Utility of synthetic datasets to refine the particular matter source profiles for chemical mass balance analysis

Source apportionment of particulate matter samples using the chemical mass balance (CMB) approach is undertaken from a known set of information regarding chemical species present in the sources.The chemical species data pertaining to a particular source often would contain interfering chemical species originating from other sources due to the mixed state of its presence in the actual state.Such mixed source profiles, when used in the CMB analysis, result in collinearity and induce several errors in source apportionment analysis.To ensure the reliability of the source apportionment process, collinearity is to be eliminated by refining the source profiles used in the source apportionment process.The utility of synthetic datasets to refine the source profile in a source apportionment process is demonstrated for the particulate matter sampled from a non-urban region in the South Indian state of Kerala.To bring out cleaner source profiles, a large sample of receptor datasets that closely resemble the field data is generated using the artificial neural network module conditional tabular generative adversarial network(CT-GAN) and principal component analysis/multi linear regression(PCA/MLR) approaches are further applied to isolate various combined sources, and the source profile information is refined using published information.The effectiveness of the refined source profile information in a CMB analysis is demonstrated using the field receptor data.