Geostats 2024

Sept. 2-6, 2024
Ponta Delgada - Azores, Portugal - University of the Azores

We're excited to announce that Geovariances will be sponsoring the 2024 edition of the Geostats Congress! Join us for insightful paper presentations, and don't miss our expert consultants' informative and interactive workshop. See you there!

Our workshop

Monday, September 2, from 2 pm to 7 pm:

Machine Learning applied to geosciences and mining

This workshop, tailored for geostatisticians, geologists, engineers, and resource estimation professionals in general, offers a holistic blend of theory and hands-on practice. Focusing on clustering, classification, and regression techniques, participants will engage in practical exercises utilizing Python programming and Isatis.neo software. These exercises centered around defining estimation domains and constructing classification and regression models will show real illustration cases.

The workshop will show the integration of Python and geostatistical software, showcasing the symbiotic relationship between programming and geostatistical analysis. Attendees will gain not only theoretical insights but also practical skills, which will enable them to implement machine-learning techniques in their own projects. By the end of the activity, we hope that the participants will understand the strategic role that machine learning can have in refining resource modeling.

👉 Be sure to register on the Geostats 2024 website.

Our paper and poster presentations

Join us for two insightful talks covering drill hole space optimization to reduce uncertainty in a iron ore deposit, and the simulation of realistic geomodels using Deep Generative Adversarial Networks.

💡Beyond drillhole spacing analysis: an integrated approach to drillhole spacing studies for uncertainty reduction in mining projects: a case study in a Brazilian iron ore deposit

Authors: G.C. Moreira (1), G.P. Scholze, (1), J.F.C.L Costa, (2), A.C. Endlein (1), R.M. Rolo (1), D.L. Usero, (1), D. Roldão (3), A.a Tente (3), A. Hübener (3)

(1) Geovariances – Datamine
(2) UFRGS, Brazil
(3) Vale, Brazil

Abstract:

Mineral resource projects depend highly on data acquisition, encompassing the entire project lifecycle, from exploration to post-mine closure. These data can include remote sensing, outcrop recognition, and samples from drilling, among other sources. The insights obtained through drilling represent the most straightforward method for comprehending the characteristics of subsurface mineral deposits, and decisions must be made to place drillholes efficiently. Traditionally, drilling strategies are defined by on-site geologists’ guidance based on experience and project requirements. However, the integration of mathematical models employing geostatistical methods has proven advantageous from numerous studies over the years.

Recently, these studies have gained popularity, mainly due to the increasing knowledge of the practitioners and the growing capabilities in processing and storage of personal computers. The outcomes of these studies can be quite sensitive to many parameters, including variograms, search strategies, and how the production scale is considered. Recognizing the absence of a one-size-fits-all solution, the paper emphasizes the significance of practitioner awareness of available options and parameter implications. Tailoring the approach to project maturity or substance type is also crucial.

This work addresses two important questions in a mineral resource project: (1) determining the ideal drilling spacing according to metrics such as converted resources or uncertainty levels and (2) where to place additional drillholes to minimize uncertainty, thus increasing knowledge about the deposit. An integrated methodology is proposed, using ordinary kriging, geostatistical simulation, and simulated annealing, which are applicable throughout different project phases, from exploration to grade control. A real iron ore deposit in Brazil serves as a case study, illustrating the practical implementation of the integrated approach. The findings help choose drillhole spacing and minimize uncertainty in mineral resource projects, providing valuable guidance for industry practitioners.

💡 Simulating structural geomodels with deep generative adversarial networks constrained by geological orientations

Authors: C. Garayt, N. Desassis, S. Blusseau, P.M. Gibert, J. Langanay, T. Romary
(1) Mines Mines Paris, PSL University, Paris, France
(2) Geovariances, Avon, France

Abstract:

This work uses deep generative adversarial networks (GANs) to simulate conditional structural geomodels, which are representations of geometric elements of geology. Geomodeling is an ill-posed problem as different geomodels can respect the same conditional data. One approach to address this problem is through geomodel simulations, which remain a challenge. Current methods, such as the potential field method based on geostatistics, struggle to both characterize all the uncertainties and produce realistic geomodels in complex geological settings.

This study proposes to use advanced GANs such as Wasserstein GAN (WGAN) to reproduce the unconditional spatial structure from a synthetic training dataset of realistic geomodels. The synthetic training dataset used is generated by Noddy using a set of geological parameters (including the number of domains, their thickness, the dip, and the folding parameters). Once trained, the GAN can simulate unconditional geomodels. This is the same idea as unconditional simulations in geostatistics. In a previous work, we presented techniques to handle hard known-domain data. This work’s main novelty is that it considers geological orientations, such as dip, to condition geomodels. To achieve this, the generator is trained to reproduce a discretized continuous function where isovalues define interfaces between geological domains. The gradient of this function, which can be computed through finite differences, provides access to geological orientations. The conditional step is performed in a Bayesian framework, where the GAN defines a prior distribution and conditional data defines the likelihood, inducing a posterior distribution. Samples from this posterior distribution are obtained using the Metropolis-adjusted Langevin algorithm, a Monte Carlo Markov Chain (MCMC) algorithm.

Finally, the proposed approach is compared with simulations of the potential fields method. This approach enables access to a wider range of geological scenarios allowing a better handling of uncertainties.

Geovariances sponsors Geostats 2024

12th International Geostatistics Congress

Visit the conference website →