Robust statistical methods for the GPS network analisysis

The possible presence of outliers in observations requires a refined statistical analysis for the traditional, as wella as GPS networks. After a least square adjustment is performed, tradiional methods essentially consist in a standardised residual analysis. Tha so-called "preliminary diagnostic procedures" allow the identification of outliers and their removal. The "Baarda data snnoping", the "t test" and the "Danish method" are well known statistical methods commonly applied in Geodesy. In the last few years a class of alternative estimation strategies has been proposed, known as robust methods (Huber, Hampel, Rousseeuw). A major advantage in the use of such methods is that the parameters obtained are only slightly influenced by outliers in the observations. Two different applications of robust estimation are discussed in the paper; their application to a simulated GPS network and to a real GPS net for deformation control is presented.