Can radiometric data improve lithology mapping and geological understanding through unsupervised classification?

Citation

Smillie Z, Demyanov V, Mckinley J & Cooper M (2020) Can radiometric data improve lithology mapping and geological understanding through unsupervised classification?. 13th International Conference on Geostatistics for Environmental Applications 2020, Parma, Italy, 01.07.2020-03.07.2020. https://2020.geoenvia.org/#proceedings-and-pictures

Abstract
Pattern classification algorithms can enhance the pattern recognition and prediction of large multivariate data sets, that otherwise would be difficult to detect. These techniques can be used to visualise the contribution or role of various features in shaping the patterns of a large data set. Self-organising map (SOM) is an unsupervised classification tool that is trained by competitive learning. The method is useful in analysing and visualising high-dimensional data, based on principles of vector quantification of similarities and clustering in a high-dimensional space. The method can be used to perform prediction, estimation, pattern recognition of large data sets. One main advantage of the SOM is that it can be applied to categorical and continuous variables making the tool ideal for analysing a complex combination of geological feature such as rock classifications, ages, geochemical composition, terrain elevations, etc. We here employ the tool to predict geological features using geophysical data, mainly the airborne geophysical data acquired through the Tellus project 2011/12. Tellus radiometric data present a high-resolution data set (Line spacing of 200 m and point spacing of 60 m). The data characterise the K, U and Th distribution associated with the natural geological features in Northern Ireland. The SOM of the radiometric data displayed patterns that are evidently associated with both bedrock and superficial geology. However, the addition of other natural features, such as terrain elevations modifies the clarity of the clusters and contribute to the prediction of geological formations. The SOM enhances the visualisation and recognition of the signals of geochemical variations within the bedrocks, although now concealed with superficial deposits. These advantages of SOM, combined with the high-resolution nature of the radiometric data input, presents an efficient tool to improve or complement conventional geological mapping techniques especially for “hard to recognise” stages of igneous rock emplacements, rock mass zonation and alteration/contact zones and also provides fundamental attempt toward understanding geological processes.