Empirical Bayes optimal fertilizer decisions

Abstract

Average fertilizer recommendations may be quite far from optimal for any given situation. Therefore, it is of interest to develop adaptive strategies, which base recommendations on plot-specific data, such as soil or plant measurements, or past yield trials for the site in question. We consider here the use of such data when fertilizer response is described by a random parameter model, since such models often provide a realistic description of fertilizer data. The objective function is taken to be the crop yield minus the fertilizer cost. When plot-specific data are available, it is shown that the optimal fertilizer strategy is obtained by maximizing the expectation of the objective function, conditional on the data. Simulations show the gain that can be obtained from using past yield measurements.