Modern Geological Modeling: from Geostatistics to Machine Learning (and, of course, AI)

Geological modeling is central to understanding the subsurface—whether for resource exploration, groundwater management, geothermal energy, or assessing geological risks. These models aim to integrate diverse and often sparse data into coherent representations of subsurface structures and properties.

Traditionally, geostatistics has provided a powerful framework to create geological models and to handle spatial uncertainty, allowing interpolation and simulation of rock properties based on limited observations.
However, as geological settings grow in complexity and data volumes increase, classical approaches reach their limits. Recent advances in computational geometry offer new possibilities: they enable compact, low-dimensional representations of complex geological structures that remain interpretable and physically plausible. These representations are particularly well-suited for integration with geophysical inversion methods and allow for efficient probabilistic workflows with meaningful uncertainty quantification.

In this talk, we explore the evolution from traditional geostatistical techniques to modern approaches in geological modeling, including the use of machine learning and AI. We highlight how large-scale pre-trained models—such as foundation models and LLMs—can provide a path to include contextual understanding. Through examples from joint geological-geophysical modeling and inversion, we illustrate how these novel tools offer not only technical improvements but also a conceptual shift in the use of geological models as scientific tools to understand the subsurface.