National-scale exposure prediction for long-term concentrations of particulate matter and nitrogen dioxide in South Korea☆
Graphical abstract
Introduction
There has been increasing evidence supporting the association between long-term exposure to air pollution and human health (Pope and Dockery, 2006, Hoek et al., 2013). Most cohort studies of long-term air pollution and health have particularly focused on traffic-related air pollutants such as particulate matter less than or equal to 10 or 2.5 μm in diameter (PM10 or PM2.5), nitric oxide (NO), and nitrogen dioxide (NO2) (Beelen et al., 2014, Kaufman et al., 2016, Laden et al., 2006, Pope et al., 2004).
One of the most important challenges faced by these studies has been the unavailability of individual-level air pollution measurements. Measurements are only available at limited numbers of monitoring sites. To overcome this limitation, many cohort studies have adopted exposure prediction models based on statistical approaches and regulatory air quality monitoring data, and estimated air pollution concentrations at participants’ homes. The most common statistical modeling approaches employed have been regression methods, namely, land use regression and geostatistical techniques such as kriging (Hoek et al., 2008, Jerrett et al., 2005). Land use regression explains the spatial variability of air pollution using predictors, and kriging additionally incorporates spatial correlation structures. These models often considered hundreds of geographic variables, computed in the Geographic Information System (GIS), as predictors representing local or regional pollution sources. Either a limited number of selected variables or a few summary predictors estimated by dimension reduction were included (Eeftens et al., 2012, Sampson et al., 2011). In addition to geographic variables, other studies improved the prediction models by including satellite imagery data or air quality model outputs as predictors in the models (Kloog et al., 2011, Lindstrom et al., 2013, van Donkelaar et al., 2015 Young et al., 2016).
Although these exposure prediction models have mostly been developed for city areas (Eeftens et al., 2012, Hoek et al., 2008; Keller et al., 2015), some studies have expanded the modeling domains to a national scale while retaining the high spatial resolution (Hystad et al., 2011, Knibbs et al., 2014 Novotny et al., 2011; Sampson et al., 2013, Vienneau et al., 2010). There are large areas without regulatory air quality monitoring sites in the vicinity, which hamper the assessment of air pollution concentrations that the people residing in such areas are exposed to. This limitation also prevents resulting health analyses using the existing national-scale health data from large cohorts or government-generated databases. Developing national exposure prediction models would support exposure assessments and health analyses on a national scale to provide important policy-related information, such as national patterns of air pollution and health risks and areas that need to be targeted for air pollution reduction and that contain vulnerable populations.
Most national exposure prediction models have been developed in North America and Western Europe (Hystad et al., 2011; Novotny et al., 2011; Sampson et al., 2013, Vienneau et al., 2010). National prediction models that operate in different areas can provide insights into the differences and similarities between countries. The aim of this study was to develop a national-scale exposure prediction approach with fine-scale spatial variability using universal kriging and dimension-reduced predictors for estimating annual average concentrations of PM10 and NO2 during 2010 in South Korea. We selected the year 2010 for our study, as there were sufficient numbers of monitoring sites available and a quinquennial population and housing census, a data source of a large portion of geographic variables, was conducted this year (Yi et al., 2016). To gain insights into the performance of our prediction models, we compared our primary approach with two alternative exposure prediction approaches. Using our exposure prediction models, we also investigated the distributions of the predicted PM10 and NO2 concentrations across residential census output areas to assess population exposure levels.
Section snippets
Regulatory air quality monitoring data
We obtained hourly PM10 and NO2 concentrations measured at the 294 regulatory air quality monitoring sites operating in 2010 in South Korea (48 million people in 100,033 km2) from the National Institute of Environmental Research (Fig. 1). The South Korean regulatory air quality monitoring network consists of four types of monitoring site: urban background, urban roadside, regional background, and national background (MOE, 2011). Urban background sites were established for monitoring air
Regulatory monitoring data
Out of the 277 regulatory monitoring sites inlcuded in our models, about 40% of the sites were located in seven metropolitan cities (Table 1). The median distance between the closest regulatory monitoring sites was 3.14 km, with shorter distances in metropolitan cities (median = 2.85 km) than in provinces (3.99 km).
The mean annual average concentrations of PM10 and NO2 for 2010 across 277 regulatory monitoring sites in South Korea were 51.63 μg/m3 (SD = 8.58) and 25.64 ppb (11.05), respectively
Discussion
To the best of our knowledge, this is the first study to develop a national-scale exposure prediction approach that estimates annual average air pollution concentrations at individual residential locations in South Korea. Our national prediction approach based on universal kriging and PLS regression gave modest and good model performances for PM10 and NO2, respectively. The PLS predictors of PM10 and NO2 concentrations were positively correlated with geographic variables representing traffic,
Conclusions
Assessment of air pollution concentrations in South Korea, with its heavily urban-focused population exposed to high air pollution levels, could provide new scientific knowledge regarding exposure prediction and health analyses. Our national-scale exposure prediction models for estimating individual-level concentrations of PM10 and NO2 is the first such study carried out for South Korea, which will allow national-scale epidemiological studies to assess national patterns of risks attributable to
Funding
This research was primarily supported by the Basic Science Research Program through National Foundation of Korea (NRF) funded by the Ministry of Education (2013R1A6A3A04059017); and additional support was provided by the Environmental Health Action Program funded by the Korea Ministry of Environment (MOE) (2014001360003, 2014001360002).
References (35)
- et al.
Effects of long-term exposure to air pollution on natural-cause mortality: an analysis of 22 European cohorts within the multicentre ESCAPE project
Lancet
(2014) - et al.
Comparison of exposure estimation methods for air pollutants: ambient monitoring data and regional air quality simulation
Environ. Res.
(2012) - et al.
A review of land-use regression models to assess spatial variation of outdoor air pollution
Atmos. Environ.
(2008) - et al.
Association between air pollution and coronary artery calcification within six metropolitan areas in the USA (the Multi-Ethnic Study of Atherosclerosis and Air Pollution): a longitudinal cohort study
Lancet
(2016) - et al.
Assessing temporally and spatially resolved PM2.5 exposures for epidemiological studies using satellite aerosol optical depth measurements
Atmos. Environ.
(2011) - et al.
A national satellite-based land-use regression model for air pollution exposure assessment in Australia
Environ. Res.
(2014) - et al.
Pragmatic estimation of a spatio-temporal air quality model with irregular monitoring data
Atmos. Environ.
(2011) - et al.
A regionalized national universal kriging model using partial least squares regression for estimating annual PM2.5 concentrations in epidemiology
Atmos. Environ.
(2013) - et al.
Comparison of land-use regression models between Great Britain and The Netherlands
Atmos. Environ.
(2010) - et al.
Temporal stability of land use regression models for traffic-related air pollution
Atmos. Environ.
(2013)
Development of long-term spatiotemporal models for ambient ozone in six metropolitan regions of the United States: the MESA air study
Atmos. Environ.
Basics of point-referenced data models
A national prediction model for components of PM2.5 and measurement error corrected health effect inference
Environ. Health Perspect.
Effect of air pollution control on life expectancy in the United States: an analysis of 545 U.S. counties for the period from 2000 to 2007
Epidemiol
Development of land use regression models for PM(2.5), PM(2.5) absorbance, PM(10) and PM(coarse) in 20 European study areas; results of the ESCAPE project
Environ. Sci. Technol.
The computation of geographic variables for air pollution prediction models in South Korea
Environ. Health Toxicol.
Spatial modeling of PM10 and NO2 in the continental United States, 1985-2000
Environ. Health Perspect.
Cited by (40)
Alzheimer's disease-like cortical atrophy mediates the effect of air pollution on global cognitive function
2023, Environment InternationalUsing big data searching and machine learning to predict human health risk probability from pesticide site soils in China
2022, Journal of Environmental ManagementCitation Excerpt :Pollutant toxicity assessment is a step to calculate the carcinogenic and non-carcinogenic harmfulness of a pollutant to the human body by using pollutant the carcinogenic slope or extrapolation model of the reference dose (CMEP, 2019). The majority of related studies have focused on the numerical simulation of human exposure concentration (Kim and Song, 2017), parameter optimization of the exposure model (Tighe et al., 2013), and uncertainty analysis of the exposure process (Yang et al., 2018). Ji et al. (2016) predicted HHR by comparing the measured pollutant concentration with the control value derived from the human dose effect, but this approach entails high uncertainty due to the absence of pollutant exposure assessment (Yang et al., 2018).
Structural and functional alterations of subjects with cement dust exposure: A longitudinal quantitative computed tomography-based study
2022, Science of the Total EnvironmentCitation Excerpt :The protocols of data collection of PM10 and NO2, and the description of air pollution model in CODA cohort were previously reported in detail (Kim et al., 2017a; Han et al., 2019; Kim and Song, 2017). From PM10 and NO2 measured in an every hour, at 277 regulatory monitoring sites in 2010, annual average concentrations of PM10 and NO2 at participants' home addresses were estimated using a universal kriging framework that consists of summary predictors of about 300 geographic variables and spatial correlation of air pollution concentrations (Kim and Song, 2017). From the CODA data, we extracted 311 baseline CDE subjects from Kangwon National University Hospital (KNUH).
Long-term exposure to particulate air pollution and incidence of Parkinson's disease: A nationwide population-based cohort study in South Korea
2022, Environmental ResearchCitation Excerpt :Moreover, it recently released the National Health Insurance Service-National Sample Cohort (NHIS-NSC), which is a nationwide, population-representative cohort (Lee et al., 2017b). Previously, we developed a national-scale exposure prediction model to estimate the annual average concentrations of PM2.5 and PM10 in South Korean residences (Kim and Song, 2017). In this study, we aimed to investigate the long-term association between PM2.5 and PM10 exposure and incident PD using a nationwide population-based cohort of South Korean citizens.
- ☆
This paper has been recommended for acceptance by Charles Wong.