GNSS geodetic velocity prediction using ensemble tree models in Abu-Dabbab, Egypt

Abstract

Estimating Global Navigation Satellite System (GNSS) velocities is essential for understanding crustal deformation and motion. This work employs the Random Forest (RF) and Gradient Boosting Machines (GBM), two machine learning (ML) techniques, to estimate horizontal velocities at specific locations using GNSS data. Crustal deformation data were acquired through Global Positioning System (GPS) techniques, with positions of eleven stations determined from eight GPS measurement campaigns. Eighty percent of the GNSS velocity data from stations in the Abu-Dabbab region were used for training, while twenty percent were reserved for testing the models. RF demonstrated superior performance in estimating east geodetic GPS velocities with the lowest mean absolute error (MAE), while GBM excelled in predicting north geodetic GPS velocities, also achieving the lowest MAE. The maximum differences between model-predicted and reference velocities were 0.09 mm/year for RF and 0.1 mm/year for GBM, underscoring the precision of these methods. Despite data constraints the study confirms the efficacy of ML techniques, particularly RF and GBM, in providing accurate GNSS velocity estimates.