Complementary co-kriging: spatial prediction using data combined from several environmental monitoring networks

We consider the problem of optimal spatial prediction of an environmental variable using data from more than one sampling network. A model incorporating spatial dependence and measurement errors with network-specific biases and variances serves as the basis for the analysis of the combined data from all networks. We develop the associated optimal prediction methodology, which we call complementary co-kriging because (a) data from each network complements the other, and (b) the solutions to several prediction problems of interest are co-kriging predictors. A hypothetical example illustrates how much better the complementary co-kriging predictor can be, when compared to the ordinary kriging predictors from each network alone and to a ‘naive’ combined predictor. We use the methodology to obtain optimal predictions of wet nitrate concentration data over the eastern U.S. using data combined from the National Atmospheric Deposition Program/National Trends Network (NADP/NTN) and the Clean Air Status and Trends Network (CASTNet).