A0559
Title: Contiguous segmentation of second-order non-stationary spatial processes
Authors: ShengLi Tzeng - National Chung Hsing University (Taiwan) [presenting]
Bo-Yu Chen - Purdue University (United States)
Hsin-Cheng Huang - Academia Sinica (Taiwan)
Abstract: Geostatistics, as a subfield of statistics analyzing and modelling spatial data, commonly assumes the process of interest to be stationary. Although this assumption is typically satisfied in small areas, it may not be appropriate for larger areas or more complex spatial phenomena. The nonstationary problems where the spatial covariance changes over the spatial domain are considered. To address this issue, a novel method for testing stationarity is proposed. Our method utilizes robust local estimates of spatial covariances to derive a test statistic. The data locations are then clustered using Voronoi tessellations. Furthermore, in cases where the stationary assumption is violated, a method is provided for identifying nonstationary features by partitioning the region into more homogeneous and close-to-stationary subregions. The optimal number of partitions can be determined using the Bayesian information criterion, ensuring our method is accurate and efficient. Notably, our proposed method applies to irregularly spaced data, making it a versatile tool for exploring a wide range of spatial data sets.
