*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
School of Civil and Environmental Engineering
Ph.D. Thesis Defense Announcement
Development and Application of Data Fusion and Source Apportionment Methods over the
Contiguous United States
By:
Nirupama Senthilkumar
Advisor:
Dr. James Mulholland (CEE), Dr. Armistead Russell (CEE)
Committee Members:
Dr. Jennifer Kaiser (CEE), Dr. Pengfei Liu (EAS), Dr. Howard Chang (Emory
University)
Date and Time: Wednesday, June 8, 2022 9:00AM
Location: SEB122, Zoom
Exposure to air pollution has been linked to numerous adverse health effects such as cardiovascular
diseases, pulmonary diseases, cancer, and increased morbidity. Having accurate air quality exposure
estimates are important to understanding the drivers of negative health outcomes. Air quality simulations
and observational data are used as inputs in health analysis to estimate exposure to air pollution.
However, observational data are limited spatially and temporally while air quality simulated data have
biases associated. This dissertation presents multiple computational techniques to provide
spatiotemporally accurate and complete air quality and source impacts fields for health analysis. A data
fusion method along with a random forest technique is used to generate fused fields for particulate, gas,
and trace metal species at a 12km resolution for the years 2005-2014. The data fusion method combines
gridded simulations from the community multiscale air quality (CMAQ) model and point source
observational data to create more accurate spatiotemporally complete air quality fields. The data fusion
method creates high temporal correlations at observational locations for all species studied. The random
forest approach uses land use variable information to correct spatial bias in annual average for fused field
products. The data fusion and random forest method showed large improvements in spatial and temporal
correlation for major particulate and gas species, and moderate improvements for trace metal pollutants.
The fused field products were then used in a receptor model source apportionment analysis for
particulate matter. A receptor model, chemical mass balance with gas constraints (CMBGC), was applied
in each 12km fused field grid cell to generate spatiotemporally complete source impact fields for 10
particulate matter sources: gasoline vehicles, diesel vehicles, dust, biomass burning, coal combustion,
ammonium sulfate, ammonium bisulfate, ammonium nitrate, secondary organic carbon, and salt. A
CMBGC model was also applied to each 12km CMAQ grid cell to compare the improvements made in
source impact fields from applying the data fusion and random forest correction. The comparison showed
that data fusion was necessary to produce accurate source impact fields.
The implications from this research show that data fusion can provide large improvements in air quality
fields for health analysis. Fused fields are also able to provide spatiotemporally complete particulate
matter source impact fields that match source impacts generated from observations. The daily data fused
fields for 22 species and daily source impact fields are made available for future health and air quality
analysis.