Resolving ozone vertical gradients in air quality models

Models severely overestimate surface ozone in the Southeast US during summertime and this overestimation has implications for the design of air quality regulations. We use the GEOS-Chem model to interpret ozone observations from aircraft (SEACRS), ozonesondes (SEACIONS), and surface sites (CASTNET) in August-September 2013. After correcting for a 30-50 % NOx emission overestimate in the US EPA National Emission Inventory, we find that the model is unbiased relative to aircraft observations below 1 km. However, surface observations of maximum daily 8-h average (MDA8) ozone are still biased high in the model (averaging 48 ± 9 ppb) compared to observations (40 ± 9 ppb). The low tail in the observations (MDA8 ozone < 25 ppb) is associated with rain and is not captured by the model. The model bias decreases by 3 ppb when accounting for the subgrid vertical gradient between the lowest model level (centered 60 m above ground) and the measurement altitude (10 m). The model underestimates low cloud cover, but this underestimate is insufficient to explain the remaining surface ozone bias because the response of model ozone to cloud cover is weaker than observed. Midday ozonesondes at Huntsville, Alabama show mean decreases in ozone from 1 km to the surface of 4 ppb under clear-sky and 7 ppb under low cloud, whereas the model decreases by only 1 ppb under both conditions. By contrast, potential temperature below 1 km is well-mixed in both the observations and the model. The observations thus imply a strong asymmetry between top-down and bottom-up mixing that is missing from GEOSChem and appears to be insufficiently represented in current air quality models. A sensitivity simulation reducing top-down eddy diffusion and removing top-down non-local vertical transport of ozone can reproduce the observed ozone gradients in the mixed layer.