GNSS-based velocity estimation using linear and machine learning approaches, with strain analysis in the Baltic Sea Region

In this paper, horizontal velocities are calculated using Linear Regression (LR) and Least Squares Support Vector Machines (LS-SVM) machine learning approaches to evaluate crustal deformation and tectonic stress models with the help of data provided by 42 GNSS stations along the Baltic coasts. Strain analysis for regional tectonic dynamics was performed with the help of estimated velocities based on daily GNSS observations processed in GIPSY-X software. The obtained velocity values showed statistical agreement between LR and LS-SVM at 40 stations, with LR providing lower standard deviations (±0.03–0.43 mm/year) and higher reliability for linear trends. Strain analysis reveals extensional stresses near stations MUS2, SUR4, PYRK and HAN1 due to crustal stress, while compressional stresses are observed around OSKL, KUN0, WARN and SAS2, which are probably affected by the Leba Ridge-Riga-Pskov Fault Zone. Although the optimized LS-SVM method via grid search and radial basis function kernels is advantageous for nonlinear data, it is considered more appropriate to use LR since it requires more computational resources. This study proposes the use of hybrid models (LR+LS-SVM) to capture complex deformation patterns and proves the effectiveness of LR for velocity estimation in tectonically stable regions. The findings not only provide important information for seismic hazard assessment and coastal management but also contribute to the understanding of the Baltic Sea geodynamics.