Dr. Angela Rodrigues - Honours Projects

Mapping retrograde alteration footprints for mineral exploration using Machine Learning in Broken Hill, Australia

Supervisors: Angela Rodrigues, Chris Voisey, Peter Betts
Support Offered: This is a lab-based project, with focus on ArcGIS Pro (licensing through Monash University) and Python based coding.  Assistance may be provided for high-resolution data purchases or for spectral ground-truthing of selected samples.
Partner organisation: Collaboration with CSIRO Mineral Resources Discovery Program is possible - TBC

The Broken Hill Block is a highly prospective Paleo Proterozoic province with complex hydrothermal and deformation history. Of particular interest is the high-grade metasomatic alteration followed by retrograde overprint, recorded in white mica chemistry. During the Olarian Orogeny (~1600 Ma), prograde granulite facies metamorphism caused white mica destruction, while retrograde hydrothermal upper-greenschist to amphibolite alteration during the Delamerian Orogeny (~530 Ma) originated structurally controlled white mica along faults and shear zones.
White mica (or sericite) is a common alteration mineral found in ore deposits and can be used to study ore formation. Its chemistry can be used as a proxy to ore deposits and associated alteration events due to the Tschermak Substitution, which causes a shift in the 2200nm absorption feature, due to a change in the proportions of Al:Si and Fe:Mg in the crystal lattice. The spectral gradients in this
absorption feature provide key insights on the alteration footprint, fluid pathways and vectoring information to mineralisation. This relationship has been demonstrated in the Broken Hill Block by Cudahy, 2016.
The major aim of this project is automated mapping of retrograde alteration in the Broken Hill Block using airborne hyperspectral data (either HyMap or PRISMA, depending on availability) using Machine Learning methods, with particular focus on white mica chemistry discrimination and implications for exploration. Several research avenues are possible depending on student’s interests: 1) Model
performance comparison between spectral feature engineering and raw spectral data as inputs; 2)
Build spatially and geologically aware models, by including available geological and structural data in the modelling stage; 3) prospectivity analysis, 4) transfer-learning studies.

Cudahy, T., 2016. Mineral Mapping for Exploration: An Australian Journey of Evolving Spectral Sensing Technologies and Industry Collaboration. Geosciences, 6, 52. DOI: 10.3390/geosciences6040052

Can airborne geophysical data improve geological mapping from spectral data? A multimodal Machine-Learning approach

Supervisors: Angela Rodrigues, Peter Betts
Support offered: This is a lab-based project, with focus on ArcGIS Pro (licensing through Monash University) and Python based coding. Assistance can be provided for high-resolution data purchases.
Partner organisation: Collaboration with CSIRO Mineral Resources Discovery Program is possible - TBC

Spectral data is an increasingly used tool for automated geological mapping and shows promising potential in discriminating different types of lithology and alteration, particularly when it comes to hyperspectral data (e.g., Jafari et al., 2026). This project’s main aim is to assess how geophysical data may affect automated lithological mapping from spectral data. Automated lithological mapping based on spectral data may be limited due to several factors, such as surface weathering, alteration and overlapping absorption features. Geophysical data, such as magnetic, gravity and radiometric data, may help resolve some of the limitations of spectral data by adding a physical layer of information to the models.
The project would involve the development of a manual benchmark lithological interpretation (from existing geological maps), and subsequent application of a supervised machine learning algorithm for automated lithological mapping (e.g., Extreme Gradient Boosting - XGBoost). At a minimum, two models would be generated and compared (hyperspectral only vs hyperspectral + geophysical data), and comparison should demonstrate where and why geophysical data add information to the model, using Explainable AI (XAI) techniques such as SHAP (SHapley Additive exPlanations). The study area can be adapted to student’s interests, although the geological exposure
at the surface should be agreeable with the bedrock geology.

Reference:
Jafari, L., Sanislav, I. V., Jarihani, B., Duce, S., & Koci, J. (2026). Machine Learning- and Remote
Sensing-Based Lithological Mapping Using VNIR + SWIR PRISMA Hyperspectral and ASTER
Multispectral Datasets in Northwest of Queensland. Minerals, 16(7), 720.
https://doi.org/10.3390/min16070720

Ask Anglea Rodrigues for further details