Efficient clustering of GNSS stations for processing using double differences

The rapid growth of GNSS networks poses significant challenges for efficiently processing large datasets using double-difference techniques. In this study, we introduce a novel clustering algorithm, qmeans, which is based on bisecting k-means, to partition GNSS networks into smaller, manageable subnetworks or clusters for double-difference processing. We explore the trade-offs between cluster size, computational cost, and solution quality using a comprehensive dataset of approximately 1200 stations distributed across México, the United States, and Canada. Our results demonstrate that partitioning the network into clusters of 20–30 stations with 6 overlap stations between clusters can reduce processing time by ~ 20%, while larger clusters of 40–50 stations with 10 overlap stations slightly improve solution precision. We show that the number of shared stations between clusters impacts both the computational efficiency and the precision of the final solution, with higher counts leading to better precision but also increased processing time. The qmeans algorithm is integrated into the open-source Parallel.GAMIT software, offering a scalable, flexible solution that can be applied to large GNSS networks. Our work sets a foundation for selecting optimal subnetwork sizes based on specific needs of a GNSS processing project, enabling faster processing without significantly sacrificing solution quality.