Development of roadway link screening criteria for microscale carbon monoxide and particulate matter conformity analyses through application of classification tree model

• Main Author: Shafi, Ghufran
• Published: Georgia Institute of Technology 2009
• Subjects: Sensitivity analysis; Environmental impacts; Project conformity analyis; Carbon monoxide; Combustion gases; Automobiles Motors Exhaust gas; Mathematical models
• Online Access: http://hdl.handle.net/1853/28222

The impacts of emissions sources of carbon monoxide and particulate matter pollution levels for projected level conformity assessment and National Environmental Policy Act (NEPA) analyses are usually estimated through computer-aided models. Because of the involvement and interaction of a large number of variables that affect formation of CO and PM hot spots, exhaustive impact assessment studies can be time consuming. This is especially true for complex urban projects consisting of numerous roadways whose potential CO and PM impacts on surrounding neighborhoods must be disclosed. A highway project may consist of hundreds of roadway links, therefore undertaking project level conformity analysis without screening tools can be computationally resource intensive. CALINE4, a line source emission modeling tool, is used to predict downwind CO and PM concentrations for various receptors to generate a learning dataset for development of screening rules. This research has developed statistical screening criteria based on Classification and Regression Tree modeling that can be used to eliminate those links from the CALINE4 analysis whose contribution of pollutant concentration to a particular receptor site are insignificant. For the purpose of this study, any link that contributes a concentration of 0 ppm of CO or 0 µg/m3 of PM to a particular receptor site is termed insignificant for the corresponding pollutant. The model uses seven predictor variables, namely wind speed, wind directional variability, linear emission flux, link length and receptor polar coordinates. Response vector has two classes of pollutant concentrations namely significant and insignificant which are obtained by conversion of numerical values of pollutant concentration according to above mentioned criterion, thereby converting a regression problem into categorical or classification problem. The developed rules based on constructed model were validated through test samples and can be applied to future dataset to classify and screen out the insignificant links in highway planning analyses. The screening tool also allows analysts to prepare gridded pollution concentration predictions for use in environmental justice analyses.