AAAR 36th Annual Conference October 16 - October 20, 2017 Raleigh Convention Center Raleigh, North Carolina, USA
Abstract View
Evaluation of the Transferability of Resolved Vs Unresolved Land Use Regression Models for Traffic-Related Air Pollutants
KEROLYN SHAIRSINGH, Cheol H. Jeong, Greg J. Evans, SOCAAR, University of Toronto
Abstract Number: 239 Working Group: Aerosol Exposure
Abstract Land use regression (LUR) models are statistical models used to predict intra-urban variability of air pollution and are frequently employed in mapping human exposure to traffic-related air pollutants. While LUR models developed for a specific city should be able to predict concentrations in other cities with similar infrastructure, land use, topography and climate; this transferability of models generally has been poor.
In this study, we resolved mobile monitored high-resolution concentrations of black carbon (BC), ultrafine particles (UFP), nitric oxide (NO) and nitrogen dioxide (NO2) into local and background signals, using time averaged minimum values, to investigate whether separation of the ambient measurements will improve LUR transferability. LUR models based on the unresolved (total) and resolved (local and background) input data were then developed for Toronto, Canada. These unresolved and resolved models were transferred to cities outside of the model domain, and their predictive performances were evaluated.
Our results showed that resolved models moderately improved the transferability of UFP and NO2. The resolved and unresolved models displayed R2 of 0.54 and 0.54 for UFP and 0.60 and 0.62 for NO2, respectively. When transferred to cities outside of the model domain, the local models displayed the lowest reduction in R2 for all pollutants, while background models transferred better than total models for UFP and NO2. This resulted in better transferability of the resolved models for NO2 (R2=0.36) and UFP (R2=0.3) when compared to their unresolved models (NO2:R2=0.26, UFP: R2=0.22).