We propose a hierarchical clustering methodology for Global Navigation Satellite System (GNSS) data that is applicable from local to global scales. We first adapted the conventional 2D velocity clustering metric for global-scale applications by implementing parallel translation in differential geometry. Then, we combined it with an Euler pole-based metric to incorporate the physical nature of plate motions, achieving advantages in identifying tectonic structures. This hybrid metric approach is examined through two case studies at different spatial scales to determine whether it can accurately identify tectonic plate and crustal block boundaries; one study utilizes global-scale data from the ITRF2008 plate motion model, and the other focuses on a local-scale study in Taiwan. Results obtained by using the hybrid metric consistently align better with geological data than those using either the 2D or Euler vector-based metrics alone. The proposed method is computationally efficient, enabling us to conduct two types of stability assessments: examining the robustness of clusters with synthetic noise contamination and performing leave-one-out analysis. These stability tests are demonstrated to be feasible within practical time frames. 

Key points