The Effect of Covariance between Baseline Components on the Reliability of GNSS Networks: Results for a Highly Redundancy Network

Abstract

The most recent version of the reliability theory has been applied to describe the ability of a measurement system to detect, identify and remove outliers for a given probability level. However, the applications of that theory have been addressed to the simulated leveling networks. Here, however, we apply the theory within the context of satellite positioning-based network with real data. We have tested whether the covariances between the baseline components have effect or not on the reliability. We show that covariances increase the success rate in terms of outlier identification and, therefore, improves the reliability of the network. In comparison to a null covariance scenario, the minimal identifiable outlier for 80% of correct identification has decreased about ~30% for the ΔX and ΔY components, and ~14% for ΔZ component.  Futhermore, the increase of the significance level had the same proportion of improvement in both scenarios (null and non-null covariance). For high significance levels (α > 0,1) and measurements systems with good redundancy (ri > 0,5), however, the reliability of a stochastic model with null covariances is actually close to that of a full covariances scenario. In the absence of a more realistic stochastic model (e.g.: full covariance between baseline components) and systems with good local redundancy (ri > 0,5), one should choose high critical values ( k< 2,8).