ELIMINATION OF SOME UNKNOWN PARAMETERS AND ITS EFFECT ON OUTLIER DETECTION

Abstract

Outliers in observation set badly affect all the estimated unknown parameters andresiduals, that is because outlier detection has a great importance for reliableestimation results. Tests for outliers (e.g. Baarda’s and Pope’s tests) are frequentlyused to detect outliers in geodetic applications. In order to reduce the computationaltime, sometimes elimination of some unknown parameters, which are not ofinterest, is performed. In this case, although the estimated unknown parameters andresiduals do not change, the cofactor matrix of the residuals and the redundancies ofthe observations change. In this study, the effects of the elimination of the unknownparameters on tests for outliers have been investigated. We have proved that theredundancies in initial functional model (IFM) are smaller than the ones in reducedfunctional model (RFM) where elimination is performed. To show this situation, ahorizontal control network was simulated and then many experiences wereperformed. According to simulation results, tests for outlier in IFM are more reliablethan the ones in RFM.