On the detection of outliers and the determination of reliability in geodetic networks
The detection and removal of all gross errors or local systematic effects from geodetic data is of paramount importance when the quality of a Least Squares Estimation is concerned. In recent years, the above need, as well as the decrease of the computational cost compared to the cost of observational procedures, has led to the development of a number of sophisticated techniques for the systematic tackling of the problem, based on statistical tests. Several approaches are reviewed in this study, but the main weight is given to the most systematic and effective up to now – the post-adjustment techniques. The use of these techniques, and the analysis and comparison of their philosophies and sensitivities, are illustrated by simple numerical examples. A systematic strategy for error detection and elimination is proposed with special emphasis on survey networks. The finite sensitivity of the employed techniques may leave undetectable outliers in the model. Their magnitude as well as their undesirable outliers in the model. Their magnitude as well as their undesirable effect on the final solution are assessed by the concepts of internal and external reliabilities. Problems encountered with the detection of small gross errors and the resistance of networks to distortions cause by the presence of inconsistent observations, are also illustrated by analyzing the strength of a real geodetic network. Certain limitations in the classical approaches have led to the study of alternatives which are more robust to outliers than is the Least Squares Method, an overview of which is also given.