AAAR 31st Annual Conference
October 8-12, 2012
Hyatt Regency Minneapolis
Minneapolis, Minnesota, USA
Abstract View
A Bayesian – Based Ensemble Technique for Source Apportionment of PM2.5
SIVARAMAN BALACHANDRAN, Howard Chang, James Mulholland, Armistead Russell, Georgia Institute of Technology
Abstract Number: 522 Working Group: Source Apportionment
Abstract A Bayesian-based ensemble average of source apportionment results is used to develop new source profiles for use in a chemical mass balance (CMB) source apportionment (SA) of fine particulate matter (PM2.5). The approach estimates source impact uncertainties from a short-term application of four individual SA methods: four receptor-based models and one chemical transport model, the community multiscale air quality (CMAQ) model. The uncertainties for each method are used as weights in the ensemble-averaged source impacts.
The Bayesian based posterior distribution has a weakly informative prior distribution and treats the root mean square errors (RMSEs) between each method source impact and the ensemble average as the updated data. For each day of the short term application of the four SA methods, source impact uncertainties are sampled from the Bayesian-based posterior distribution. These uncertainties are used as weights to determine an ensemble average. Since the Bayesian analysis uses the RMSE, which requires knowledge of the ensemble average, iteration is required. All methods are treated equally to determine the initial ensemble average. The RMSEs from this first estimate of the average is used to determine a posterior distribution of SA method uncertainties, which are subsequently used to determine a new ensemble average. This process is repeated until the ensemble average converges. A Monte Carlo technique is used to estimate a distribution of Bayesian ensemble –based source impacts for each day in the ensemble.
These distributions of source impacts are then used to determine distributions of two seasonally-based source profiles. For each day in a long term PM2.5 data set, 10 source profiles are sampled from these distributions and used in a CMB application resulting in 10 SA results(for each day). This formulation results in a distribution of daily source impacts rather than a single value with an estimated uncertainty.