ABSTRACT
As with conditioned Latin hypercube sampling, spatial response surface sampling is an experimental design adapted for spatial surveys. With response surface sampling one assumes that some type of low order regression model can be used to accurately approximate the relationship between the study variable and the covariates. Sampling units are selected to implicitly optimise the estimation of the quadratic regression model. Spatial response surface sampling is illustrated with the electromagnetic measurements of the apparent electrical conductivity on the 80 ha Cotton Research Farm in Uzbekistan. The coordinates of the central composite response surface design points are multiplied by a factor so that a large proportion p of the bivariate standardised principal component scores of the population units is covered by the circle that passes through the design points. The model-based minimisation criterion is the average correlation of the sampling points. This criterion requires as input the parameters of a residual correlogram.
