Keywords:Tsunami deposits, machine learning
Correlation of tsunami deposits is very important to assess local tsunami size and frequency. In general, paleotsunami researchers conduct coring at several sites and correlate tsunami deposits in each core based on lithology and chronology. However, it is difficult to correlate tsunami deposits based only on lithology if the coring points are separated. Also, measuring numerous 14C age is costly. Herein, we propose a methodology to correlate tsunami deposits based on geochemical features together with machine learning technique. As geochemical proxy, we measured elements of tsunami deposits by using a core scanner that can conduct non-destructive and high-resolution XRF analysis. To correlate tsunami deposits based on elemental data quantitatively, we used SVM (support vector machine), machine learning technique for classification. We compare the results of correlations by machine learning and that by conventional approach based on lithology and chronology. As a result, we found that machine learning is useful to find the possible way of correlation while it is difficult to correlate tsunami deposits based only on machine learning. Therefore, machine learning is useful for correlation of tsunami deposits as a supplementary tool of other methods, and it is beneficial for us to decrease the number of samples for dating and to enhance reliability of correlation.