Anomaly detection in geostatistical models with application to groundwater level data in the Gaza Coastal Aquifer

In geostatistics, the detection of anomalous observations has a particular importance because of the changes they can create in environmental and geological patterns. Few methods for detecting such observations in univariate data have been proposed for the spatial case, namely sample influence function (SIF), kriging, Intrinsic Random Functions (IRF), and geostatistical functional data. This article reviews the main outlier detection procedures in the context of geostatistics, and due to the absence of a numerical comparison between them, this article obtained the cut-off points of these methods for three different variogram models, and evaluated their performance via a simulation study. The results show that for all detection methods and the three considered models, there is an inverse relationship between the level of contamination and power of performance. In addition, the SIF for the cubic variogram model outperforms the exponential and Matérn. Because of the peculiarities of the Gaza Strip, as regards Palestine water condition, and for illustration purposes, we consider real groundwater level data in the Gaza Coastal Aquifer, where a set of possible outliers were identified.