Improvement of geoid accuracy using convolution neural networks and empirical rules

Résumé

Abstract We improve the accuracy of global geoid models (GGMs) by utilizing deep convolutional neural networks (CNNs) and empirical rules. The geoid heights obtained from the global geopotential models (GGM), EGM2008, and XG2019, are validated using geometrical geoid heights obtained from GNSS/leveling data. We used AlexNet, LeNet, and VGG16 CNN models to investigate the improvement of geoid accuracy and found that VGG16 model shows the most significant improvement among all the models. The VGG16 model improves the geoid accuracy of EGM2008 and XGM2019e by about 39 cm. Additionally, we apply empirical rules for outliers detection to eliminate unreliable GNSS/leveling data and found that this further improves the geoid accuracy by 13 cm for EGM2008 and 12 cm for XGM2019e. We also used a 7-parameter stochastic model to remove existing systematic errors, impacting 9 mm on AlexNet and VGG16.

Citer ce document

Accès au document

Ce document est hébergé sur son dépôt institutionnel d'origine.

Auteur(s)

Statistiques

Consultations : 1

Téléchargements : 0