Utilizing deep learning-based feature engineering for effective geographical origin subdivision and classification of environmental soil samples in South Korea

초록

The integration of complementary geochemical analyses with deep learning techniques offers a powerful approach for determining the geographical origin of environmental samples at the national scale. This study presents methodologies that enhance the accuracy of geology-based geographic origin determination for soil samples across South Korea by integrating geochemical and geological data. It addresses the challenges of integrating Sr isotopes with multivariate geochemical variables while maintaining geological interpretability through the use of autoencoder-based deep learning algorithms, which enable efficient feature engineering for complex data analysis. Based on an analysis of 412 soil samples collected throughout South Korea, a geographic origin classification model was developed, establishing a novel framework for environmental sample provenance analysis. The analysis identified six distinct geographical origins in South Korea, each characterized by unique tectonic settings, bedrock ages, and lithologies. Broad regions underlain by granite bedrock were predominantly classified into the same origin, regardless of their tectonic context. These findings demonstrate the effectiveness of combining isotopic and geochemical data using advanced analytical techniques, significantly improving the accuracy and efficiency of origin tracing. The methodological advancements presented in this study have broad applicability across a variety of fields, including agriculture, forensic science, and archaeology.

키워드