Abstract
Because the concentrations of atmospheric ozone (O3) are often overestimated by chemical transport model (CTM) simulations, a bias correction is useful when applying CTMs to evaluations of the environmental impacts of O3. In this study, we used the CTM results for the hourly O3 distribution over Japan during 2012, and we used the O3 observations in 2012 to evaluate three different methods of model bias correction. These consisted of two statistical methods and one machine learning (ML) method. Although all three methods significantly reduced the model bias in reproducing the observed hourly O3 concentrations, the ML method also improved the scores of the mean error and correlation coefficient. We also evaluated the model performance for the daily maximum 8-hour average (DMA8h) O3 and for the accumulated exposure to O3 above a threshold of 40 ppbv (AOT40). The ML method performed the best in both DMA8h and AOT40. Our findings suggest that the ML method could consistently reduce the model bias and error of the O3 indices, thus improving the evaluation of impacts of O3 on human health and crop yields.
