Reducing PM2.5 and ozone concentrations is important to protect human health and the environment. Chemical transport models, such as the Community Multiscale Air Quality (CMAQ) model, are valuable tools for exploring policy options for improving air quality but are computationally expensive. Here, we statistically fit an efficient polynomial function in a response surface model (pf-RSM) to CMAQ simulations over the eastern U.S. for January and July 2016. The pf-RSM predictions were evaluated using out-of-sample CMAQ simulations and used to examine the nonlinear response of air quality to emission changes. Predictions of the pf-RSM are in good agreement with the out-of-sample CMAQ simulations, with some exceptions for cases with anthropogenic emission reductions approaching 100%. NOx emission reductions were more effective for reducing PM2.5 and ozone concentrations than SO2, NH3, or traditional VOC emission reductions. NH3 emission reductions effectively reduced nitrate concentrations in January but increased secondary organic aerosol (SOA) concentrations in July. More work is needed on SOA formation under conditions of low NH3 emissions to verify the responses of SOA to NH3 emission changes predicted here. Overall, the pf-RSM performs well in the eastern U.S., but next-generation RSMs based on deep learning may be needed to meet the computational requirements of typical regulatory applications.
Predicting the Nonlinear Response of PM2.5 and Ozone to Precursor Emission Changes with a Response Surface Model
