Optimal sampling designs for nonparametric estimation of spatial averages of random fields

Optimal designs of sampling spatial locations in estimating spatial averages of random fields are considered. The random field is assumed to have correlated values according to a covariance function. The quality of estimation is measured by the mean squared error. Simple nonparametric linear estimators along with sampling designs having a limiting density are considered. For a large class of locally isotropic random fields, we argue for the asymptotic optimality of simple linear estimators. The convergent rates of the mean squared error and optimal limiting densities of sampling designs are determined in every dimension. An example of simulation is given.