Elsevier

Atmospheric Environment

Volume 64, January 2013, Pages 312-319
Atmospheric Environment

Temporal stability of land use regression models for traffic-related air pollution

https://doi.org/10.1016/j.atmosenv.2012.09.056Get rights and content

Abstract

Background

Land-use regression (LUR) is a cost-effective approach for predicting spatial variability in ambient air pollutant concentrations with high resolution. Models have been widely used in epidemiological studies and are often applied to time periods before or after the period of air quality monitoring used in model development. However, it is unclear how well such models perform when extrapolated over time.

Objective

The objective of this study was to assess the temporal stability of LUR models over a period of 7 years in Metro Vancouver, Canada.

Methods

A set of NO and NO2 LUR models based on 116 measurements were developed in 2003. In 2010, we made 116 measurements again, of which 73 were made at the exact same location as in 2003. We then developed 2010 models using updated data for the same predictor variables used in 2003, and also explored additional variables. Four methods were used to derive model predictions over 7 years, and predictions were compared with measurements to assess the temporal stability of LUR models.

Results

The correlation between 2003 NO and 2010 NO measurements was 0.87 with a mean (sd) decrease of 11.3 (9.9) ppb. For NO2, the correlation was 0.74, with a mean (sd) decrease of 2.4 (3.2) ppb. 2003 and 2010 LUR models explained similar amounts of spatial variation (R2 = 0.59 and R2 = 0.58 for NO; R2 = 0.52 and R2 = 0.63 for NO2, in 2003 and in 2010 respectively). The 2003 models explained more variability in the 2010 measurements (R2 = 0.58–0.60 for NO; R2 = 0.52–0.61 for NO2) than the 2010 models explained in the 2003 measurements (R2 = 0.50–0.55 for NO; R2 = 0.44–0.49 for NO2), and the 2003 models explained as much variability in the 2010 measurements as they did in the 2003 measurements.

Conclusion

LUR models are able to provide reliable estimates over a period of 7 years in Metro Vancouver. When concentrations and their variability are decreasing over time, the predictive power of LUR models is likely to remain the same or to improve in forecasting scenarios, but to decrease in hind-casting scenarios.

Highlights

► Land use regression (LUR) models of traffic air pollution can be extended over time. ► LUR model application is improved for forecasting vs hind-casting scenarios. ► Epidemiologic studies can reliably apply LUR models over ∼7 year time periods.

Introduction

Land use regression (LUR) models are now frequently used as a cost-effective approach for assessing intra-urban air pollution contrasts (Briggs et al., 1997; Hoek et al., 2008; Kashima et al., 2009; Allen et al., 2011). Typically, air pollutants are measured at multiple sites and associated with potentially predictive geographic attributes (e.g., land use, population density, traffic patterns) to build regression models that can be rendered in a Geographic Information System (GIS) to predict concentrations at unmeasured locations (Henderson et al., 2007). Once a LUR model is developed, individual exposure to air pollution can be estimated by geocoding addresses of interest (e.g. subject's residential addresses, schools, etc.) and determining the modeled concentrations at those locations (Henderson et al., 2007).

Many epidemiological studies have adopted LUR models for exposure assessment over periods of different years. In most cases, models were developed over a single year and extrapolated forward (Morgenstern et al., 2008; Ryan et al., 2008) or backward (Brauer et al., 2008; Cesaroni et al., 2008; Clark et al., 2010; Karr et al., 2009a,b; Morgenstern et al., 2006, 2008; Slama et al., 2007) in time to assess the exposures of individual study subjects during specific time periods of interest. In these studies, the gap between the time of the exposure estimate and the LUR model development ranged from 0 to 7 years. Most studies have directly applied models over time without any adjustments, and therefore do not account for temporal variability in exposures due to long-term changes in air pollutant concentrations. The few studies that have accounted for these changes, have used yearly calibration (Molter et al., 2010) or by monthly adjustment (Brauer et al., 2008; Clark et al., 2010; Gan et al., 2011; MacIntyre et al., 2011; Slama et al., 2007). Such approaches may be inadequate because they typically rely on measurements at fewer locations than used to develop the LUR models themselves, and assume that temporal changes are constant across the study area. More generally, spatial prediction models derived from continuous measurements may be better at describing temporal trends but typically are more limited in spatial resolution. Interpolation methods such as kriging, relying on deterministic and stochastic geostatistical techniques (Jerrett et al., 2005), have mainly been used at the regional and national scales (Bell, 2006; Liao et al., 2006). However, as ordinary kriging does not account for factors such as terrain or localized patterns (Jerrett et al., 2005), it cannot reveal small-scale spatial variation (Aguilera et al., 2007).

Land use and traffic emissions change over time. Therefore spatial contrasts of pollutant concentrations predicted from these characteristics may not remain stable. This instability may lead to exposure misclassification, which can affect epidemiological results. For example, if pollutant concentrations decrease more rapidly at some locations than others, exposure estimates based on these initial concentrations will have greater misclassification with respect to concentrations over the long-term. Accordingly, our objective was to evaluate the validity of applying LUR models to periods that are temporally distinct from the time for which the model was built.

Specifically, we evaluated the temporal stability of a LUR model to retrospectively and prospectively estimate subjects' chronic exposure to nitrogen oxides. In Vancouver, LUR models for NO and NO2 were developed in 2003 (Henderson et al., 2007) and have been applied in several epidemiological analyses (Brauer et al., 2008; Clark et al., 2010; Gan et al., 2011; MacIntyre et al., 2011). In this study, we made new measurements in 2010 update the model. The temporal stability of both the original and new LUR models over the 7-year period was evaluated by comparing model predictions with measured spatial contrasts between 2003 and 2010. Findings have specific relevance to longitudinal studies where the exposure period of interest spans several years. In addition, we also sought to enhance the 2010 models by including new variables that were not tested previously.

Section snippets

Materials and methods

In 2003, models were built for NO, NO2, fine particulate matter (PM2.5) and light absorbance (ABS) (Henderson et al., 2007). Using the same methods developed for the 2003 models, we repeated NO and NO2 measurements, updated the input predictor variables, and constructed new models (referred to herein as the 2010 models).

NO and NO2 measurements

Table 2 summarizes descriptive statistics of measured NO and NO2 concentrations in 2010. The NO concentrations ranged from 2.5 to 49.8 ppb, and were log-normally distributed (n = 116) with a geometric mean (geometric standard deviation (gsd)) of 13.6 (1.9) ppb. The NO2 concentrations ranged from 3.1 to 17.6 ppb, and were normally distributed with an arithmetic mean (sd) of 10.9 (3.3) ppb.

Quality control showed no major bias or error. The NO and NO2 measurements from co-located Ogawa samplers

Discussion

The primary motivation of this study was to evaluate the temporal stability of LUR models when they are used to retrospectively or prospectively estimate long-term exposure in epidemiological studies. The 2003 LUR model explained as much variability in the 2010 measurements (forecasting) as it did in the 2003 measurements, indicating no loss of applicability of the model when prospectively predicting spatial contrasts over a 7-year period. Our 2010 LUR model explained as much variability in the

Conclusion

In this study, LUR models were able to predict concentrations in either forecasting or hindcasting. For study areas where pollutant concentrations are decreasing over time, the predictive power of a local LUR model is likely to be retained or improved in forecasting, but decreased in hindcasting applications. The findings can inform the assignment of exposure in studies of the health effects of chronic exposure to air pollution.

Acknowledgments

This work was supported by AllerGen NCE Inc. (The Allergy, Genes and Environment Network), a member of the Network of Centres of Excellence Canada Program.

References (34)

  • J.G. Su et al.

    A land use regression model for predicting ambient volatile organic compound concentrations in Toronto, Canada

    Atmospheric Environment

    (2010)
  • G.J.M. Velders et al.

    Meteorological variability in NO2 and PM10 concentrations in the Netherlands and its relation with EU limit values

    Atmospheric Environment

    (2009)
  • I. Aguilera et al.

    Using land-use regression modeling to estimate exposure to VOCs in a cohort of pregnant women

    Epidemiology

    (2007)
  • Allen, R., Gombojav, E., Barkhasragchaa, B., Byambaa, T., Lkhasuren, O., Amram, O., Takaro, T., Janes, C.R. An...
  • D.O. Atari et al.

    Spatial variability of ambient nitrogen dioxide and sulfur dioxide in Sarnia, Chemical Valley, Ontario, Canada

    Journal of Toxicology and Environmental Health – Part A-Current Issues

    (2008)
  • M. Brauer et al.

    A cohort study of traffic related air-pollution impacts on birth outcomes

    Environmental Health Perspectives

    (2008)
  • D.J. Briggs et al.

    Mapping urban air pollution using GIS: a regression-based approach

    International Journal of Geographical Information Science

    (1997)
  • Cited by (179)

    View all citing articles on Scopus
    View full text