Hybrid Adjustment of Very-Long-Baseline GNSS Networks Using Robust Estimation with Residual Clustering

Global Navigation Satellite Systems (GNSSs) have become indispensable in establishing geodetic networks, owing to their capability to deliver high-accuracy positioning. However, the accuracy and reliability of GNSS networks can be compromised by measurement outliers and suboptimal adjustment techniques. This issue arises particularly in very-long-baseline GNSS networks, where some baseline solutions may appear suspicious. In such cases, applying the Constrained Least-Squares (CLS) technique in network adjustment does not guarantee accurate estimates, as its performance deteriorates significantly in the presence of suspicious baselines. This paper presents a novel algorithm for very-long-baseline GNSS network adjustment that enhances robustness and reliability through an integrated optimization framework. The proposed method combines robust M-estimation and baseline-wise residual clustering to achieve more reliable solutions under diverse observational conditions. Unlike the CLS approach, the proposed hybrid algorithm strengthens the overall resilience of the adjustment process against both isolated and grouped errors. Various GNSS network datasets representing different configurations are tested, with baseline lengths ranging from 227 km to 1980 km. The results demonstrate that the proposed hybrid algorithm outperforms the CLS technique in terms of geodetic accuracy and the distribution of observation residuals. It is shown that the hybrid technique can mitigate the effect of suspicious baseline solutions and provide improved residual stability. In addition, the geodetic accuracy improves by up to 71.4% and 90.0% in the horizontal and vertical components, respectively, at a single station, with average improvements of 22.1% and 22.9% among the stations showing improvement.