Interpreting Deepkriging for spatial interpolation in geostatistics

Fabian Leal-Villaseca, Edward Cripps, Mark Jessell, Mark Lindsay

Abstract: Highlights•Deepkriging, a spatial neural network, outperforms benchmarks in 3D interpolation.•The interpretability method enhances efficiency by focusing on meaningful groups.•The feature importance results align with domain knowledge.•Widely applicable methodology for interpolation problems in the geosciences.