The application of LiDAR and machine learning in managing historical mining features in Victoria, Australia

Authors: Herley, S; Riley, C; Tran, L; Divko, LG; Eid, R; Silver, E; Dang, L Download Paper Open access courtesy of:

DOI https://doi.org/10.36487/ACG_repo/2615_80

Cite As:
Herley, S, Riley, C, Tran, L, Divko, LG, Eid, R, Silver, E & Dang, L 2026, 'The application of LiDAR and machine learning in managing historical mining features in Victoria, Australia', in AB Fourie, G Boggs, J Heyes & M Tibbett (eds), Mine Closure 2026: Proceedings of the 19th International Conference on Mine Closure, Australian Centre for Geomechanics, Perth, pp. 1-13, https://doi.org/10.36487/ACG_repo/2615_80



Abstract:
Accurate mapping of historical mining features is critical for their management, providing essential information for a wide range of stakeholders – from local communities and land managers planning daytoday activities to emergency services responding to incidents. This study presents a large-scale machine- learning workflow for detecting and classifying previously unrecorded legacy mining features using Light Detection and Ranging (LiDAR) data and digital elevation models (DEMs). Two detection methods were developed to identify potential mining features: a sink detector which identifies hydrologic sinks in the DEM, and a noise detector which identifies clusters of noise-classified points in the LiDAR point cloud. Subsets of these detections were manually labelled by humans as ‘mine shaft’, ‘surface workings’ or ‘not sure/NA/other’ using visual inspection. These were used to train and test the machine learning image classification model. Statistical and spatial filtering techniques were then applied to refine the results. The method was applied as part of a broader program to establish a centralised and comprehensive database of mining features using available historical records. Its implementation has increased the number of recorded legacy mining features in Victoria’s historical database from approximately 36,000 to more than 250,000. These results demonstrate the value of using LiDAR data, DEM and machine learning to improve the identification and management of historical mining features, and provide a rapid, scalable approach for jurisdictions to better map and manage legacy mining features.

References: