LUR models for particulate matters in the Taipei metropolis with high densities of roads and strong activities of industry, commerce and construction

Highlights

• The ESCAPE LUR modeling approach can be applied to the Taipei metropolis.

• Incorporating local variables relevant to PM emissions improve model performance.

• Road area is a good surrogate for traffic intensity data in the Taipei metropolis.

• PM_2.5 and PM_2.5 absorbance models yielded 96% and 95% of explained variability.

Abstract

Traffic intensity, length of road, and proximity to roads are the most common traffic indicators in the land use regression (LUR) models for particulate matter in ESCAPE study areas in Europe. This study explored what local variables can improve the performance of LUR models in an Asian metropolis with high densities of roads and strong activities of industry, commerce and construction. By following the ESCAPE procedure, we derived LUR models of PM_2.5, PM_2.5 absorbance, PM₁₀, and PM_coarse (PM_2.5–10) in Taipei. The overall annual average concentrations of PM_2.5, PM₁₀, and PM_coarse were 26.0 ± 5.6, 48.6 ± 5.9, and 23.3 ± 3.1 μg/m³, respectively, and the absorption coefficient of PM_2.5 was 2.0 ± 0.4 × 10^− 5 m^− 1. Our LUR models yielded R² values of 95%, 96%, 87%, and 65% for PM_2.5, PM_2.5 absorbance, PM₁₀, and PM_coarse, respectively. PM_2.5 levels were increased by local traffic variables, industrial, construction, and residential land-use variables and decreased by rivers; while PM_2.5 absorbance levels were increased by local traffic variables, industrial, and commercial land-use variables in the models. PM_coarse levels were increased by elevated highways. Road area explained more variance than road length by increasing the incremental value of 27% and 6% adjusted R² for PM_2.5 and PM₁₀ models, respectively. In the PM_2.5 absorbance model, road area and transportation facility explain 29% more variance than road length. In the PM_coarse model, industrial and new local variables instead of road length improved the incremental value of adjusted R² from 39% to 60%. We concluded that road area can better explain the spatial distribution of PM_2.5 and PM_2.5 absorbance concentrations than road length. By incorporating road area and other new local variables, the performance of each PM LUR model was improved. The results suggest that road area is a better indicator of traffic intensity rather than road length in a city with high density of road network and traffic.

Introduction

Land use regression (LUR) has been widely used and has rapidly become an important approach to predict long-term average pollutant concentration at an intra-urban scale. Recently, the European Study of Cohorts for Air Pollution Effects (ESCAPE) project described the development and performance of LUR models of 20 European study areas for particulate matter (PM) (Eeftens et al., 2012a). Based on the exposures estimated from the LUR models, this project illustrated an association between mortality and average annual exposure to fine particles (Beelen et al., 2014). Several studies have also developed LUR models for assessing intra-urban contrast of PM in North America (Henderson et al., 2007, Moore et al., 2007, Ross et al., 2007). The above-mentioned European and North American LUR models yielded a predictive capacity (as R²) ranging between 35% and 94% for PM_2.5, 39% and 97% for PM_2.5 absorbance (i.e., soot), 50% and 90% for PM10, and 32% and 81% for PM_coarse (PM_10–2.5). To estimate the concentrations of air pollutants at any point in the area, air pollution data and predictor variables are required. The selection of variables hence plays an important role in developing LUR models. Most study areas in Europe and North America, such as the Netherlands, Munich, London, Los Angeles, and New York City, share the same characteristics. However, cities in rapidly developing countries, such as Taipei, Taiwan, usually have different land use attributes from those in developed countries. One of such variable is the traffic-related predictor of road length, which was suitably used in several LUR studies (Hoek et al., 2008) as a good indicator of traffic intensity in European and North American cities, but may not be good enough in traffic-jammed cities like Taipei. Under such circumstances, road area can be an alternative to road length as a better land use variable for traffic intensity in LUR models in Asian cities with crowded roads. To our knowledge, no research has yet focused on the role of road area as a surrogate for traffic intensity when traffic flow data was not available. Elevated highways, which have been commonly constructed to increase road capacity across the city and soothe the traffic jams in city centers in Asian cities, were seldom considered in previous LUR models. Intra-city variation in air pollution can be better described by including elevated highways in the LUR models as one previous study reported that overpass structures could affect the distribution of air pollutants (Tong et al., 2011).

The continuous construction of buildings and infrastructures to meet residents' living demands is another common feature in rapidly urbanized cities. Previous LUR models have overlooked spatial variation in particulate air pollution arising from fugitive emissions of such construction activities, which are even more important in Asian cities. The mix of residential, commercial, and industrial areas in one small city district is not uncommon in Asian cities, where zoning policies are not as strictly enforced as in most European and North American cities. The individual contribution of residential, commercial, and industrial areas to spatial variation in PM needs to be adequately considered when they are included in LUR models.

Rivers and green lands in cities, by contrast, can alleviate air pollution in nearby areas by restricting or diluting pollution emissions. River, a component of the green land variable in the ESCAPE LUR models, can also be treated as an independent variable as it can cover a larger area in some urban environments and its air pollution dilution effects can be different, compared to urban green land.

This study aimed at developing LUR models by following the ESCAPE modeling procedures to characterize the spatial distribution of PM_2.5, PM_2.5 absorbance, PM₁₀, and PM_coarse. We focused on the role of specific variables that are important and are associated with emissions of PM in the Taipei metropolis but which have rarely been used in European and North American LUR studies. Further, we investigated whether the performance of PM LUR models in Taipei can be equivalent to those of European cities in the ESCAPE project by incorporating specific new variables. Results of this study could provide a higher spatial resolution of estimation of exposure to PM for future health studies in Taipei, Taiwan.