A0369
Title: Methods for structural change detection in the trend function of random fields
Authors: Sheila Goerz - TU Dortmund University (Germany) [presenting]
Roland Fried - TU Dortmund University (Germany)
Abstract: Sudden changes in the structure of the data can occur not only in temporal data/time series but also in spatial and spatiotemporal data. In all cases, it is important that structural breaks are recognized reliably. Change point detection in spatial data is addressed. The already available methods for this type of problem are either not designed for continuous data, impose very strict assumptions or are computationally expensive. A method for detecting an arbitrary number of changes of any type is proposed in the mean of a time series that partitions the data into blocks and calculates the Ginis mean difference of the block means. This idea is extended to the detection of changes in spatial data such as satellite imagery or disease spread data. Gini's mean difference is not limited to other choices of change point statistics that are applied to the block means. The asymptotic behavior of suitable test statistics is investigated under the hypothesis of a constant mean when applied to independent spatial data. Simulation studies indicate that the tests work well not only for independent data but also for moving average-type random fields. In ongoing work, extensions to further kinds of random fields and more general spatial dependence structures are elaborated.
