Efficient borehole targeting for ground-control of airborne electromagnetic (AEM) survey results. Issue 1 (1st December 2019)

• Record Type: Journal Article
• Authors: Haynes, Marcus; Ley-Cooper, Yusen
• Is Part Of: ASEG Extended Abstracts (Online). Volume 2019:Issue 1(2019)
• Page Start: 1
• Page End: 5
• Publication Date: 2019-12-01
• Subjects: machine learning -- airborne electromagnetic -- borehole assessment -- random forest -- AusAEM; Prospecting -- Geophysical methods -- Periodicals; Prospecting -- Geophysical methods; Periodicals
• DOI: 10.1080/22020586.2019.12073091
• Languages: English
• ISSNs: 2202-0586

Abstract

Summary: AusAEM is a Geoscience Australia program to collect broad-spaced (~20 km) airborne electromagnetic (AEM) data at the regional scale. The AusAEM data is being used to map the thickness and character of sedimentary and regolith cover across northern Australia. To maximise the utility of the collected data, it is important that subsequent interpretation can integrate the best available ground-control information. Typically such information is provided by boreholes. Prior to AEM data collection, we manually assessed boreholes according to a suite of metadata including spatial location information, depth, quality of lithological information, and the availability of geophysical wireline logging. These assessments are then used to deviate the planned AEM flight lines to intersect high-quality boreholes. However, this process proved prohibitive in the Pilbara; a mature mineral province with extensive drilling. Even after filtering for depth (>50 m), there are ~78, 000 mineral exploration boreholes in the current survey area. New methods are clearly required to enable the efficient prioritisation of borehole targets. To this end, we have used the results of previous manual borehole assessments to train and validate a machine learning algorithm for the purpose of identifying priority borehole targets. We find that a detailed manual assessment of boreholes can be closely replicated using a substantially reduced suite of borehole metadata. While the quality of assessment is mildly reduced, and there is a loss of borehole-specific information, the trade-off is a process that is ~137, 000 times faster. In practical terms, this has enabled a validated quality assessment to be conducted over an area of extensive drilling, a feat which would have proved prohibitive without machine learning.