Estimation of greenhouse gas emissions from a tracer gas study

Heilmann, Cory
Major Professor
Philip Dixon
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Organizational Unit
Journal Issue

This thesis models the emission of three greenhouse gases that exist in nature, CH4, CO2, and N2O, from a hoop structure, using the artificial introduction of a tracer gas, SF6, which exists in nature at levels below detection limit. Hoop structures are facilities used to house pigs before their slaughter. Many other studies of hoop structure emissions measure only one sample of the tracer gas and greenhouse gas at a time. However, the data sets in our study consist of 25 to 45 samples of each gas, taken on fixed grids of points. We will construct models to account for the relation between the greenhouse gases and SF6 as well as spatial relations in the data set. We will use these models along with the known emission rate of SF6 to estimate the relative rate of emission of the greenhouse gases;We fit a Bayesian hierarchical model to the data sets. In this model, we relate the pointwise concentrations of one greenhouse gas and SF6 and then analyze the posterior distribution of a parameter representing the relative rates of emission of the greenhouse gas and SF6. We assume lognormal measurement errors of the greenhouse gas and SF6 around the true concentration of each gas;We also fit geostatistical models to estimate the rates of emission of these gases. We consider block kriging, block co-kriging, and lognormal block kriging to estimate the concentration of each gas. An advantage of geostatistical models over the Bayesian hierarchical model is that we do not assume strict proportionality of the concentrations of the gases. These estimates can be related to the relative rate of emission of the gases. Due to the small size of these data sets, we take into consideration the uncertainty of the variogram parameters and how this uncertainty affects block kriging averages and variances;We use simulations from both geostatistical models and Bayesian hierarchical models to determine superiority of one set of models in terms of coverage probabilities, bias, and length of coverage sets or confidence intervals. We also address the concern of spatial design for the geostatistical models.