Zahra Al Mualem

Lithium exploration from hard rock is critical to support global energy transition initiatives. Efficient prospection requires robust methods to identify lithium-enriched pegmatites, which can be mapped by delineating hydrothermal alteration zones (HAZ) that are indicative of concentrated presence of minerals [1]. Remote sensing data coupled with machine learning (ML) offers a significant potential to conduct a cost-effective, regional-scale, and eco-friendly exploration of lithium [2]. This study establishes a framework for evaluating ML classification algorithms by leveraging legacy surface geology maps as a benchmark for validation. We assess the performance of multiple ML algorithms including Random Forest (RF), Principal Component Analysis (PCA), Support Vector Machine (SVM), and Neural Networks (NN). We applied the methods to Landsat 8 & 9 data for mapping HAZ [3]. Algorithm inputs comprised systematically conditioned remote sensing derivatives (band ratios, spectral indices, and RGB composites) optimized for mineralogical discrimination. Validation utilized spatially explicit legacy geological data as ground-truth proxies. Our analysis quantifies key performance metrics (e.g., overall accuracy, precision, recall, Kappa) for each algorithm against geological maps. Results underscore that no single algorithm universally outperforms others across diverse geological settings. Instead, the concurrent application of multiple algorithms significantly enhances prospectivity mapping reliability. This validated approach leverages lithium exploration by exploiting legacy geological data and publicly available remote sensing data with the end product being a scalable methodology for prioritizing lithium exploration targets that will lead to informed and data-driven lithium reconnaissance in understudied desert terrains. 

---------------------------------------------------------------

References

[1] S. S. Alarifi, R. El‑Qassas, A. Omar, A. Al-Saleh, P. Andráš and A. Eldosouky, "Remote sensing and aeromagnetic mapping for unveiling mineralization potential: Nuqrah Area, Saudi Arabia," Springer, vol. 10, 2024. 
[2] H. Shirmard, E. Farahbakhsh, D. Müller and R. Chandra, "A review of machine learning in processing remote sensing data for mineral exploration," Elsevier, 2022. 
[3] O. O. Osinowo, A. Gomy and M. Isseini, "Mapping hydrothermal alteration mineral deposits from Landsat 8 satellite data in Pala, Mayo Kebbi Region, Southwestern Chad," Elsevier, vol. 11, 2021. 

Co-author/s:

Ahmad Ramdani, Petroleum Engineer, Saudi Aramco.

Taqi Al-Yousuf, Lead Geophysicist, Saudi Aramco.

Pavel Golikov, Geophysical Specialist, Saudi Aramco.