A J Abstract 681, Paris 2006ISEA-ISEE Conference Spatial Analysis Of Air Pollution And Development Of A Land-Use Regression Model In An Urban Airshed Shaibal Mukerjee1*, Luther Smith2, Xiaojuan Liao2, Lucas Neas3, Casson Stallings2, and Mary Johnson3 1U.S. EPA/ORD/NERL E205-02, RTP, NC, USA; 2Alion Science & Technology, Durham, NC, USA; 3U.S. EPA/ORD/NHEERL, RTP, NC, USA Introduction All sampling in Detroit, Michigan, USA area during Summer 2005. Passive sampling of VOC & N02 at 25 local elementary (PK-6) schools in Detroit and Dearborn Public Schools. Sampling also done at 2 State of Michigan regulatory (reg) sites. This poster presents the exposure component of the EPA/ORD Detroit Children's Health Study (DCHS). Overall spatial assessment & development of land-use regression (LUR) model will be discussed. Approach based on spatial approach in El Paso Children's Health Study (Smith et al., Atmos Environ 40 (2006) 3773- 3787). Methods • VOCs: Carbopack X sorbent thermal desorption tubes (Supelco) (McClenny et al. JEM8 (2006) 263-9) • N02: Ogawa (Model 3300) (Mukerjee et al JAWMA 54 (2004) 307-19) • Sampled 6 weeks during stable air masses & low winds 1 Week-long sampling ¦ exposures mimic chronic Fig. 1. Schools and their Locations Relative to Enumeration Districts (EDs) in Detroit Area Si ¦M, JL.I rrKv. - 1 . i ^ A Schools • Regiittory (reg) sites Mayor .-wkK Enumeration districts (EDs) ~ City limti EJ EO 57 ,-irea not studied lipr — 4 i p 4 a ¦ ; ¦ Outline of Approach 1. Correlation analysis to determine ancillary (GIS) variables for prediction 2. Pattern analysis to select school sites 3. Comparison of EDs (first 2 digits of 2000 US Census tract number) 4. Comparison of regulatory (reg) sites with neighboring schools to assess representativeness of the regulatory sites for gaseous pollutants (Fig.1) 5. Development of LUR model for VOCs & N02 (Statistical programming in SAS® 8) Potential Ancillary Variables • Data from SE Michigan Council of Governments, National Center for Education Statistics, 2000 US Census, EPA TRI & NEI emissions inventories. • Variable types (relative to schools) from GIS: ¦ Distance (m) to nearest road of various traffic volumes (Dist_90KP = distance to road segment > 90,000 cars/day) ¦ Traffic intensity (vehicles per day/km) within set distances (lnt_1000 = intensity within 1000 m radius) ¦ Housing unit density (units/km2 i census block) ¦ Population density (people/km2 census block; Pop_Den500 = population density from census tract(s) within 500 m) ¦ Distance (m) to large VOC, PM2 5, Manganese point sources (VOC_Big_Dist, PM25_Big_Dist, Mn_Big_Dist - respectively) ¦ Distance (m) to nearest Border X-ing in in Choice of Variables & Selection of Schools for Monitoring • Correlation analysis: -Same correlation structure desired for monitored & un-monitored schools -Avoided strong correlation among chosen ancillary variables • Potential ancillary variables chosen: -Dist_50KP -Dist_90KP — Int_1000 -Pop_Den500 -VOC_Big_Dist -PM25_Big_Dist -Mn_Big_Dist -Distance to Border X-ing • Schools chosen (see Fig. 1) based on their ancillary variables (above) and had to be representative of study area; 4 to 5 schools selected per ED ED & Regulatory (reg)/School Comparisons of Air Pollutants • ED Comparison Tests (5% level) -Overall Kruskal-Wallis test - Pairwise multiple comparisons (analogous to t-test) - Modified Dunn's test (Dunn Technometrics6 (1964) 241-252) • Reg/School Comparisons -Comparison of regulatory site data to corresponding range of school data -Dixon's r10 ratio (5% level) (DixonAnn Math Stat 21 (1950) 488-506; 22 (1951) 68-78) Results Based on distance & traffic intensity values, 25 schools chosen were generally representative of variability of study area & their ED. Similar correlation structure among ancillary variables between monitored & remaining schools. Overall Spatial Analysis of Air Pollution Data from Chosen Schools Overall & pairwise comparisons of EDs suggested only N02 showed coarse spatial difference. For N02, ED 52 (high traffic/industrial impacted area) significantly higher than ED 54 (residential area) - see Fig. 2. Fig. 4 - Comparison of N02 & VOCs from reg sites (yellowed) and respective schools (red and blue regions). Pollutants at reg sites within range of their schools. If outside of range, no significant difference (Dixon's r10 ratio, 5% level). Initial Regression Results • Natural log transformation applied for all pollutant and some ancillary variables. • Only ancillary variables that behaved linearly used. • Weighted regressions used. Table 1. Initial regression models. Only significant (5% level) coefficients reported Fig. 2. Median Values of N02, Total BTEX, & Styrene Over All Schools and By Each ED i HI EO 50 M EO 51 M £0 52 ZZ3 £0 53 ¦ ED54 ttm £0 57 • All Schools Signrficant Difference i—-i(Modified Dunn's Test} ¦If ¦¦ I I. Pollutant Intercept Dist_50KP DiSt_90KP IntJOOO Pop_Den500 VOC_ Big_Dist PM25_Big_Dist Mn_Big_Dist Border R2* (%) Ln N02 4.43 -.052 (Ln)" .023 (Ln) -.095 (Ln) -0.082 (Ln) -1.5 xior5 82 Ln Benzene 6.23 -5.04X10"® 5.99 X10"5 -3.8 x10"5 1.23 x10"5 43 Ln Toluene 7.42 -4.17 X10"5 1.92 x10"5 31 Ln Ethyl benzene 10.0 -.458 (Ln) 1.79 x10"5 63 Lno- xylene 6 -2.34 X10"5 .239 (Ln) .092 (Ln) -.345 (Ln) 60 Ln m,p- xylene 8.43 -1 X104 -2.81 X10"5 -1.51 X10"6 .166 (Ln) -.215 (Ln) 55 Ln 1.3- butadiene 4.56 1 X104 -2.95 X10"5 43 Ln Styrene 3.75 -2.47x10-' -2.03 X10'5 43 *R2 from original scale; ** Ln=natural logarithm of variable Discussion Overall spatial analysis results suggested mobile source effect throughout study area for VOCs. Only N02 exhibited coarse spatial difference between traffic/industrial-dominated city district versus a more residential district (see Fig. 2). In this study, regulatory sites were representative of neighboring locations (see Fig. 3). Regressions (in Table 1) were not as successful as hoped a priori. Highest R2 values obtained for N02. The relatively poor R2 values (and the overall spatial results) may be due to the fact that the hourly winds were found to be blowing in roughly equal proportions from each compass quadrant during each week of the study. Winds were almost always light to calm for the entire six-week period. Thus, the sites were subjected to multiple influences for every measurement period. Siting issues may have also contributed to these findings. Collinearity present in some of the regressions; suggested by coefficient values and collinearity diagnostics. Next efforts will address collinearity and cross-validation will be applied to the final predictive equations. Based on above tabulated results, traffic influences were important predictor variables for the Detroit/Dearborn area. In El Paso, traffic intensity (lnt_1000) was only important for predicting N02 (Smith et al., Atmos Environ 40 (2006) 3773-3787). In addition, distance from border crossing showed a consistent decline with increasing distance in El Paso; in Detroit, this was only the case for N02. These preliminary findings from the Detroit area suggest possible local differences should be factored in when attempting to derive common exposure metrics from data collected in different urban air sheds. Suggestions for Future Research • Additional monitoring to assess seasonal variability • Additional monitoring for PM probably would indicate better local spatial variability • Epidemiological study relating results to health data pending Acknowledgements and Disclaimer - We thank Felicia Venable, Mathew Sam, and Priscilla Morris of the Detroit Public Schools Department of Environmental Health & Safety and Don Ball of Dearborn Public Schools as well as school principals and staff for access to the schools. We also thank the technical support of Dr. Hunter Daughtrey, Karen Oliver, Herb Jacumin, Chris Fortune, Mike Wheeler, and Dennis Williams all from Alion Science and Technology. The United States Environmental Protection Agency through its Office of Research and Development funded and managed the research described here under contract EP-D-05-065 to Alion Science and Technology, Inc. It has been subjected to Agency review and approved for publication. ------- |