+1 (202) 780-3397
Comparative evaluation of chemical mass balance and multivariate receptor models using synthetic data
7th World Congress on Chemistry
November 13-15, 2017 Athens, Greece
Georgios Argyropoulos and C Samara
Aristotle University of Thessaloniki, Greece
Posters & Accepted Abstracts : J Biotech and Phyto
Abstract:
In the field of atmospheric sciences, the term receptor models is customarily used to describe top-down approaches for airpollution assessment, i.e. methods that begin by sampling air in a given area, in order to match common chemical and physical characteristics between source and air pollution samples. Source identification and quantification is realized by employing statistical analyses, widely known, under the term Multiple Linear Regression (MLR). There are two fundamental categories of RMs, Chemical Mass Balance (CMB) models, which assume full knowledge of the compositions of emissions, and multivariate models, which apportion sources on the basis of observations at the receptor site, alone. One of the first documented uses of RMs for air-quality management was by the United States Environmental Protection Agency (EPA), back in the early 80s. Since then, RMs have gradually become familiar to policymakers all over the world, as there were vast improvements, not only in the MLR methods that can now be performed by modern computers but also in the chemical speciation techniques that provide RMs with input data. Nevertheless, there are still major concerns regarding RMs, such as the influence of personal judgment to model results, as well as the lack of a standard methodology for quantifying uncertainty levels, since, in the real world, one cannot check the model output, against the actual values of source contributions. This study presents a comparative evaluation of RMs, using synthetic input datasets, i.e. where the values of source contributions are already known. Synthetic data were generated by inducing random variations to reference values, with the use of deterministic procedures, widely known as “pseudo-random number generators”. Virtual receptors have been set to match conditions that can actually hinder model performance, such as large measurement errors, collinearity between source profiles, strong correlations between the temporal variations of source contributions etc. The simulation includes the newest versions of CMB and multivariate receptor models, as well as some of the previous ones that are still in use, by the scientific community. Particular emphasis has been placed on a recently developed computational procedure, the so-called Robotic Chemical Mass Balance (RCMB), which has been considered to be a mathematical optimization of previous CMB models, minimizing personal judgment. Preliminary results indeed confirm the superiority of RCMB over the human modeler, if the latter one has under or overdetermined source profile input data.
Biography:
Georgios Argyropoulos is a postdoctoral researcher in the Department of Chemistry, at the Aristotle University of Thessaloniki (AUTH), Greece. His educational background includes an MSc degree in Chemical Engineering, an MSc degree in Environmental Chemistry, and a PhD degree in Receptor Modeling, all received from AUTH. One of his major research interests is the use of statistical techniques, such as multivariate analysis, for source apportionment of atmospheric pollutants. He has participated in numerous research projects, including the LIFE Environment programme, funded by the European Commission. Recently, he also received a Fellowship of Excellence for Postgraduate Studies in Greece, from the State Scholarships Foundation (IKY) of Greece, in the framework of the Hellenic Republic – Siemens Settlement Agreement.
PDF HTML