Laukamp, Carsten, Beattie, Emma, and LeGras, Monica (2023) Modelling geochemical indices from hyperspectral drill core data from the Eucla Basin basement. Journal of Geochemical Exploration, 253. 107293.

Preview

PDF (Published Version) - Published Version Available under License Creative Commons Attribution Non-commercial No Derivatives. Download (18MB) | Preview

 

26

Abstract

Whole rock geochemical analysis of drill core is crucial for rock classification and understanding the evolution of a geological province. This becomes even more important when access to fresh rock samples is hindered by considerable overburden, such as in the case of the basement to the Eucla Basin sediments in Australia. A partial least squares (PLS) regression method was trialled to model geochemical indices, commonly applied for classification of igneous rocks, from hyperspectral visible-near infrared (VNIR; 350 to 1200 nm), shortwave infrared (SWIR; 1200 to 2500 nm) and thermal infrared (TIR; 6000 to 14,500 nm) reflectance spectra of three drill cores intersecting the basement of the Eucla Basin. Modelling of the Mg# (Mg/Mg + Fe) achieved a high correlation between the calculated and predicted values (i.e. R2 > 0.9) at an RMSE of about 10 %. The 2200 to 2400 nm wavelength region of the SWIR was key in modelling the Mg#, with amphibole and chlorite-related absorption features overlapping with the most important input bands in the PLS model. Modelling of the silicon, calcium, iron, magnesium index (SCFM; SiO2/(SiO2 + CaO + FeO + MgO)) required the TIR wavelength range, where diagnostic spectral signatures of silicates are located. PLS-based modelling of the SCFM achieved even a high correlation (i.e. R2 > 0.8) when the model was developed for one drill core but applied to another drill core that intersects the same lithologies. The third geochemical index trialled was the aluminium saturation index (ASI; Al2O3/(Na2O + K2O + CaO), producing high correlation between calculated and predicted ASI when using either, the VNIR-SWIR and the TIR wavelength ranges as input. Our study demonstrates how PLS regression can utilise hyperspectral data to determine geochemical indices, without the cost of full geochemical analysis of every sample in a wide array of rock types. The higher spatial resolution of the modelled geochemistry and the extrapolation of the modelling results to intervals that haven't been sampled for geochemistry supports stratigraphic correlation between drill cores. Furthermore, the combined analysis of geochemical and hyperspectrally-derived mineralogy helps mapping the intensity of weathering of basement as well as mineralogical and physicochemical gradients potentially related to hydrothermal systems.

Actions (Repository Staff Only)

Item Control Page