Optimizing Data Assimilation Parameters for Improved CMAQ PM2.5 Estimates Over the United States to Inform Epidemiological Studies

PM_2.5 exposure estimates for United States at high spatial and temporal resolution are highly desirable to conduct further epidemiological studies focused on identifying more and less harmful types of particulate matter. Data assimilation via optimal interpolation (OI) was used to improve the Models-3 Community Multiscale Air Quality Model (CMAQ) modeling system estimates of aerosol pollution. Model predicted concentrations over North America for 2002 with and without assimilation of MODIS satellite-based aerosol optical depth are compared to PM_2.5 measurements for performance evaluation.

Our results show that OI can be done in three successful optimization approaches.  Each approach is characterized by slightly different sets of relative error statistics on the model and the satellite observations, and also the different choices of temporal and spatial aggregation in modeled AODs and OI parameters.   The first successful optimization scheme uses a spatially- and temporally-invariant settings and simple temporal (monthly) averaging of model and satellite data.  The second approach uses either customized OI settings for specific regions or time periods.  The third approach uses customized settings specific to time periods and geographical locations.  In all cases, we find it is critical to exclude from optimization MODIS data from months and regions where it tends to degrade rather than improve model skill.   Results from the first approach can improve average fractional bias from 1.31 to 0.59 (relative to IMPROVE PM_2.5 measurements) and from 0.34 to 0.22 (relative to STN PM_2.5 measurements).  The other two schemes have better performance, but at the expense of requiring more month- and region- specific settings.  The robustness of the technique has been applied by training the OI system on odd months of 2002 and testing settings on even months, and vice versa.