Evaluation of high-resolution predictions of fine particulate matter and its composition in an urban area using PMCAMx-v2.0
Abstract. Accurately predicting urban PM2.5 concentrations and composition has proved challenging in the past, partially due to the resolution limitations of computationally intensive chemical transport models (CTMs). Increasing the resolution of PM2.5 predictions is highly desired to better inform air quality and emissions controls policies that protect public health, and also to address issues related to environmental justice. A nested grid approach using the CTM PMCAMx-v2.0 was used to predict PM2.5 at 36 × 36 km, 12 × 12 km, 4 × 4 km, and 1 × 1 km resolution for a domain largely consisting of Allegheny County and the city of Pittsburgh in southwestern Pennsylvania, US during February and July 2017. Performance of the model in reproducing PM2.5 concentrations and composition was evaluated at the finest scale using measurements from regulatory sites as well as a network of low-cost monitors. Total PM2.5 mass is reproduced well by the model during the winter period with low fractional error (0.3) and fractional bias (+0.05) when compared to regulatory measurements. Comparison with speciated measurements during this period identified small underpredictions of PM2.5 sulfate, elemental carbon (EC), and organic aerosol (OA) offset by a larger overprediction of PM2.5 nitrate (bias = +1.4 µg m-3, fractional bias = +0.81). In the summer period, total PM2.5 mass is underpredicted with fractional bias of -0.39. Here, PM2.5 nitrate is overpredicted again with a large fractional bias (+0.7) but significantly lower magnitude (+0.4 µg m-3). Underpredictions in PM2.5 sulfate and EC contribute to the negative prediction bias of total PM2.5 (-0.4 µg m-3 and -0.2 µg m-3, respectively), however the largest underprediction is seen for summer OA (bias = -1.9 µg m-3, fractional bias = -0.41). In the winter period, the model performs well reproducing the variability between urban measurements and rural measurements of local pollutants such as EC and OA. This effect is also captured well in the summer for EC, although the OA performance here is less consistent because much more of this OA is secondary and transported from outside of the inner modeling domain. Comparison with total PM2.5 concentration measurements from low-cost sensors yielded similar results with slightly higher overpredictions seen in the winter (fractional bias = +0.24) and lower underpredictions seen in the summer (fractional bias = -0.27). Inconsistencies in PM2.5 nitrate predictions in both periods are believed to be due to errors in partitioning between PM2.5 and PM10 modes and motivate improvements to the treatment of dust particles within the model. The underprediction of summer OA would likely be improved by updates to biogenic SOA chemistry within the model, which would result in an increase of long-range transport SOA seen in the inner modeling domain. These improvements are obvious topics for future work towards model improvement. Comparison with regulatory monitors showed that increasing resolution from 36 × 36 km to 1 × 1 km improved both fractional error and fractional bias by 0.04 in February 2017. In July 2017, fractional error decreased by 0.05 and fractional bias improved by 0.07 with increasing resolution. Improvements at all types of measurement locations indicated an improved ability of the model to reproduce urban-rural PM2.5 gradients at higher resolutions.
