top of page

Machine Learning in Geophysics

Machine learning algorithms (MLAs, aka AI) are powerful tools used to recognize patterns in high-dimensional data. MLAs are efficient methods for producing geological maps or improving existing maps as well as predicting subsurface resources. MLAs can be categorized into two groups: supervised and unsupervised methods. Supervised methods utilise training data that are comprised of a set of known observations. These data train a classification model (in the MLAs), which enables us to predict the group/class labels of samples previously unseen. The unsupervised method has no training data, and it identifies groups/classes within data. In MLAs, we can use geophysical, geochemical, radiometric, geological and remote sensing data.
P.S. Because of the policy of real data, information on figures has been removed. 
Geology Mapping Using Machine Learning and Geophysic Data

Geology Mapping Using Machine Learning and Geophysical Data

Numerous datasets (including geophysical, geochemical and geological) that are covering the study area can provide good knowledge of the subsurface. However, the integration of many datasets can be challenging for the interpretation as well. MLAs can overcome this challenge using an automatic method to recognize patterns in datasets. MLAs are efficient methods for producing geological maps or improving existing maps. They start with training, and, when learned, patterns will be applied to the datasets to generate predictions for data-driven classification and regression problems. MLAs show a good performance in predicting categories from training data. MLAs are one of the most effective methods for the prediction and classification of mineralizations as well as lithological units and geological structures.

Machine Learning and Geophysical Inversion

Machine Learning and Geophysical Inversion

While geophysical inversions reconstruct geological structures based on physical responses and geometries, machine learning methods use patterns in data to provide predictions and classifications (e.g. of lithology or mineralization). For a geometry (structure), different physical properties (such as density, conductivity, etc...) give different types of geophysical data. But in the inversion, different types of geophysical data from one geometry with different physical properties might not give the same model. Therefore, the geological models reconstructed from the inversion of geophysical data might be different. At this point, MLAs can find the pattern in the models to provide a single model using classification. 

bottom of page