Common mode errors (CMEs) widely exist in the coordinate time series of regional GNSS networks, severely affecting the estimation of deformation parameters and extraction of seasonal signals. In order to improve the accuracy of the time series, principal component analysis (PCA) is often used for spatiotemporal filtering. Due to various reasons, GNSS time series inevitably contain missing values. The traditional approach is to use interpolation methods to fill in the missing values before applying PCA filtering. However, interpolation methods are only suitable for time series with a low missing rate. In cases with a high missing rate, improper interpolation may introduce false information. Additionally, GNSS coordinate time series consist of unequal precision sequences, where the precision information is provided by international official organizations alongside the coordinate sequence. The classical PCA method does not consider the precision information of the time series. Based on the theory of best low-rank approximation, a team has proposed an extended PCA (EPCA) method that can directly handle incomplete time series with varying precision, without the need for prior data interpolation. Compared to their previously proposed modified PCA (JoG, 2014) and weighted PCA (RS, 2018) methods, the extended approach not only achieves better filtering performance (Figure 1) but also reduces computational time (Figure 2).

Fig. 1 The RMSEs of EPCA and MPCA for the equal weight and weighted cases

Fig.2 The ratio of MPCA to EPCA on computation times

The research findings were published in the prestigious journal "IEEE Transactions on Geoscience and Remote Sensing (TGRS)". The research work was funded by the National Natural Science Foundation of China (41974002, 42274005, 42192532).

Paper citation:

Kunpu Ji, Yunzhong Shen, Qiujie Chen, Tengfei Feng, (2023). Extended Principal Component Analysis for Spatiotemporal Filtering of Incomplete Heterogeneous GNSS Position Time Series. IEEE Transactions on Geoscience and Remote Sensing. Doi:10.1109/TGRS.2023.3277460.