Tutorial on Fixed Rank Kriging (FRK) of CO2 Data

In this document, we describe Fixed Rank Kriging (FRK), an approach to the analysis of very large spatial datasets. Such datasets now arise in many fields; our focus is on satellite measurements of CO2. FRK predictors and standard errors can be computed rapidly, even for datasets with a million or more observations. FRK relies on a so-called spatial random effects (SRE) model, which assumes that the process of interest can be expressed as a linear combi-nation of spatial basis functions, plus a fine-scale-variation component. Here, we describe in detail all steps involved in the analysis of a spatial dataset using FRK, we illustrate the steps using a synthetic dataset, and we provide Matlab code on an accompanying website