The paper presents a spatio-temporal statistical model of agricultural yield prediction based on spatial mixtures of distributions. The proposed method combines several hierarchical and sequential Bayesian estimation procedures that allow the general problem to be addressed with a series of simpler tasks, providing the required flexibility of the model while decreasing the complexity associated with the large dimensionality of the spatial data sets. The data used for the study are 1970 - 2009 annual Iowa state county level corn yield data. The spatial correlation hypothesis is studied by comparing the alternative models using the posterior predictive criterion under squared loss function.