Spatial data compression via adaptive dispersion clustering

In this article, we introduce a method of spatial data compression, which we call Adaptive Spatial Dispersion Clustering (ASDC). It is specifically designed to reduce the size of a spatial dataset in order to facilitate subsequent spatial prediction. Unlike with traditional data and image compression methods, the goal of ASDC is to create a new dataset that will be used as input into spatial prediction methods, such as traditional kriging or Fixed Rank Kriging, where using the full dataset may be computationally infeasible. ASDC can be classified as a lossy compression method and is based on spectral clustering. It aims to produce contiguous spatial clusters and to preserve the spatial correlation structure of the data so that the loss of predictive information is minimal. Through simulations, we demonstrate the predictive performance of these adaptively compressed datasets for several scenarios. ASDC is compared to two other data-reduction schemes, one using local neighborhoods and one using simple binning. We also present an application to remotely sensed sea-surface temperature data.