Lee, E.H. and W.E. Hogsett. 2001. Interpolation of temperature and non-urban ozone exposure at high spatial resolution over the western United States. Climate Research 18:163-179. WED-00-051
In order to assess the impact of natural and anthropogenic stresses on forest ecosystems, it is necessary to interpolate air temperature and tropospheric ozone (03) exposure values at high spatial resolution over complex terrain. The proposed interpolation approach was selected because of its ability to (1) account for the effect of elevation on temperature and their effects on tropospheric ozone, (2) use auxiliary data at higher spatial resolution than the variables of interest to improve the precision and accuracy of the prediction surfaces, (3) handle large amounts of data, and (4) provide not only a prediction at nonsampled locations but also a prediction standard deviation. The approach used auxiliary digital elevation model (DEM) data at 1 km resolution to improve the precision and resolution of the predictions for temperature and 03 exposure in the western United States. Initially, the study area was stratified into 03 regions based on seasonality and variability of monthly SUMO6 values at 111 stations for the period 1990-1992 using rotated principal component analysis. Monthly mean daily maximum air temperatures were spatially interpolated using loess nonparametric regression and kriging of the loess residuals and interpolated to 2 km grid points of a DEM and to the ambient air quality monitoring points. Monthly 03 exposures were spatially interpolated using loess fits to relate 03 levels to elevation, predicted temperature, and the geographic coordinates and interpolated to 2 km grid points of a DEM. The elevation-based spatial interpolation procedure produced accurate and precise temperature and 03 exposure surfaces which had desirable statistical properties and were logically consistent with local topographical features and atmospheric conditions known to influence 03 formation and transport. The leave-one-out cross-validation mean absolute error was 0.93EC for the monthly mean daily maximum temperature and 1.93 ppm-h for the monthly SUMO6 index for June 1990 for the western United States, comparable to published results for other regions at smaller spatial scales with less complex terrain.