Identifier
etd-07092009-100152
Degree
Doctor of Philosophy (PhD)
Department
Chemical Engineering
Document Type
Dissertation
Abstract
The adverse health e_x000B_ffects of tropospheric ozone around urban zones indicate a substantial risk for many segments of the population. This necessitates the short term forecast in order to take evasive action on days conducive to ozone formation. Therefore it is important to study the ozone formation mechanisms and predict the ozone levels in a geographic region. Multivariate statistical techniques provide a very e_x000B_ffective framework for the classifi_x000C_cation and monitoring of systems with multiple variables. Cluster analysis, sequence analysis and hidden Markov models (HMMs) are statistical methods which have been used in a wide range of studies to model the data structure. In this dissertation, we propose to formulate, implement and apply a data-driven computational framework for air quality monitoring and forecasting with application to ozone formation. The proposed framework integrates, in a unique way, advanced statistical data processing and analysis tools to investigate ozone formation mechanisms and predict the ozone levels in a geographic region. This dissertation focuses on cluster analysis for identi_x000C_fication and classi_x000C_fication of underlying mechanisms of a system and HMMs for predicting the occurrence of an extreme event in a system. The usefulness of the proposed methodology in air quality monitoring is demonstrated by applying it to study the ozone problem in Houston, Texas and Baton Rouge, Louisiana regions. Hierarchical clustering is used to visualize air flow patterns at two time scales relevant for ozone buildup. First, clustering is performed at the hourly time scale to identify surface flow patterns. Then, sequencing is performed at the daily time scale to identify groups of days sharing similar diurnal cycles for the surface flow. Selection of appropriate numbers of air flow patterns allowed inference of regional transport and dispersion patterns for understanding population exposure to ozone. This dissertation proposes to build HMMs for ozone prediction using air quality and meteorological measurements obtained from a network of surface monitors. The case study of the Houston, Texas region for the 2004 and 2005 ozone seasons showed that the results indicate the capability of HMMs as a simpler forecasting tool.
Date
2009
Document Availability at the Time of Submission
Release the entire work immediately for access worldwide.
Recommended Citation
Pakalapati, Swathi, "Integrated data-driven techniques for environmental pollution monitoring" (2009). LSU Doctoral Dissertations. 675.
https://repository.lsu.edu/gradschool_dissertations/675
Committee Chair
Jose A. Romagnoli
DOI
10.31390/gradschool_dissertations.675