High-resolution emissions and concentrations of carbon monoxide and fine particle black carbon in Fort Collins, Colorado: development of a Bayesian uncertainty modeling and evaluation framework

Accurate, high-resolution data on air pollutant emissions and concentrations are needed to understand human exposures and health effects and to manage pollutant sources. Quantification of uncertainties is also needed. Bayesian approaches are promising for systematic uncertainty analysis that combines information from measurements and modeling. The work presented discusses an emissions inventory and concentration estimates for carbon monoxide (CO) and fine particle (PM2.5) black carbon for the city of Fort Collins, Colorado. The development of a Bayesian framework for updating estimates of emissions and concentrations (and exposures) in multiple stages, using measurement data, is also presented. The emissions inventory was constructed using the 2008 National Emissions Inventory (NEI). The spatial and temporal allocation methods from the Emission Modeling Clearinghouse data set are used to downscale the NEI data from annual and county-level resolution. Data from the Continuous Emissions Monitoring System (CEMS) for the years 2001-2010 were used to define the hourly temporal structure and variability in power plant emissions. Onroad mobile source emissions were estimated by combining a bottom-up emissions calculation approach (using emission factors and activities) for large roadway links within Fort Collins with a top-down spatial allocation approach for other roadways. Vehicle activity data for road links were obtained from local 2009 travel demand model results. The AERMOD Gaussian plume dispersion model was used to estimate air pollutant concentrations. Multiple years of available meteorological data are used to capture temporal variability in transport. A Bayesian stochastic ensemble approach is used to generate distributions of concentrations for each spatial location and time frame. Input data for ensemble members are sampled from distributions defined from the emissions inventory and meteorological data. Simulated distributions are compared with fixed-site monitoring data and personal concentration monitoring data collected along commute routes in Fort Collins. Posterior (updated) distributions of concentrations are calculated using likelihood functions derived from the measurement data. Results are presented on the detailed emissions inventory for Fort Collins, estimated to represent year 2013, with a focus on the temporal and spatial structure of emissions. Quantitative estimates of variability and uncertainty in dispersion model input data are highlighted and translated to parametric and non-parametric representations of data distributions, both for emissions and meteorology. Modeled concentrations of CO and PM2.5 black carbon at high spatial resolution (up to 100 m near commute routes) for representative daytime cycles (at hourly resolution) are presented and compared with measurement data. Results presented here will inform understanding of variability and uncertainty in emissions and concentrations (and ultimately exposures and health impacts). This work will also facilitate the development of a systematic approach for applying Bayesian updating to a real-world air pollution case application.