Masked Autoregressive Flow for Geochemical Anomaly Detection with Application to Li–Cs–Ta Pegmatites Exploration of the Superior Craton, Canada

In mineral exploration, geochemical anomaly detection aims at identifying areas where geochemical properties differ from the surrounding areas, indicating possible mineralization. Robust outlier detection can help better identify potential anomalies. However, standard outlier detection techniques tend to work only in low-dimensional and Gaussian space, hence the need of a more robust outlier detection technique that can be used in the space of geochemical elements, which has high complexity and dimensionality. In this paper, a novel machine learning-based outlier detection technique is proposed. The masked autoregressive flow (MAF) was used to model the density of the high-dimensional geochemical space. Once successfully trained, the MAF provides a Gaussian space on which standard outlier detection techniques (here robust Mahalanobis distance) can be applied more successfully. The proposed method was applied to a high-quality lake sediment geochemical data acquired in Quebec, Canada, in an area with known Li–Cs–Ta (LCT) pegmatites. Results are very encouraging, with the detection of many of the known occurrences of LCT pegmatites and the discovery of potential new targets for further exploration. Hence, the method described here can be used to explore for LCT pegmatites.