Improving predictability of high-ozone episodes through dynamic boundary conditions, emission refresh and chemical data assimilation during the Long Island Sound Tropospheric Ozone Study (LISTOS) field campaign

Although air quality in the United States has improved remarkably in the past decades, ground-level ozone (O-3) often rises in exceedance of the national ambient air quality standard in nonattainment areas, including the Long Island Sound (LIS) and its surrounding areas. Accurate prediction of high-ozone episodes is needed to assist government agencies and the public in mitigating harmful effects of air pollution. In this study, we have developed a suite of potential forecast improvements, including dynamic boundary conditions, rapid emission refresh and chemical data assimilation, in a 3 km resolution Community Multiscale Air Quality (CMAQ) modeling system. The purpose is to evaluate and assess the effectiveness of these forecasting techniques, individually or in combination, in improving forecast guidance for two major air pollutants: surface O-3 and nitrogen dioxide (NO2 ).Experiments were conducted for a high O-3 episode (28-29 August 2018) during the Long Island Sound Tropospheric Ozone Study (LISTOS) field campaign, which provides abundant observations for evaluating model performance. The results show that these forecast system updates are useful in enhancing the capability of this 3 km forecasting model with varying effectiveness for different pollutants. For O-3 prediction, the most significant improvement comes from the dynamic boundary conditions derived from the NOAA operational forecast system, National Air Quality Forecast Capability (NAQFC), which increases the correlation coefficient (R) from 0.81 to 0.93 and reduces the root mean square error (RMSE) from 14.97 to 8.22 ppbv, compared to that with the static boundary conditions (BCs). The NO2 from all high-resolution simulations outperforms that from the operational 12 km NAQFC simulation, regardless of the BCs used, highlighting the importance of spatially resolved emission and meteorology inputs for the prediction of short-lived pollutants. The effectiveness of improved initial concentrations through optimal interpolation (OI) is shown to be high in urban areas with high emission density. The influence of OI adjustment, however, is maintained for a longer period in rural areas, where emissions and chemical transformation make a smaller contribution to the O-3 budget than that in high-emission areas. Following the assessment of individual updates, the forecasting system is configured with dynamic boundary conditions, optimal interpolation of initial concentrations and emission adjustment, to simulate a high-ozone episode during the 2018 LISTOS field campaign. The newly developed forecasting system significantly reduces the bias of surface NO2 prediction. When compared with the NASA Langley GeoCAPE Airborne Simulator (GCAS) vertical column density (VCD), this system is able to reproduce the NO2 VCD with a higher correlation (0.74), lower normalized mean bias (40 %) and normalized mean error (61 %) than NAQFC (0.57, 45% and 76 %, respectively). The 3 km system captures magnitude and timing of surface O-3 peaks and valleys better. In comparison with lidar, O-3 profile variability of the vertical O-3 is captured better by the new system (correlation coefficient of 0.71) than by NAQFC (correlation coefficient of 0.54). Although the experiments are limited to one pollution episode over the Long Island Sound, this study demonstrates feasible approaches to improve the predictability of high-O-3 episodes in contemporary urban environments.