Sampling in Precision Agriculture, Optimal Designs from Uncertain Models

If farmers are to manage the soil–crop system efficiently through variable application of fertilizers within fields they require information of the within-field variation of soil properties. To ensure that precision agriculture is cost-effective, soil sampling must be as efficient as possible. This chapter demonstrates the potential to optimize the design of soil sampling schemes if the variation of the target property is represented by a linear mixed model. If the parameters of the model are known prior to sampling we see that it is possible to optimize the sampling design with a numerical algorithm known as spatial simulated annealing. In general the parameters are unknown when the sample scheme is designed and the model is fitted to the observations. However it can be sufficient to assume a model which was fitted to a previous survey of the target variable over a similar landscape. When we do not have existing information about the variation of the target variable multi-phase adaptive sampling schemes may be used. We describe such a scheme for a survey of top-soil water content. The data are analysed as they are collected, and the sample design is modified to ensure that it is suitable for the particular target variable. The technologies described in this chapter represent the state of the art for sampling design in the geostatistical context. We discuss the developments required for them to be implemented as standard tools for precision agriculture.