Bedrock maps serve as a crucial source of information in mineral exploration, offering valuable insights into the distribution of the elements of a mineral system. They play a vital role in guiding exploration efforts and generating vectors for locating potential mineral deposits.
In this blog article, we will discuss how AI methods can enhance bedrock mapping, paving the way for more accurate maps and efficient mineral exploration.
Limitations of traditional approaches
Traditionally, bedrock mapping in remote or under the cover areas relied on visual interpretation of geophysical layers, such as magnetic and radiometric data, alongside the extrapolation of known field data collected by geologists to estimate lithologies in unsampled areas. While this approach has proven effective in many scenarios, it often suffers from a high degree of subjectivity and becomes increasingly challenging in greenfield areas, where limited field data is available.
What can AI bring to the table?
With the advent of advanced machine learning algorithms, particularly clustering techniques, we are witnessing a transformative shift in how bedrock maps are generated. By leveraging multivariate remote sensing datasets—including hyperspectral satellite imagery, RADAR data, and airborne geophysical data—these numerical tools facilitate pattern recognition and can be used to generate preliminary maps in the absence of reliable and/or detailed geological maps. This capability far surpasses the limitations of human interpretation, allowing exploration of complex, high-dimensional datasets that were previously beyond our reach.
How can Computer Vision enhance data-driven bedrock maps?
Computer vision models can serve as excellent tools for extracting inherent patterns from data and enhancing the accuracy of AI-based maps. Unlike classic AI tools, these models are capable of learning from the spatial context within geoscientific datasets, thereby improving predictions by understanding the connections between different pixels and regions in an image. This capability is particularly powerful in mineral exploration, as events being mapped (e.g., mineralization) are often correlated in space.
Integrating multisource geophysical data to assess subsurface geology
Bedrock mapping in areas with extensive surficial deposits is challenging, particularly in regions with limited outcrop exposure e.g. deserts, glaciated or forrested terrains. Integrating airborne magnetic and electromagnetic imagery offers valuable insights to enhance existing bedrock maps and guide future field campaigns. For instance, convolutional neural networks (CNNs), a type of computer vision architecture, can be utilized for multisource data fusion and clustering without requiring labeled data. These networks can learn to identify relevant patterns across multiple scales, grouping pixels based on local and regional similarities.
This approach was applied in Outokumpu, Finland, a region 60% covered by till deposits and characterized by diverse rock types, including metasedimentary rocks and mafic and felsic igneous rocks. We trained the model using total magnetic intensity data, apparent resistivity, and both the real and imaginary components of the electromagnetic field. Ultimately, ten classes were selected based on expert evaluation and existing knowledge of the area.
Fig. 1 Integration of magnetic and electromagnetic datasets using convolutional neural networks to generate AI-based bedrock maps. The resulting product displays 10 clusters, representing the lithological diversity of the subsurface.
Using AI to leverage hyperspectral satellite imagery for bedrock mapping in arid and desert regions
Hyperspectral sensors collect information from very narrow bands of the electromagnetic (EM) spectrum, enabling the mapping of subtle changes in rock chemistry down to the mineral composition level. However, the EM radiation captured by these sensors is mostly absorbed or back-scattered at the surface and does not penetrate deeply. Therefore, hyperspectral imagery is most effective for bedrock mapping in areas with high exposure, typically in arid and desert environments with low vegetation and minimal surficial deposit coverage.
We selected a region in Namibia with ideal conditions for utilizing hyperspectral satellite imagery to demonstrate our method. In this example, we applied a CNN to define 12 clusters using a data cube containing 285 spectral bands from the Earth Surface Mineral Dust Source Investigation (EMIT). These bands capture radiance across wavelengths ranging from 280 to 2500 nanometers. The results provide valuable insights for a preliminary analysis of the area's geology and, when combined with field data, can refine existing maps and produce a final exploration-oriented map.
Fig. 2 Using a data cube of 285 images from the EMIT project to generate an AI-based bedrock map. The resulting product identifies 12 clusters. The composite image was created using bands from the visible, near-infrared, and shortwave infrared regions.
Conclusion
In conclusion, AI is a powerful tool for helping geologists extract valuable information from geoscientific datasets to advance mineral exploration. Compared to classic machine learning algorithms, computer vision methods are notably good at learning local and regional features, facilitating the generation of reliable and accurate bedrock maps. This capability allows the integration of various data sources, including airborne magnetic and EM data and hyperspectral satellite imagery, according to the geological context and characteristics of the area of interest.
Ready to enhance your mineral exploration efforts? Our AI-driven solutions deliver AI bedrock maps and integrate diverse datasets to tackle your toughest challenges.
Contact us today and let’s unlock your project’s full potential!
Author: Victor Silva dos Santos, Data Scientist, Equivest
Comments