The rapid increase in the volume of geospatial data over recent years has added more challenges to processing these data using traditional methods. Thus,  geospatial applications have brought High-Performance Computing (HPC) into the mainstream and further increased its use in the spatial statistics field. ExaGeoStat is one example of an HPC software that enables large-scale parallel generation, modeling, and prediction of large geospatial data via covariance matrices. Unlike other existing tools, which typically rely on approximations to deal with the vast data volume on day-use machines, ExaGeoStat allows the processing of big geospatial data using modern HPC hardware in the exact mode without approximation. Therefore, ExaGeoStat makes it possible to fairly compare various approximation methods on synthetic datasets via large-scale simulations, not only to the true models but also to the exact solutions provided by ExaGeoStat. In 2021 and 2022, we hosted two competitions to assess the performance of existing methods/tools in the estimation and prediction of the given spatial and spatio-temporal datasets. Those datasets were synthetic datasets generated from specified statistical models, in particular, covariance models, using ExaGeoStat.

This year, we are hosting a third competition with different objectives and datasets. Instead of point estimation and prediction, the 2023 competition focuses on the construction of confidence and prediction intervals. It includes four sub-competitions, i.e., 1a, 1b, 2a, and 2b. The datasets were generated from stationary Gaussian random fields with an isotropic Matérn covariance function. There are several datasets with various designs of irregularly spaced locations. We provide two training dataset sizes, 90K and 900K, and two testing dataset sizes, 10K and 100K, to accommodate participants with different levels of computing resources.