You are here:
USE OF AUXILIARY DATA FOR SPATIAL INTERPOLATION OF OZONE EXPOSURE IN SOUTHEASTERN FORESTS
Citation:
Phillips, D., E. H. Lee, A. A. Herstrom, W. Hogsett, AND D. Tingey. USE OF AUXILIARY DATA FOR SPATIAL INTERPOLATION OF OZONE EXPOSURE IN SOUTHEASTERN FORESTS. Environmetrics 8(1):43-61, (1997).
Description:
In order to assess the impact of tropospheric ozone on forests, it is necessary to quantify ozone exposure on regional scales. Since ozone monitoring stations are widely scattered and mostly concentrate in urban and suburban areas, some form of spatial interpolation is necessary to estimate ambient air quality at non-monitored forested sites. This paper examines four different interpolation procedures for estimating ozone exposure as quantified by the SUM06 index (sum of hourly concentrations at or above 0.06 ppm) for July 1988 within the geographic range of loblolly pine (Pinus taeda L.) in the southeastern United States. The first three methods, inverse distance weighting (INVD), inverse distance squared weighting (INVD2), and ordinary kriging, are routinely used interpolators. In addition, we examined the use of cokriging, a geostatistical procedure which utilizes additional information from correlated auxiliary variables to aid in estimation, with a synthetic ozone exposure potential index (EPI). This index incorporated monthly data on anthropogenic emissions of nitrous oxides (NO-x), average daily maximum temperature, wind directional frequencies, and distance downwind from anthropogenic NO-x sources. The EPI index had a correlation coefficient of 0.626 with monitored SUM06 measurements. Cross-validation indicated that the accuracy and precision of the interpolators increased in the order: INVD lt INVD2 lt ordinary kriging lt cokriging. Cokriging with EPI exhibited 44-62 per cent lower mean error, 0-10 per cent lower mean absolute error, 6-14 per cent lower error standard deviation, and 4-10 per cent higher predicted verus observed r-2 values than the other three methods. These results show the potential for use of additional covariate data for prediction of ozone exposure in forested areas. However, increased monitoring of forested areas is needed to more adequately characterize forest exposure to ozone and its consequences.