Impact of offsets on GNSS time series stochastic noise properties and velocity estimation

Abstract

This study systematically investigates the efficiency of RMLE and MLE algorithms for noise model identification using various information criteria and examines the influence of offsets on noise properties and velocity estimation. Through experimental analysis on simulated GNSS time series, we confirm that RMLE and MLE accurately discriminate among four different noise models using AIC, BIC, and BIC_tp information criteria. For time series longer than 15 years, RMLE is recommended as the default method, while MLE outperforms RMLE for short time series when the underlying noise model is FN, PL, or GGM. RMLE proves more sensitive to the RW component than the MLE algorithm, particularly with shorter time series (less than 10 years). We also confirm that the addition of offsets does not transform non-GGM optimal noise models into GGM noise models, further supporting that GGM noise models in GNSS time series are not an artifact and that residual offsets are not the underlying cause of GNSS position time series exhibiting GGMWN characteristics. For selected real GNSS time series without offsets from Extended Solid Earth Science ESDR System, we further confirm that PLWN and FNWN remain the primary noise processes best describing GNSS time series. A noteworthy discovery is the observation of more RW components (approximately 19.22 %) than in previous research with 15-year long-term GNSS time series using the RMLE algorithm. Additionally, we still find GGM as the optimal noise model in the vertical component, ensuring it is real and not an artifact. Furthermore, even with artificially induced frequent offsets, we observe that only about 0.35 % of sites’ optimal noise models change from non-GGMWN models to GGMWN, consistent with the simulated GNSS time series experiments.