Neighborhood-Level NO2 Machine Learning Models Toward Sustainable Monitoring Systems

• Main Author: Frazier-Duncan, Janay A.W
• Format: Dissertation
• Language: English
• Published: ProQuest Dissertations & Theses 01.01.2023
• Subjects: Artificial intelligence; Civil engineering; Environmental engineering
• ISBN: 9798358489226

Although many researchers have studied aspects of monitoring and modeling vehicle traffic emissions, more attention is needed for smaller than regional scale monitoring. National Ambient Air Quality Standards (NAAQS) are set by the Environmental Protection Agency (EPA) for six criterion pollutants and use a system of federally recognized analytical techniques to track compliance with those requirements known as Federal Equivalent Methods or Federal Reference Methods (FEMs or FRMs). FEM monitoring networks may be effective for assessing air pollution at a regional spatial resolution, which may not be suitable for monitoring at the neighborhood level due to their high cost and technological complexity. This research aims to bridge the knowledge gap in understanding the types of modeling sufficient for neighborhood-level emissions monitoring and how environmental and traffic engineers use them to enhance regional monitoring system networks. This quantitative study investigates the knowledge and comprehension gap of what is perceived from historical traffic data about the level of emissions occurring from on-road pollution and the ability to forecast levels without purchasing new systems but utilizing existing traffic count stations. The conceptual framework was based on emissions modeling measurements using meteorological data, volume count, and other emissions to assess the ability to build a forecasting model. The research questions were developed with the goal of creating a model that forecasts Nitrogen Dioxide (NO2) levels in Washington, D.C. Using historical data collected from data produced by the Department of Energy & Environment (DOEE) in support of the Environmental Protection Agency EPA and the Metropolitan Washington Council of Governments (MWCOG), which describes real-world transportation and traffic data for Washington, D.C., to support the District Department of Transportation (DDOT). Data were modeled using WEKA (a data mining machine learning tool) and analyzed using Python and Minitab. The main conclusions observed are which multiple statistical tests and machine learning (ML) algorithms evaluated had the highest performance, what model features were determined to be the most significant in forecasting NO2 levels using traffic data and other features for Washington D.C., and based on those models can ranges of NO2 and their corresponding levels can be forecasted for other smart cities based on the framework developed in this praxis. Based on these findings, the key recommendations are to have Washington, D.C., utilize the model in this praxis by using existing systems to improve the monitoring network.The study's findings could further advance the understanding of monitoring and modeling of forecasting of hazardous NO2. This contribution has implications for positive neighborhood-level monitoring system improvements by identifying reactive emissions monitoring practices, which can lead to high-quality continuous reduction in vehicle NO2 emission impacts.