You are here:
Evaluation of land use regression models in Detroit, Michigan
Citation:
JOHNSON, M. M., L. Smith, S. MUKERJEE, R. W. WILLIAMS, E. E. HUDGENS, C. Stallings, H. A. OZKAYNAK, A. F. VETTE, C. W. CROGHAN, AND L. M. NEAS. Evaluation of land use regression models in Detroit, Michigan. Presented at International Society for Environmental Epidemiology, Pasadena, CA, October 12 - 14, 2008.
Impact/Purpose:
research results
Description:
Introduction: Land use regression (LUR) models have emerged as a cost-effective tool for characterizing exposure in epidemiologic health studies. However, little critical attention has been focused on validation of these models as a step toward temporal and spatial extension of land-use regression model prediction for application in multiple study populations. This presentation evaluated LUR models developed for Detroit Children’s Health Study (DCHS) using outdoor residential measurements from several Detroit-area studies. Methods: LUR estimated outdoor concentrations of benzene, toluene, ethylbenzene, mp and o-xylenes (BTEX) and nitrogen dioxide (NO2) were compared with outdoor residential NO2 and BTEX measurements collected under Detroit Exposure Aerosol Research Study (DEARS) and Mechanistic Indicators of Childhood Asthma-Air Monitoring (MICA-Air). LUR models were developed under DCHS based on outdoor air monitoring conducted at neighborhood schools during summer 2005. Monitoring data were used in conjunction with representative traffic, point sources, and other variables to develop LUR models. The DEARS study collected winter and summer outdoor measurements for Detroit homes from 2005-2007. MICA-Air collected outdoor residential measurements in fall 2006. Model predictions were evaluated through qualitative comparisons and linear regression analyses. Results: Qualitative comparison of estimated concentrations and measurements showed good agreement between low, low-medium, medium-high, and high outdoor residential concentrations for NO2 and BTEX species. Preliminary analysis showed good agreement between estimated and measured NO2 for outdoor residential concentrations within the range of measurements used to develop the LUR model. Estimated and measured BTEX species showed less agreement. Conclusions: Preliminary results suggest that land-use regression models were better able to characterize outdoor residential NO2 compared with BTEX species for the Detroit area models. Factors influencing the predictive power of these models including seasonality, pollutant species, sampling timeframes, and geographic area will be discussed further. These analyses provide insights on the strengths and limitations of applying LUR models in temporal and spatially heterogeneous study populations. Although this work was reviewed by EPA and approved for publication, it may not necessarily reflect official Agency polic