Mon Aug 24, F. Flechtner
• Sorted by Date • Sorted by Last Name of First Author •
Feng, Yong, Chang, Guobin, Qian, Nijia, Huan, Yueyang, Cao, Yu, and Sun, Yinxiao, 2026. Probabilistic principal component analysis of time variable gravity data considering measurement errors' covariance matrix. Advances in Space Research, 78(2):527–546, doi:10.1016/j.asr.2026.04.008.
• from the NASA Astrophysics Data System • by the DOI System •
@ARTICLE{2026AdSpR..78..527F,
author = {{Feng}, Yong and {Chang}, Guobin and {Qian}, Nijia and {Huan}, Yueyang and {Cao}, Yu and {Sun}, Yinxiao},
title = "{Probabilistic principal component analysis of time variable gravity data considering measurement errors' covariance matrix}",
journal = {Advances in Space Research},
keywords = {GRACE, Principal component analysis, Covariance matrix, EM algorithm},
year = 2026,
month = jul,
volume = {78},
number = {2},
pages = {527-546},
abstract = "{Time-variable gravity data are often analyzed with non-parametric
decompositions such as principal component analysis. However,
for upstream products such as GRACE Level-2 solutions, the
observations can contain heterogeneous and correlated
measurement errors, whose correlations are reflected by a non-
diagonal error covariance matrix. Fully accounting for this
covariance information poses challenges to conventional PCA-type
processing. To address this issue, we propose a new
probabilistic PCA method that explicitly incorporates the
measurement-error covariance structure. The proposed model
decomposes the data into three parts: signals, noises, and
measurement errors. The signal term is further represented by a
small number of components that are orthogonal in both the
temporal and spatial domains, consistent with the
interpretability of many PCA-based approaches. The method is
formulated in a measurement-equation-based probabilistic
framework, and model parameters are estimated optimally in the
maximum-likelihood sense. An efficient expectation--maximization
(EM) algorithm is developed for iterative parameter estimation,
and a suboptimal variant is also tested experimentally. Using
GRACE Level-2 data as an example, the proposed approach
successfully suppresses stripe errors for signal reconstruction
and provides a compact set of orthogonal components for
spatiotemporal signal analysis. Experimental results demonstrate
that the proposed methods perform well in both signal
reconstruction and spatiotemporal characterization.}",
doi = {10.1016/j.asr.2026.04.008},
adsurl = {https://ui.adsabs.harvard.edu/abs/2026AdSpR..78..527F},
adsnote = {Provided by the SAO/NASA Astrophysics Data System}
}
Generated by
bib2html_grace.pl
(written by Patrick Riley
modified for this page by Volker Klemann) on
Mon Aug 24, 2026 17:14:25
GRACE-FO
Mon Aug 24, F. Flechtner