A Logit-Based Binary Classifier of Tsunamigenic Earthquakes for the Northern Pacific Ocean

Abstract

In this study, logistic regression was used as a tool for deriving the binary classifier of tsunamigenic and nontsunamigenic earthquakes in the near-source zone. Earthquake source depth and moment magnitude were considered as predictors. The training dataset consisted of 767 M6.0+ submarine earthquakes, including 80 tsunamigenic and 687 nontsunamigenic events that occurred in the northern part of the Pacific Ocean from 1960 to 2020. The target area has already experienced significant and catastrophic tsunamis. The current analysis clearly showed that the data-driven logit model had a significantly lower false discovery rate relative to the threshold magnitude criteria that are widely used by tsunami warning agencies. At the same time, the balanced accuracy was about 71%, which suggests optimism for accurate tsunami forecasting based on the rapid interpretation of earthquake source parameters.