Blog – Analysis and expert opinion
5 MIN. READ

Results of Project GEMA – Generation of High-Resolution Maps and Advanced Interpretation of Geological Data through Artificial Intelligence

Over its 18 months of execution (August 2023 – February 2025), the GEMA project has achieved significant progress in the integration of artificial intelligence applied to geophysics, reaching technical milestones that bring value to both research and applied exploration.

Proyecto GEMA

    • Development of an international database
      As a starting point, an international geophysical database was built, integrating radiometric, magnetic, hyperspectral, satellite, and digital terrain model data. This data collection, obtained from different geographies and under diverse operational conditions, has made it possible to create a robust repository for training and validating artificial intelligence models, ensuring the generalization capacity of the solutions developed.
    • Advanced processing methodologies
      The project has established a reproducible pipeline for data processing and normalization, including the homogenization of spatial reference systems (CRS) and the correction of common artifacts in magnetic data. These methodologies have substantially reduced data preparation times, improving the quality and traceability of the information used in modeling.
    • Artificial intelligence models developed
    1. Magnetic artifact correction (Levelling): An approach based on Bayesian models and spatial interpolation was implemented, resulting in an algorithm that significantly reduces the need for manual adjustments. This tool, also deployed as a plugin, has proven useful for homogenizing airborne geophysical surveys more quickly and consistently.
    2. Automatic lineament segmentation: Convolutional neural networks (CNN, U-Net architecture) were trained to detect and trace geological structures in magnetic maps. The results achieved high accuracy (with a Dice coefficient close to 0.8 in training sets), demonstrating the ability of these techniques to automate tasks that traditionally required months of manual interpretation.
    • Geomagnetic prediction for directional drilling (IFR2): A spatio-temporal model (INLA-SPDE) was developed to predict high-resolution variations of the magnetic field in drilling zones. This has optimized borehole orientation and reduced trajectory errors, being validated even against specific geomagnetic events recorded during the validation campaign.

    Validation and quantitative results
    Progress was verified through rigorous validation processes:

    • Levelling algorithms were compared against manually corrected flight data, showing high spatial consistency and improved correlation coefficients (R²).
    • Lineament models were evaluated with standard metrics (precision, recall, F1, and Dice), confirming their ability to generalize in regions not included in the training.
    • The geomagnetic prediction model was validated by minute-by-minute comparison of Fx, Fy, and Fz components against independent stations, achieving low mean absolute errors (MAE) and successfully anticipating episodes of magnetic variability.

    Geographical scope of validation
    The methodologies were tested in diverse international geological scenarios, reinforcing their versatility. Pilot areas included:

    • Africa: Uganda, Democratic Republic of the Congo (Katanga), Nigeria.
    • Europe: France, Finland, and Russia.
    • Asia: Kazakhstan, Mongolia, and Turkey.
    • North America: Canada.

    This geographical reach demonstrates the global ambition of the project and ensures that the developed tools are applicable in different geological contexts.

    Funding
    The GEMA project has a total budget of €1,418,908, of which €1,206,071.80 corresponds to the contribution of the European Union (Gross Equivalent Subsidy). This support has been essential for the development of innovative artificial intelligence solutions and their application in geophysical exploration.

 

Related news

Contact us.

FILL THE FORM