(1).
- Melo, A. and Li, Y. [2021]. Geology differentiation by applying unsupervised machine learning to multiple independent geophysical inversions. Geophysical Journal International, 227, 2058-2078.
- Sun, J. and Li, Y. [2016] Joint inversion of multiple geophysical data using guided fuzzy c-means clustering. Geophysics, 81(3), ID37- ID57.
- Zhang, M., Battig, E.and Li, Y. [2025] First field trial of ergodic sampling for airborne magnetic survey. The Leading Edge, 44(11), pp.831-837.
- Zhang, M. and Li, Y. [2023] Ergodic sampling: acquisition design to maximize information from limited samples. Geophysical Prospecting, 72(2), 435-467.
- Zhang, M. and Li, Y. [2026] Line-based ergodic sampling and application in airborne magnetic data acquisition. Journal of Applied Geophysics, 250, article 106234.
