Deep Learning and Kriging

Kriging, a geostatistical method for optimal linear prediction, is widely used in various domains including mining, hydrology, and environmental and health sciences, as well as in computational experiments. This technique bears a close resemblance to kernel methods in machine learning. Despite the swift advancement of deep learning techniques, their application in spatial prediction remains relatively underexplored. In this context, we introduce a deep neural network-based approach for spatial prediction, designed to be resilient against variations in both the mean function and the kernel function.