Daily ambient air pollution metrics for five cities: Evaluation of data-fusion-based estimates and uncertainties

Publication date: June 2017
Source:Atmospheric Environment, Volume 158
Author(s): Mariel D. Friberg, Ralph A. Kahn, Heather A. Holmes, Howard H. Chang, Stefanie Ebelt Sarnat, Paige E. Tolbert, Armistead G. Russell, James A. Mulholland
Spatiotemporal characterization of ambient air pollutant concentrations is increasingly relying on the combination of observations and air quality models to provide well-constrained, spatially and temporally complete pollutant concentration fields. Air quality models, in particular, are attractive, as they characterize the emissions, meteorological, and physiochemical process linkages explicitly while providing continuous spatial structure. However, such modeling is computationally intensive and has biases. The limitations of spatially sparse and temporally incomplete observations can be overcome by blending the data with estimates from a physically and chemically coherent model, driven by emissions and meteorological inputs. We recently developed a data fusion method that blends ambient ground observations and chemical-transport-modeled (CTM) data to estimate daily, spatially resolved pollutant concentrations and associated correlations. In this study, we assess the ability of the data fusion method to produce daily metrics (i.e., 1-hr max, 8-hr max, and 24-hr average) of ambient air pollution that capture spatiotemporal air pollution trends for 12 pollutants (CO, NO2, NOx, O3, SO2, PM10, PM2.5, and five PM2.5 components) across five metropolitan areas (Atlanta, Birmingham, Dallas, Pittsburgh, and St. Louis), from 2002 to 2008.Three sets of comparisons are performed: (1) the CTM concentrations are evaluated for each pollutant and metropolitan domain, (2) the data fusion concentrations are compared with the monitor data, (3) a comprehensive cross-validation analysis against observed data evaluates the quality of the data fusion model simulations across multiple metropolitan domains. The resulting daily spatial field estimates of air pollutant concentrations and uncertainties are not only consistent with observations, emissions, and meteorology, but substantially improve CTM-derived results for nearly all pollutants and all cities, with the exception of NO2 for Birmingham. The greatest improvements occur for O3 and PM2.5. Squared spatiotemporal correlation coefficients range between simulations and observations determined using cross-validation across all cities for air pollutants of secondary and mixed origins are R² = 0.88–0.93 (O3), 0.81–0.89 (SO4), 0.67–0.83 (PM2.5), 0.52–0.72 (NO3), 0.43–0.80 (NH4), 0.32–0.51 (OC), and 0.14–0.71 (PM10).Results for more spatially heterogeneous (larger spatial gradients) pollutants of primary origin (NOx, CO, SO2 and EC), tend to be better than those for relatively homogeneous pollutants of secondary origin. Generally, background concentrations and spatial concentration gradients reflect interurban airshed complexity and the effects of regional transport, whereas daily spatial pattern variability shows intra-urban consistency in the fused data. With sufficiently high CTM spatial resolution, traffic-related pollutants exhibit gradual concentration gradients that peak toward the urban centers. Ambient pollutant concentration uncertainty estimates for the fused data are both more accurate and smaller than those for either the observations or the model simulations alone.

Graphical abstract

http://ift.tt/2nabiTz