Robustness Analysis


Described in this report are the results of the investigations undertaken by two collaborating research groups at the University of New Brunswick and The University of Calgary under a DSS research contract #23244-9-4198/01-SS for the Geodetic Survey Division of the Canada Center for Surveying. The investigations addressed the problem of geodetic network analysis techniques, and proposed alternatives to the standard statistical analysis techniques designed to analyse network sensitivity to gross errors and blunders. The original aim of the investigations was to study the differences between and merits of two such alternative techniques; the reliability technique, introduced by Baarda and implemented by The University of Calgary group. It was discovered at the beginning of the investigation that these two techniques are very much complementary: that is, the weakness of each is in the area of the strength of the other. It was decided this to combine the two techniques into one, which we call "robustness analysis". Experiments with both simulated and real networks have shown that robustness analysis works very well in depicting the strong and the weak points in the network must be studied in the sense of scales, shear, or local twist, which have to be judged in three independent senses. The strength/weakness of a network must be studied in the sens of scales, shear, or local twist, each of which provides a different picture of strength. These three indicators (primitives) cannot be combined into a single scalar indicator. It has been concluded that robustness analysis should be carried out side-by-side with the standard statistical analysis form which it differs fundamentally. It is recommended that the Canadian federal specifications for horizontal geodetic networks be amended to include pertinent prescriptions as far as desired robustness is concerned, i.e., specific robustness to be achieved through meeting robustness tolerance limits.