Machine Learning Assists in Exploration of Geo-Resources

Scientists are smart, but they are still just blood meat-sacks. The thing about humans is it’s hard for us to look at a large amount of data. We forget things and get distracted. The huge amount of seismic data that we humans have collected in becoming a struggle for us to wrap our fleshy minds around. Now, researchers have a machine learning-based tool that can help in the automatic interpretation of this data.

Being able to detect subsurface geologic features from seismic data taken from the surface is of critical importance in understanding the basin evolution, geotectonic, resource exploration, and other processes that cause earthquakes. Over the years, humanity’s acquisition of seismic data keeps rising, and all that data is great and all, but it’s not very useful when its too much data to be interpreted. This task can range from extremely difficult to impossible. There are high-performance computers that analyze all this data and provide reports, but even those reports are difficult for human analysts to understand.

To solve this problem, the WIHG (Wadia Institute of Himalayan Geology) developed a neural-network-based system that automatically interprets 3D seismic data. This all-new approach is made possible by processing something called a ‘meta-attribute.’

Intrusions between old layers of sedimentary rock contribute significantly to the build-up and movement hot magma, which leads to an overburden. This is thought to be a structural trap for accumulations of hydrocarbons in the sedimentary basin. For example, in New Zealand, there are saucer-shaped magmatic sills that are embedded within layers of earth that were created between 145 and 33.9 million years ago, respectively. This succession has resulted in forced folds and hydrothermal vents forming above this area.

Scientists at WIHG were able to record this scenario. They were able to do so by computing Fluid Cube and Sill Cube meta-attributes. These are virtual attributes that are created by combining many discrete seismic attributes using an AI-based method. You can find more information on this study in ‘Tectonophysics,’ a peer-reviewed journal.