Using Unsupervised Clustering for Analyzing Airborne Gamma-Ray Spectrometry Data

Abstract

Unsupervised clustering analysis is becoming a useful tool in several fields, as can group similar entities together. Airborne gamma-ray spectrometry (AGRS) is widely used for geological mapping, mineral exploration and environmental studies, as it measures the natural radiation of potassium (K), uranium (eU) and thorium (eTh) in rocks and weathered materials. Ternary maps derived from K, eU and eTh provides an image of the three elements together, however, it's interpretation relies on the interpreter. In this work we apply unsupervised clustering on principal component analysis images derived from primary variables (potassium, uranium and thorium) to automatically map geological units in Mara Rosa- Goias, Brazil. We applied Gaussian Mixture Models (GMMs) as an unsupervised clustering method that automatically estimates the number of clusters in a dataset and was poorly explored in the context of geological mapping.