Michanowicz, Andrew / R.
(2015)
Hybrid Dispersion/ Land Use Regression Modeling for Improving Air Pollutant Concentration Estimates.
Doctoral Dissertation, University of Pittsburgh.
(Unpublished)
Abstract
The overall objective of this dissertation was to examine the utility of incorporating source-meteorological interaction information from two commonly employed atmospheric dispersion models into the land use regression technique for predicting ambient NO2 and PM2.5. Ultimately, we are interested in obtaining highly resolved spatiotemporal pollutant estimates to examine the attenuation of health effect estimate bias that may result from exposure model misspecification. A multi-pollutant sampling campaign was conducted across six successive weekly sampling sessions in the summer and winter seasons of 2011-2013 in Pittsburgh, PA. As a preliminary investigation, predictions from a roadway dispersion model (Caline3) were included as an independent predictor in pre-constructed winter season LUR models for NO2. Caline3 output improved out-of-sample model fitness and added an additional portion of unexplained variation (3-10% by leave-one-out cross-validated R2) in NO2 observations compared to the standard LUR models. Correspondingly, the AERMOD dispersion model was implemented to predict PM2.5 from local and regional stationary sources in a similar hybrid framework. As per cross-validated R2 and RMSE, AERMOD predictions improved overall model fitness and explained an additional 9-13% in out-of-sample variability in summer and winter PM2.5 models. Both dispersion model output functioned similarly when incorporated into standard LUR models, effectively displacing the respective GIS-based covariates, corroborating model interpretability, and capturing the greatest degree of improvements at nearby, high-density source locations. To examine the potential for spatially-differential exposure measurement improvement in health effect estimation studies, we applied LUR and hybrid LUR/ dispersion model PM2.5 predictions to non-sampled locations and observed non-Berkson-type measurement error only when the modeling domain was restricted to a near-source (<1km) environment. By a simple stochastic simulation, we demonstrated that a well characterized dispersion-derived geographic covariate, defined by a robust variance about the monitoring locations, can theoretically result in less exposure measurement error and exposure misclassification. Therefore, highly refined spatiotemporal information can improve out-of-sample prediction accuracy; however, the statistical fidelity remains constrained by the degree of source contribution captured by monitoring locations. These findings have important public health implications for understanding air pollutant exposure measurement error derived from typical LUR studies. In the absence of a spatially dense monitoring network, we demonstrated that AERMOD can produce a spatiotemporally resolved prediction surface compared to typical GIS-based covariates across a large urban-to-suburban domain with pertinent pollutant sources and complex topography.
Share
| Citation/Export: |
|
| Social Networking: |
|
Details
| Item Type: |
University of Pittsburgh ETD
|
| Status: |
Unpublished |
| Creators/Authors: |
| Creators | Email | Pitt Username | ORCID  |
|---|
| Michanowicz, Andrew / R. | arm73@pitt.edu | ARM73 | |
|
| ETD Committee: |
|
| Date: |
28 January 2015 |
| Date Type: |
Publication |
| Defense Date: |
1 December 2014 |
| Approval Date: |
28 January 2015 |
| Submission Date: |
22 November 2014 |
| Access Restriction: |
1 year -- Restrict access to University of Pittsburgh for a period of 1 year. |
| Number of Pages: |
117 |
| Institution: |
University of Pittsburgh |
| Schools and Programs: |
School of Public Health > Environmental and Occupational Health |
| Degree: |
DrPH - Doctor of Public Health |
| Thesis Type: |
Doctoral Dissertation |
| Refereed: |
Yes |
| Uncontrolled Keywords: |
NA |
| Date Deposited: |
28 Jan 2015 15:57 |
| Last Modified: |
15 Nov 2016 14:25 |
| URI: |
http://d-scholarship.pitt.edu/id/eprint/23582 |
Metrics
Monthly Views for the past 3 years
Plum Analytics
Actions (login required)
 |
View Item |
![[img]](http://d-scholarship.pitt.edu/style/images/fileicons/application_pdf.png)
![[feed]](/images/feed-icon-32x32.png){kind=icon}