Mapping Air Pollutant Concentrations from Point Monitoring Data I: Declustering and Temporal Variance

This paper discusses statistical methods for mapping monitored air quality data. A key issue in mapping air pollutant concentrations is the proper accounting, or weighting, of high monitor station density urban areas and low station density rural areas. Improperly weighting of urban and rural monitors biases estimates in rural areas. The methods presented are designed to reduce this bias by incorporating two new estimation weights. The spatial configuration of monitoring stations is incorporated into a weight that "declusters " densely monitored areas by treating a spatial cluster of sites as one during the estimation. The second weight is based on the temporal variance of pollutant concentrations at a monitoring station and restricts the region of influence of a high variance monitoring sites. The new estimation methods are used to estimate summertime 90th percentile ozone concentrations over the eastern half of the U.S. for 1991-1995. Incorporating the declustering and temporal variance weights into the estimation restricts the spatial spreading of urban ozone concentrations into non-urban areas. Cross validation analysis is used to compare the new interpolation method with standard inverse distance squared interpolation and ordinary kriging. In densely monitored areas, the new method generates estimates that are similar to those generated by other methods. But, in sparsely monitored areas, the new method reduces the mean absolute estimation error by about 30 % compared with inverse distance and 10 % compared with ordinary kriging