Spatial Resolution Requirements for Traffic-Related Air Pollutant Exposure Evaluations

Abstract

Vehicle emissions represent one of the most important air pollution sources in most urban areas, and elevated concentrations of pollutants found near major roads have been associated with many adverse health impacts. To understand these impacts, exposure estimates should reflect the spatial and temporal patterns observed for traffic-related air pollutants. This paper evaluates the spatial resolution and zonal systems required to estimate accurately intraurban and near-road exposures of traffic-related air pollutants. The analyses use the detailed information assembled for a large (800 km²) area centered on Detroit, Michigan, USA. Concentrations of nitrogen oxides (NO_x) due to vehicle emissions were estimated using hourly traffic volumes and speeds on 9,700 links representing all but minor roads in the city, the MOVES2010 emission model, the RLINE dispersion model, local meteorological data, a temporal resolution of 1 hr, and spatial resolution as low as 10 m. Model estimates were joined with the corresponding shape files to estimate residential exposures for 700,000 individuals at property parcel, census block, census tract, and ZIP code levels. We evaluate joining methods, the spatial resolution needed to meet specific error criteria, and the extent of exposure misclassification. To portray traffic-related air pollutant exposure, raster or inverse distance-weighted interpolations are superior to nearest neighbor approaches, and interpolations between receptors and points of interest should not exceed about 40 m near major roads, and 100 m at larger distances. For census tracts and ZIP codes, average exposures are overestimated since few individuals live very near major roads, the range of concentrations is compressed, most exposures are misclassified, and high concentrations near roads are entirely omitted. While smaller zones improve performance considerably, even block-level data can misclassify many individuals. To estimate exposures and impacts of traffic-related pollutants accurately, data should be geocoded or estimated at the most-resolved spatial level; census tract and larger zones have little if any ability to represent intraurban variation in traffic-related air pollutant concentrations. These results are based on one of the most comprehensive intraurban modeling studies in the literature and results are robust. Recommendations address the value of dispersion models to portray spatial and temporal variation of air pollutants in epidemiology and other studies; techniques to improve accuracy and reduce the computational burden in urban scale modeling; the necessary spatial resolution for health surveillance, demographic, and pollution data; and the consequences of low resolution data in terms of exposure misclassification.

Keywords: Air pollution; Exposure; Exposure misclassification; Traffic.

Figure 1

Modeled road network for Detroit. Blue shaded areas shows city of Detroit and population density. Orange rectangle is region for high (10 m) resolution analysis. Locations of the three NO_x monitoring sites are indicated.

Figure 2

NO_x concentrations in μg/m³ around the I75/I94 junction for four scenarios: (a) monthly average; (b) maximum 24-hr average; (c) 98^th percentile 1-hr average, and (d) 24-hr average for Jan. 19, 2010. I75 runs roughly N-S; I94 runs SW-NE. Concentration scale shown in inset. Receptor grid uses 10 m spacing over a 1.0 ×1.2 km area.

Figure 3

Distribution of error measures for highest 24-hr average and estimates from 10 to 160 m from known values. Absolute concentration differences (ΔC) for (A) nearest-neighbor and (B) inverse-distance-weighted estimates. Relative concentration differences (RΔC) for (C) nearest-neighbor and (D) inverse distance-weighted estimates.

Figure 4

Distance stratified analysis of relative errors (concentration differences or RΔC) for inverse-distance weighted estimates and estimates from 10 to 160 m from known values. A. 95^th percentile errors. B. Percentage of receptors meeting 25% relative error criterion at given distance.

Figure 5. Maximum 24-hr NO_x concentrations in Detroit for four geographic units: A: parcel; B: blocks; C: tracts; D: ZIP code

Figure 6

Cumulative distribution of population exposed to maximum 24-hour NO_x concentrations estimated at parcel, block, census tract and ZIP code levels.