Unsupervised classification applications in enhancing lithological mapping and geological understanding - A case study from Northern Ireland

Abstract

Using pattern classification algorithms can help recognise and predict patterns in large and complex multivariate datasets. Utilising competitive learning, self-organising maps (SOMs) are known unsupervised classification tools that are considered very useful in pattern classification and recognition. This technique is based on the principles of vector quantification of similarities and clustering in a high-dimensional space, where the method can handle the analysis and visualization of high-dimensional data. The tool is ideal for analysing a complex combination of categorical and continuous spatial variables, with particular applications to geological features. In this paper, we employ the tool to predict geological features based on airborne geophysical data acquired through the Tellus project in Northern Ireland. SOMs are applied through 8 experiments (iterations), incorporating the radiometric data in combination with geological features, including elevation, slope angle, terrain ruggedness (TRI), and geochronology. The SOMs proved successful in differentiating contrasting bedrock geology, such as acidic versus mafic igneous rocks, while data clustering over intermediate rocks was not as apparent. The presence of a thick cover of glacial deposits in most of the study area presented a challenge in the data clustering, particularly over the intermediate igneous and sedimentary bedrock types.