A Proposal of an Iterative DIA Testing Method for Multiple Outlier Identification

Outliers are the observations affected by unexpected errors identified in the observation model using a Detection-Identification-Adaptation (DIA) procedure. The commonly used data snooping method is well-established and recommended for a single outlier. However, iterative data snooping (IDS) has a serious weakness when multiple outliers appear. More precisely, IDS identifies outliers iteratively one-by-one and, therefore, a good observation once specified as an outlier can no longer be specified as a good one. A solution that addresses this weakness of IDS can be the so-called backward-forward strategy proposed in the past to identify unstable points in geodetic monitoring networks. After the backward step (final IDS iteration), points specified as unstable are iteratively one-by-one re-specified as stable in the forward step. In this study, we present another DIA testing procedure that addresses the discussed IDS weakness. The core idea is to iteratively increase the number of outliers specified in consecutive iterative steps. To evaluate the quality of the results obtained from the proposed method in relation to the conventional IDS and the backward-forward IDS, we conducted a single-point positioning pre-experiment under controlled conditions. The probabilities of all possible testing decisions under multiple outliers and the biases of final parameter estimators were computed for the three discussed iterative DIA testing procedures.