Customer story: Assessing the internal structure of a Tailings Storage Facility

Background

The safety of tailings storage facilities is arguably one of the most important missions of mining operations to ensure a sustainable extraction of critical raw materials. As the largest mining company in South America, Vale S.A. aims at exceeding the state of practice in geoscience to maximize safety and minimize risk.

In 2025, EMerald Geomodelling carried out a Proof of Concept (PoC) for Vale, using machine learning techniques to integrate existing geophysical data with geological-geotechnical surveys from drillings—including tactile-visual descriptions and sample values ​​from drilling holes at one of Vale S.A.’s tailing storage facilities. The main objective was to develop geological models that delimit interfaces between different tailings materials and lithologies and provide information on the associated geological uncertainties.

How it was done

For the first time, an extensive spectra of geophysical, geotechnical and geological investigations were integrated into EMerald’s machine learning algorithms to better utilize all the information available in the structure. In addition to helicopter geoscanning data, geophysical data from ERT, MASW, HVSR and GPR, together with (S)CPTu, Vane test and hydrogeological data (hydrogeological model and raw instrumentation data) were incorporated into the algorithm to develop models with high detail greater reliability.

Deliveries to the customer

This study employed two modeling workflows. The interface modelling was used to predict the depth of key geological boundaries: bottom of soft soil, phreatic surface (water table), bottom of tailings deposits, natural terrain (i.e., bottom of anthropogenic material), and bottom of residual soil.

The second workflow performed volumetric classification, where predictions were performed to produce models for grain size (fine, transitional, coarse) using the CPTU-derived Soil Behavior Type Index (Ic), and for contractive versus dilative behavior.

Customer results and actions taken

Critical parameters for save tailings management are their volume and topographic placement in the former natural terrain. Heritage tailings such as the investigated, do not always have a complete record of historical data and assumptions must be made that lead to increased uncertainty. The holistic model of pre-tailings natural ground gave Vale highly valuable insights that did not exist prior to the PoC.

With the positive outcome of the proof, work is now underway to start applying the EMerald Geomodelling^TM workflow on further tailing storage facilities to decrease risk and contribute to responsible management of these assets.

‍

“By providing uncertainty estimates and probabilistic outputs, the models provide a more robust foundation for risk-informed engineering decision making” Fernanda Matarazo, Geotechnical Projects, Vale S.A Brazil