Dr Ehsan Farahbakhsh

Ehsan develops data-driven methods for mineral exploration. He combines machine learning with geological, geophysical, geochemical and remote sensing data to predict where ore deposits are likely to occur, at scales ranging from regional maps and three-dimensional deposit models to spatio-temporal models that link mineralisation to plate tectonic and geodynamic evolution.

Much of this work has been carried out with industry partners, first through the PLATO and STELLAR projects at the University of Sydney and now through GEM-Cu at ANU, and is motivated by the need to find the critical minerals required for the energy transition. Ehsan welcomes enquiries from students and collaborators interested in data-driven mineral exploration.

Research overview

Spatio-temporal mineral prospectivity modelling

Most prospectivity models describe the present day. Spatio-temporal models instead ask when and where conditions favoured mineralisation through geological time, and whether the resulting deposits are likely to have survived deep-time landscape evolution (Tu et al., 2025). Ehsan uses machine learning to link plate tectonic reconstructions, subduction parameters and geodynamic models to dated deposits. He has applied this approach to porphyry systems in the New Guinea and Solomon Islands region (Farahbakhsh et al., 2025a), the Tethyan belt and globally, and to sediment-hosted deposits along craton edges (Shirmard et al., 2026). At ANU, he is extending this work through GEM-Cu, incorporating mantle convection models to connect mineralisation with deep-Earth processes.

Rotating globe showing the probability of porphyry copper formation along the global subduction network from 179 Ma to the present day. The animated field is the calibrated probability from a random forest classifier, estimating the likelihood that geodynamic conditions favourable to porphyry copper formation prevailed at each location and time. Because it carries no preservation information, it shows where deposits were likely to form rather than where they are likely to have survived.

Probability of porphyry mineralisation across the New Guinea region, evolving through time and culminating in the present-day map (Farahbakhsh et al., 2025a).

Three-dimensional mineral prospectivity modelling

At deposit scale, exploration decisions depend on depth as much as location. During his PhD, Ehsan developed a three-dimensional weights of evidence method that integrates geological, geophysical and geochemical data to target concealed deposits (Farahbakhsh et al., 2020a), and released it as the open-source 3DWofE package (Farahbakhsh et al., 2020b). He has also contributed to Bayesian approaches for fusing geological and geophysical data and quantifying uncertainty in 3D geological models (Olierook et al., 2021).

Two-dimensional mineral prospectivity modelling

Regional prospectivity mapping is often limited less by algorithms than by the small number of known deposits available for training. Ehsan develops machine learning frameworks for this setting, including generative adversarial networks with positive-unlabelled bagging and explainable semi-supervised deep learning (DEEP-SEAM). He has applied them to mafic–ultramafic intrusion-hosted nickel, cobalt and chromium in the Gawler Craton (Farahbakhsh et al., 2023), rare earth elements in the Curnamona Province (Luo et al., 2026), and lateritic nickel–cobalt (Wake et al., 2024) and porphyry copper–gold (Heidari et al., 2026) in the Lachlan Orogen.

Image

Probability of porphyry copper–gold mineralisation in the Lachlan Orogen, New South Wales (Heidari et al., 2026).

Geophysical and geochemical data processing for mineral exploration

Ehsan uses statistical and machine learning methods to extract exploration targets from geophysical and geochemical data. This includes unsupervised detection of mineralisation-related craton structures from geophysical data (Shirmard et al., 2025a) and full-waveform seismic tomography (Shirmard et al., 2025b), geochemical anomaly detection with methods ranging from fractal models and catchment-based approaches (Shokouh Saljoughi et al., 2018; Farahbakhsh et al., 2019) to deep anomaly detection networks (Luo et al., 2024), and estimating crustal thickness through time from geochemical data (Zhou et al., 2025).

Remote sensing for mineral exploration

Ehsan develops computer vision and deep learning methods for mapping geological features from satellite data, including hydrothermal alteration zones (Farahbakhsh et al., 2025b), tectonic lineaments (Farahbakhsh et al., 2020c) and lithological units (Shirmard et al., 2022a; Nagar et al., 2024), using multispectral data such as ASTER and Landsat as well as hyperspectral data. He has also reviewed how machine learning is used to process remote sensing data for mineral exploration (Shirmard et al., 2022b).