Uncertainty of spatial disaggregation procedures: Conditional autoregressive versus geostatistical models

Consider the problem of allocation of spatially correlated gridded data to finer spatial scale, conditionally on covariate information observable in a fine grid. Spatial dependence of the process can be captured with the conditional autoregressive structure, suitable for gridded (areal level) data. Also geostatistical methods, particularly empirical universal kriging, can be used for this purpose. In this study, we compare prediction results as well as prediction standard errors for two disaggregation procedures, based on the inventory of agricultural ammonia emissions reported in Pomeranian Voivodeship of Poland.