Better prediction of surface ozone by a superensemble method using emission sensitivity runs in Japan

Abstract

Surface O 3 exerts adverse effects on human health and vegetation. To accurately predict the surface O 3 concentration, a superensemble was made, and its predictability was evaluated by comparison with observations collected in March and July 2014 in Japan. To produce a superensemble, five ensemble simulations consisting of a control run and four sensitivity runs were performed, with half and double the amount of anthropogenic and biogenic NO x and nonmethane volatile organic compounds (NMVOCs) emissions, respectively, in the same months of the previous three years. Then, a superensemble was made based on the five emission ensemble simulations to match the observations in the previous years by using the general linear least squares method. The superensemble trained with the previous years showed significant improvements in the predictability of the model in March and July 2014, especially in terms of the mean biases and root mean square errors, compared to the control run for the same period. In March, long-range transport influenced the enhancement of O 3, especially in the western part of Japan, while in July, the effects of the local photochemical production of O 3 were dominant. Thus, we obtained a better prediction for urban locations in July. • O 3 superensemble was made trained by observations in previous years and evaluated. • The superensemble was made by five members with changing precursor gas emissions. • It is low cost (single model with five tests), but the predictability was improved.