GSBBO: a high-precision method for stress tensor inversion and its application at the great wall station in Antarctica

The current stress tensor inversion method based on the focal mechanism cannot solve problems such as the interference of too many outliers on the results and the slow speed and low accuracy caused by the excessive computation of the inversion process; therefore, we propose a new stress tensor inversion method, GSBBO (grid search, boxplot and Bayesian optimization), which combines machine learning algorithms to sieve out outlier data and improve the inversion speed and accuracy. The method first screens the focal mechanism data via a grid search and boxplot, and this process eliminates the bias of the outliers on the results. Then, to improve the speed and accuracy of the inversion results, the method further inverts the stress tensor by means of Bayesian optimization, which can obtain high-precision results quickly by means of screened datasets and machine learning algorithms. The GSBBO method is validated using artificially synthesized focal mechanism data containing random noise and outliers for three stress systems. The obtained results are compared with those of grid search and Bayesian optimization, and the GSBBO method is able to accurately identify the outliers and provide more accurate results quickly. Applying the method to the area of the Great Wall Station in Antarctica, the results show that the area experiences near-vertical compressive stress and strong northwest‒southeast extensional stress, which is consistent with the extensional stress in the area due to the subsidence of the Phoenix Plate. These findings indicate the continuing subsidence process of the Phoenix Plate in Antarctica.