Evaluation of land use regression models for nitrogen dioxide and benzene in Four US cities (original) (raw)
Related papers
A Review of Land-use Regression Models for Characterizing Intraurban Air Pollution Exposure
Inhalation Toxicology, 2007
Epidemiologic studies of air pollution require accurate exposure assessments at unmonitored locations in order to minimize exposure misclassification. One approach gaining considerable interest is the land-use regression (LUR) model. Generally, the LUR model has been utilized to characterize air pollution exposure and health effects for individuals residing within urban areas. The objective of this article is to briefly summarize the history and application of LUR models to date outlining similarities and differences of the variables included in the model, model development, and model validation. There were 6 studies available for a total of 12 LUR models. Our findings indicated that among these studies, the four primary classes of variables used were road type, traffic count, elevation, and land cover. Of these four, traffic count was generally the most important. The model R 2 explaining the variability in the exposure estimates for these LUR models ranged from .54 to .81. The number of air sampling sites generating the exposure estimates, however, was not correlated with the model R 2 suggesting that the locations of the sampling sites may be of greater importance than the total number of sites. The primary conclusion of this study is that LUR models are an important tool for integrating traffic and geographic information to characterize variability in exposures.
Evaluation of land-use regression models used to predict air quality concentrations in an urban area
Atmospheric Environment, 2010
Cohort studies designed to estimate human health effects of exposures to urban pollutants require accurate determination of ambient concentrations in order to minimize exposure misclassification errors. However, it is often difficult to collect concentration information at each study subject location. In the absence of complete subject-specific measurements, land-use regression (LUR) models have frequently been used for estimating individual levels of exposures to ambient air pollution. The LUR models, however, have several limitations mainly dealing with extensive monitoring data needs and challenges involved in their broader applicability to other locations. In contrast, air quality models can provide high-resolution sourceeconcentration linkages for multiple pollutants, but require detailed emissions and meteorological information. In this study, first we predicted air quality concentrations of PM 2.5 , NO x , and benzene in New Haven, CT using hybrid modeling techniques based on CMAQ and AERMOD model results. Next, we used these values as pseudo-observations to develop and evaluate the different LUR models built using alternative numbers of (training) sites (ranging from 25 to 285 locations out of the total 318 receptors). We then evaluated the fitted LUR models using various approaches, including: 1) internal "Leave-One-Out-Cross-Validation" (LOOCV) procedure within the "training" sites selected; and 2) "Hold-Out" evaluation procedure, where we set aside 33e293 tests sites as independent datasets for external model evaluation. LUR models appeared to perform well in the training datasets. However, when these LUR models were tested against independent hold out (test) datasets, their performance diminished considerably. Our results confirm the challenges facing the LUR community in attempting to fit empirical response surfaces to spatially-and temporally-varying pollution levels using LUR techniques that are site dependent. These results also illustrate the potential benefits of enhancing basic LUR models by utilizing air quality modeling tools or concepts in order to improve their reliability or transferability.
Spatial analysis and land use regression of VOCs and NO2 in Dallas, Texas during two seasons
Journal of Environmental Monitoring, 2011
Passive air sampling for nitrogen dioxide (NO 2 ) and select volatile organic compounds (VOCs) was conducted at 24 fire stations and a compliance monitoring site in Dallas, Texas, USA during summer 2006 and winter 2008. This ambient air monitoring network was established to assess intra-urban gradients of air pollutants to evaluate the impact of traffic and urban emissions on air quality. Ambient air monitoring and GIS data from spatially representative fire station sites were collected to assess spatial variability. Pairwise comparisons were conducted on the ambient data from the selected sites based on city section. These weeklong samples yielded NO 2 and benzene levels that were generally higher during the winter than the summer. With respect to the location within the city, the central section of Dallas was generally higher for NO 2 and benzene than north and south. Land use regression (LUR) results revealed spatial gradients in NO 2 and selected VOCs in the central and some northern areas. The process used to select spatially representative sites for air sampling and the results of analyses of coarse-and fine-scale spatial variability of air pollutants on a seasonal basis provide insights to guide future ambient air exposure studies in assessing intra-urban gradients and traffic impacts. † Electronic supplementary information (ESI) available. Tables of ancillary variables considered for use in LUR models, passive method evaluation, mean concentrations at each site, and maps of measured NO 2 and benzene concentrations. See
Journal of the Air & Waste Management Association, 2006
This paper reports on the development of a land use regression (LUR) model for predicting the intraurban variation of traffic-related air pollution in Hamilton, Ontario, Canada, an industrial city at the western end of Lake Ontario. Although land use regression has been increasingly used to characterize exposure gradients within cities, research to date has yet to test whether this method can produce reliable estimates in an industrialized location. Ambient concentrations of nitrogen dioxide (NO 2) were measured for a 2-week period in October 2002 at Ͼ100 locations across the city and subsequently at 30 of these locations in May 2004 to assess seasonal effects. Predictor variables were derived for land use types, transportation, demography, and physical geography using geographic information systems. The LUR model explained 76% of the variation in NO 2. Traffic density, proximity to a highway, and industrial land use were all positively correlated with NO 2 concentrations, whereas open land use and distance from the lake were negatively correlated with NO 2. Locations downwind of a major highway resulted in higher NO 2 levels. Cross-validation of the results confirmed model stability over different seasons. Our findings demonstrate that land use regression can effectively predict NO 2 variation at the intraurban scale in an industrial setting. Models predicting exposure within smaller areas may lead to improved detection of health effects in epidemiologic studies.
Journal of the Air & Waste Management Association, 2005
The purpose of this study was to derive a land-use regression model to estimate on a geographical basis ambient concentrations of nitrogen dioxide (NO 2 ) in Montréal, Quebec, Canada. These estimates of concentrations of NO 2 will be subsequently used to assess exposure in epidemiologic studies on the health effects of traffic-related air pollution. In May 2003, NO 2 was measured for 14 consecutive days at 67 sites across the city using Ogawa passive diffusion samplers. Concentrations ranged from 4.9 to 21.2 ppb (median 11.8 ppb). Linear regression analysis was used to assess the association between logarithmic concentrations of NO 2 and land-use variables derived using the ESRI Arc 8 geographic information system. In univariate analyses, NO 2 was negatively associated with the area of open space and positively associated with traffic count on nearest highway, the length of highways within any radius from 100 to 750 m, the length of major roads within 750 m, and population density within 2000 m. Industrial land-use and the length of minor roads showed no association with NO 2 . In multiple regression analyses, distance from the nearest highway, traffic count on the nearest highway, length of highways and major roads within 100 m, and population density showed significant associations with NO 2 ; the best-fitting regression model had a R 2 of 0.54. These analyses confirm the value of land-use regression modeling to assign exposures in large-scale epidemiologic studies.
Accounting for spatial effects in land use regression for urban air pollution modelling
Spatial and Spatio-temporal Epidemiology, 2015
In order to accurately assess air pollution risks, health studies require spatially resolved pollution concentrations. Land-use regression (LUR) models estimate ambient concentrations at a fine spatial scale. However, spatial effects such as spatial non-stationarity and spatial autocorrelation can reduce the accuracy of LUR estimates by increasing regression errors and uncertainty; and statistical methods for resolving these effects -e.g., spatially autoregressive (SAR) and geographically weighted regression (GWR) models -may be difficult to apply simultaneously.
The transferability of NO and NO2 land use regression models between cities and pollutants
Atmospheric Environment, 2011
Land use regression (LUR) models are commonly used for exposure assessment in epidemiologic studies of traffic-related air pollution. Models in different cities often contain similar predictors, suggesting that models may be transferable between cities with similar characteristics. LUR models of NO or NO 2 may also be useful for estimating exposure to other pollutants in the traffic pollution mixture. We evaluated the transferability of NO and NO 2 LUR models between Winnipeg, Manitoba and Edmonton, Alberta, and the ability of the Winnipeg NO and NO 2 LUR models to predict concentrations of benzene and toluene. In both cities, LUR models were developed based on measurements with Ogawa passive samplers at 50 locations during two 14-day sampling campaigns, while benzene and toluene concentrations in Winnipeg were measured at 46 independent locations using 3M #3500 passive badges during a single 14-day period. Locally calibrated LUR models explained more variability in NO 2 (R 2 : 0.81e0.84) than NO (R 2 : 0.55e0.56). Although models transferred to the opposite city did not perform as well as the locally calibrated models for NO 2 (R 2 : 0.37e0.52) or NO (R 2 : 0.24e0.41), the transferred models explained more variability than simple binary or continuous road proximity metrics (R 2 0.19), which are commonly used in epidemiologic studies. In Winnipeg, the NO 2 LUR model explained 34% and 19% of the variation in benzene and toluene, respectively, while road proximity metrics explained 19% of the variation in both pollutants. In conclusion, epidemiologic studies will achieve better exposure assessments by developing LUR models locally and for the pollutant of interest, though transferred LUR models may provide a cost-effective improvement over road proximity metrics for assessing exposure to trafficrelated air pollution.
Atmospheric Environment, 2010
More than 25 studies have employed land use regression (LUR) models to estimate nitrogen oxides and to a lesser extent particulate matter indicators, but these methods have been less commonly applied to ambient concentrations of volatile organic compounds (VOCs). Some VOCs have high plausibility as sources of health effects and others are specific indicators of motor vehicle exhaust. We used LUR models to estimate spatial variability of VOCs in Toronto, Canada. Benzene, n-hexane and total hydrocarbons (THC) were measured from July 25 to August 9, 2006 at 50 locations using the TraceAir organic vapor monitors. Nitrogen dioxide (NO 2 ) was also sampled to assess its spatial pattern agreement with VOC exposures. Buffers for land use, population density, traffic density, physical geography, and remote sensing measures of greenness and surface brightness were also tested. The remote sensing measures have the highest correlations with VOCs and NO 2 levels (i.e., explains >36% of the variance). Our regression models explain 66e68% of the variance in the spatial distribution of VOCs, compared to 81% for the NO 2 model. The ranks of agreement between various VOCs range from 48 to 63% and increases substantially e up to 75% e for the top and bottom quartile groups. Agreements between NO 2 and VOCs are much smaller with an average rank of 36%. Future epidemiologic studies may therefore benefit from using VOCs as potential toxic agents for traffic-related pollutants.