Sound sampling design is essential for the collection of data to support reliable scientific inference and decision making for management and policy. What counts as a sound design depends on the problem of interest, and the nature of the inference that is required. Many statistical analyses of spatial data, particularly for the prediction of local values, are done in a model-based context in which data are treated as realizations of an underlying random process. In this setting it is not necessary to use probability sampling to select sample locations, and there is scope to optimize sampling arrays computationally.

In this session we shall introduce some of the concepts that underlie the optimization of spatial sampling. These include methods to ensure good spatial coverage by a sample, methods to select an optimal grid spacing for geostatistical mapping and complex methods of combinatorial optimization to minimize the expected prediction error variance under certain assumptions about the underlying model. Participants in the course will be provided with scripts for the free R platform which will allow them to use the methods that are described to solve sampling optimization problems during the course and in their research afterwards.