Identifying multivariate geochemical anomalies via tensor dictionary learning over spatial-elemental dimensionalities

Geochemical data is a typical geospatial-data with high-dimensional elemental properties. Existing methods for recognizing multivariate geochemical anomalies are limited in fully profiting from the inherent spatial-elemental structures of the geochemical data. This paper presents a novel method for the identification of multivariate geochemical anomalies, which leverages tensor dictionary learning over spatial-elemental dimensionalities to learn the joint representation of the spatial-elemental structures of geochemical data. The learned representations in spatial and elemental dictionaries are capable of capturing both the non-local features over space and the high-correlated distributed features across elements. The sparse representations over the spatial and elemental dimensions are exploited to reconstruct the geochemical background values, which permits us to identify multivariate geochemical anomalies considering the spatial-elemental structure of geochemical data. The proposed method is applied to recognize geochemical anomalies related to Au mineralization in the northwest Jiaodong Peninsula, Eastern China. By comparing with the anomalies recognized by deep autoencoder and local singularity analysis, we conclude that the tensor dictionary learning method is more effective to identify the mineralization-associated anomalies attribute to its ability to fuse spatial-elemental information of the geochemical data.