Data-driven crustal deformation model in southeastern Tibetan Plateau using unsupervised machine learning techniques for GNSS observations

Abstract

The Southeastern Tibetan Plateau (STP) is a crucial place in understanding how stress is transmitted between the India-Eurasia collision belt and the surrounding blocks. It was accustomed to applying block models to study its deformation characteristics, which can hide or ignore some information due to the artificial division of the blocks. Therefore, we use an unsupervised machine learning method to analyze the tectonic deformation of the southeastern Tibetan Plateau without any prior information constraints. Firstly, we conducted K-means clustering analysis on GNSS velocity and identified the best cluster model using gap analysis, and we used Hierarchical Agglomerative Clustering to assess the reliability of this optimal cluster. Next, we used the method of spherical strain to obtain the translation rate, rotation rate, principal strain, surface strain, and second strain invariants for each cluster. After conducting cluster analysis and strain estimation, our results indicate that the oblique convergence of the Indian plate at the eastern Himalaya tectonic syntaxis is significantly involved in the deformation of the southeastern Tibetan Plateau. On the one hand, it slows down the crustal material migration caused by the gravitational potential energy in the Tibetan Plateau, resulting in the decrease of the rotation rate from the plateau to the middle of the STP. On the other hand, it encourages the lateral expansion of the STP, increasing the rotation rate from the central part of the STP to the Nantinghe fault and Menglian fault. This model has also confirmed the eastward growth mechanism of the Tibetan Plateau. It suggests that the oblique convergence of the Indian plate along the Eastern Himalayan tectonic syntaxis may be essential for the formation of regional tectonic deformation that has been ongoing since 20 million years ago.