1164602 - Understanding Biases in Emission Inventories with the Shair Real-Time Air Quality Modeling System in Richmond, California

The Ramboll Shair model was deployed as part of the Air Rangers II project by Groundwork Richmond with funding from a Community Air Grant (CAG) to assess air quality in Richmond, California. Richmond hosts the largest oil refinery in California, numerous chemical and industrial manufacturers, a regional solid waste facility, industrial recycling facilities, a major port, major railways, Interstate 580, Interstate 80, and the Richmond-San Rafael Bridge toll plaza. The Shair model provides source apportionment based on CAMx, a photochemical grid model with tagged emissions by source category. Mobile emissions are calculated in real-time for Shair; emissions from other source categories rely on historical emission inventory data, where the most similar day and time are matched. There is resulting uncertainty in the temporal accuracy of point, non-road and area-source emissions used in the Shair model, and consequently the estimated source contributions on a short-term basis. The Shair near real-time air quality modeling system, that integrates a dense network of air quality sensors for model adjustment, has been in operation in Richmond, California since December 2019.

We will present the findings from a January through December 2020 deployment of the Ramboll Shair air quality modeling platform. We will compare PM2.5 concentrations observed by the sensors with those modeled by Shair at the sensor locations and use the estimated source allocation information to evaluate systematic model-measurement bias introduced during periods of strong contribution from specific source categories. Nearly 55% of the average PM2.5 contribution is estimated to be from regional background pollution (i.e., secondary and/or upwind of Richmond) with very low spatial variability within the domain. Area sources contribute 21% of the average and have a higher spatial variation than does background. Traffic, non-road, and point sources contribute 10%, 9% and 9%, respectively. In terms of model-measurement bias, 9 sensor locations were found to be underpredicting the concentrations relative to the measured values indicating potential unknown sources or underestimates in emission inventories, and the other 30 locations were overestimating PM2.5 concentrations relative to the measured values. A comparison of the locations where the bias occurred and contribution of each of the sources indicated that entire source groups may be over- or under- predicted. This can help drive improvements with the regional emissions inventory, which will, in turn, result in a more effective emissions reduction plan for Richmond.