About AAQR

Aims and Scope

Articles online
For contributors
Call for Papers
Guideline for the
Special Issue Proposal


Contact Us
Search for  in   Search  Advanced search  


Volume 14, No. 7, December 2014, Pages 2040-2050 PDF(199 KB)  
doi: 10.4209/aaqr.2014.01.0024   

A Comparison of Multiple Combined Models for Source Apportionment, Including the PCA/MLR-CMB, Unmix-CMB and PMF-CMB Models

Guo-Liang Shi1, Gui-Rong Liu1, Xing Peng1, Yi-Nan Wang2, Ying-Ze Tian1, Wei Wang2, Yin-Chang Feng1

1 State Environmental Protection Key Laboratory of Urban Ambient Air Particulate Matter, Pollution Prevention and Control, College of Environmental Science and Engineering, Nankai University, Tianjin, 300071, China
2 College of Software, Nankai University, No. 94 Weijin Road, Tianjin 300071, China


  • The accuracies of the predictions by different combined models were assessed.
  • Combined models can obtain reasonable results.
  • Fpeak rotation can affect the final result of PMF-CMB.
  • Marker species in the mixed source can affect the result of combined models.



A combined models was developed and applied to synthetic and ambient PM datasets in our prior works. In this study, multiple combined models, including the PCA/MLR-CMB, Unmix-CMB and PMF-CMB models, were developed and employed to analyzed the synthetic datasets, in order to understand 1) the accuracies of the predictions by multiple combined models; 2) the effect of Fpeak-rotation on the predictions of the PMF-CMB model; and 3) the relationship between the extracted mixed source profiles (in the first stage) and the final predictions. 50 predictions based on different combined model solutions were obtained and compared with the synthetic datasets. The average absolute errors (AAE), cluster analysis (CA), and PCA plots were applied to evaluate the precision of the predictions. These statistical methods showed that the predictions of the PCA/MLR-CMB and PMF-CMB model (with Fpeaks from 0 to 1.0) were satisfactory, those of the Unmix-CMB model were instable (some of them closely approached the synthetic values, while other them deviated from them). Additionally, it was found that the final source contributions had good correlation with their marker concentrations (obtained in the first stage), suggesting that the extracted profiles of the mixed sources can determine the final predictions of combined models.



Keywords: Receptor models; Synthetic datasets; Mixed source; Fpeak.



Copyright © 2009-2014 AAQR All right reserved.