Spatial subsemble estimator for large geostatistical data

We introduce the concept of spatial subsemble, a subset ensemble estimation method useful in the analysis of large spatial random field datasets. The full dataset is sampled to give small spatially structured subsets of observations whose parameters are easily estimated; these are combined using a weighting scheme based on their cross-validation prediction ability. We show that our estimator is consistent. More importantly, we compare the spatial subsemble with competing alternatives and show that our proposed procedure is both accurate and much faster than its competitor. We illustrate the use of our method using several examples from large datasets.