Abstract

This paper presents a novel and versatile framework for building ensemble spatial interpolation functions. As with all ensemble methods, the central idea is to assemble a voting scheme where a set of weak interpolation functions are combined, by using an aggregation function, to produce a strong ensemble response. In the presented scheme, voter interpolation functions are weak because they deal with a minimal portion of sample data extracted from spatial random partition elements, while the ensemble as a whole uses all the available information as much as possible. The random partitions scheme behaves as a bootstrapping strategy applied in a spatial context. Experiments show that the proposed framework has the promising ability to produce robust interpolation functions that can both scale to handle large sample data sets and deal with uncertainty quantifications, although weak voter interpolation functions are deterministic or highly data-consuming.