Assessment of hydrological loading displacement from GNSS and GRACE data using deep learning algorithms.

Area of Science:

• Geodesy
• Earth Science
• Machine Learning

Background:

• Terrestrial water storage variations cause significant crustal deformation.
• Accurate estimation of hydrological loading displacement is crucial for precise geodetic observations and reference frame maintenance.
• Conventional methods like load Green's function inversion have limitations in precision.

Purpose of the Study:

• To introduce a novel 3D Convolutional Neural Network (3D-CNN) method for estimating hydrological loading displacement.
• To compare the precision of the 3D-CNN method against conventional techniques.
• To analyze the spatiotemporal characteristics of terrestrial water storage and loading displacement in Yunnan Province.

Main Methods:

• Utilized vertical displacement time series data from Global Navigation Satellite System (GNSS) stations.
• Integrated spatiotemporal variations in terrestrial water storage from Gravity Recovery and Climate Experiment (GRACE) satellites.
• Applied a 3D Convolutional Neural Network (3D-CNN) model for displacement estimation.

Main Results:

• The 3D-CNN method demonstrated markedly higher inversion precision than conventional load Green's function inversion.
• Significant reductions in deviations were observed: max deviation decreased by 1.34 mm, absolute minimum by 1.47 mm, absolute mean by 79.6%, and standard deviation by 31.4%.
• Analysis revealed dominant annual and semi-annual cycles in terrestrial water storage and loading displacement, accounting for over 90% of the variance.

Conclusions:

• The 3D-CNN approach offers a novel and more precise method for estimating terrestrial water loading displacement (TWLD).
• TWLD exhibits significant spatial heterogeneity, strongly correlated with regional precipitation patterns.
• The integrated GRACE-GNSS TWLD model provides valuable data for high-precision terrestrial water storage inversion and geodetic applications.