Linking the air quality forecasting performance to meteorological and emissions conditions: Evaluation on a four-year practice in Southeastern United States

The high-resolution air quality forecasting system, Hi-Res, has been operationally serving the Atlanta metropolitan area since 2006. Hi-Res uses the Weather Research and Forecasting model (WRF, version 3.1) for forecasting of meteorology, the Sparse Matrix Operator Kernel Emissions model (SMOKE, version 2.3) for emissions, and the Community Multiscale Air Quality model (CMAQ, version 4.6) for chemistry and transport (Figure 1a). The emissions inventory used in Hi-Res as input to SMOKE is projected from a 2002 “typical year” inventory. Hi-Res nests its 4-km forecasting grid in a 12-km mother grid covering Georgia and portions of neighboring states and uses a 36-km outer grid over the eastern U.S. to provide air quality boundary conditions (Figure 1b). Hi-Res is run for two cycles each day: 00Z and 12Z cycles. For each cycle, Hi-Res first simulates a 66-hour period starting from 00Z or 12Z on the 36-km grid. WRF is initialized and constrained at the boundaries using 00Z or 12Z 84-hour forecast products from the North American Mesoscale (NAM) model while CMAQ is initialized from the previous forecasting cycle and uses “clean” boundary conditions for the 36-km grid (Figure 1c). Then, Hi-Res simulates the same 66-hour period on the 12-km grid and nests down to the 4-km grid for the last 51 hours. The simulations for each cycle take about 8 hours on 4 dedicated CPUs on a 64-bit platform server.

Figure 1. The Hi-Res air quality forecasting system: (a) System components, (b) Modeling domains (urban areas are shown in blue), and (c) Forecasting cycles.

 

Hi-Res air quality forecasts started as O₃ forecasts for the ozone season of 2006 and continue as O₃ and PM_2.5 forecasts since 2007. Forecast products are disseminated through the website http://forecast.ce.gatech.edu, as spatial distributions within the 4-km forecasting grid, and as temporal profiles at the locations of air quality monitoring sites, of not only O₃ and PM_2.5, but also air quality index (AQI) and winds, temperature, and precipitation. Here, we examine Hi-Res's air quality forecasting performance on O₃ and PM_2.5 for the past four years. Observational datasets of O₃ and PM_2.5, their precursor species, as well as PM_2.5 components, available from routine operating AIRS air quality monitors as well as from SEARCH, ASACA and STN networks, are utilized. We evaluate how well the forecasting system can capture the spatial and temporal variations of the two major air quality indexes. Further evaluation is conducted by examining the performance of forecasting meteorological variables, namely winds, temperatures and precipitations as integrated in the Hi-Res system. Correlation analysis is then carried out by linking O₃ and PM_2.5 performances to the impacting factors such as meteorological variables (weather conditions) and emissions conditions (normal days vs. wildfire days, holidays, major event days, etc.) Our goal is to characterize the strengths and weaknesses of Hi-Res's air quality forecasting ability in Southeastern United States.