Semester of Graduation
Spring 2026
Degree
Master of Science in Civil Engineering (MSCE)
Department
Civil and Environmental Engineering
Document Type
Thesis
Abstract
Accurate orthometric heights are referenced to the geoid and essential in surveying, engineering, and the geosciences. Precise leveling used to determine them is impractical for large-scale or remote areas. Global Navigation Satellite System (GNSS) provides rapid and precise ellipsoidal heights. Converting the ellipsoidal heights to orthometric heights requires an accurate geoid model. Global Geopotential Models (GGMs) enable this conversion but exhibit discrepancies when compared with GNSS/leveling data. This study evaluated the accuracy of five ultra-high-degree GGM (XGM2019, GECO, EGM2008, SGG-UGM-2, and EIGEN6C4). The evaluation used 5,187 GNSS/leveling benchmarks across the United States (U.S.). The analysis was conducted at 5 degree intervals up to degree 2190. Residuals between GGM-derived and GNSS/leveling-derived geoid heights were analyzed to find the optimum GGM-degree that agrees more with the GNSS/leveling and further apply Machine Learning (ML) to minimize the discrepancies. The minimum RMSE values of the geoid undulations were consistently observed at degree 2140 for all models. XGM2019 achieved the lowest GGM RMSE of 62.75 cm while EGM2008 exhibited the highest RMSE of 63.31 cm. The geoid estimation accuracy were improved by integrating ML algorithms with the GGM predictions. Three ML techniques were applied at the same 5 degree interval for each GGM to degree 2190. The ML algorithms used are the Random Forest (RF), Support Vector Regression (SVR), and Multilayer Perceptron (MLP). After the ML improvement, MLP consistently outperformed SVR and RF, achieving the highest accuracy across all GGM. GECO produced the minimum MLP RMSE of 7.7 cm at d/o 2125. The findings reveal that ultra-high degrees do not add significant improvement nor guarantee better performance after certain degree. This suggests that lower degree truncations can achieve comparable accuracy with reduced computational cost. It also demonstrates the potential of ML techniques to enhance geoid estimation accuracy when integrated with GGM predictions.
Date
3-27-2026
Recommended Citation
Akutcha, John A., "Machine Learning-Enhanced Geoid Estimation Accuracy Using Ultra-High-Degree Global Geopotential Models and GNSS/Leveling Data" (2026). LSU Master's Theses. 6373.
https://repository.lsu.edu/gradschool_theses/6373
Committee Chair
Abdalla Ahmed
LSU Acknowledgement
1
LSU Accessibility Acknowledgment
1