Smoothing Spatial Data by Estimating Mean Local Variance

摘要

A nearest neighbor nonparametric regression method is used to estimate air pollution levels at other than measured points. The method requires an appropriate smoother. Cross-validation is used to determine the appropriate smoother. An alternative method is introduced to determine an appropriate level of smoothing which involves minimizing mean local variance. Mean local variance is a function of the size of a circular window. It is minimized for two pollutants in Ohio, New York and Florida. The smoother obtained by cross-validation using Ohio's data is compared to that obtained by minimizing mean local variance.