University of Twente Student Theses


Decomposition of Particulate Matter in to its components and their prediction : Bayesian Hierarchical modeling

Shingade, Sharad Gorakh (2012) Decomposition of Particulate Matter in to its components and their prediction : Bayesian Hierarchical modeling.

[img] PDF
Abstract:Particulate Matter (PM) receives global attention due to their association with human health and environment. The effects caused by PM depends on the chemical composition, origin and particle size. Detailed knowledge of PM and its components are required for understanding their effects, source appointment studies and policy making. PM and its components measured by in situ measurement techniques are limited at few locations which leads to uncertainty in prediction. To overcome the above mentioned problem, a study which could efficiently predict the decomposition of Particulate Matter into its components was required. Thus, in this study, PM components were modeled in Bayesian hierarchical paradigm with added strength from a densely gridded covariate like CTM (chemical transport model) and AOT (aerosol optical thickness). Bayesian hierarchical modeling have an advantage over classical geostatistical modeling as it takes into account the parameter uncertainty during prediction. In this research we develop models in Bayesian paradigm considering different approach. To understand the potential of adding covariable in to modeling, a model was developed with adding CTM given covariable and another model developed with CTM covariable along with AOT data. The PM component (PM10) predicted with one model (RMSE = 0.5646) and the other (RMSE= 0.5632) shows similar value of RMSE. To incorporate PM components relationship; three models, namely, Model A, Model B and Model C were developed. Model A does not incorporate PM component relationship in to modeling and shows RMSE 0.6701. Model B incorporates the PM components relationship via adding prior knowledge about the parameter in modeling and as a result shows RMSE 0.6691. Model C incorporates PM relationship in to the mean of process as a covariable and gives RMSE 1.2676. Based on comparing the above mentioned models it was concluded that CTM and AOT both added strength in to modeling.Regarding the PM components relationship added in to modeling based on Model A, Model B and Model C we conclude that adding PM components into the mean of the process leads to a bias in prediction. Moreover, model B, which was developed with prior knowledge proved to be the most feasible approach with the least RMSE. Keywords Particulate Matter, Bayesian Hierarchical modeling, Multisource data, Geostatistical modeling
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page