Stability of a PMF derived source apportionment solution using a smaller dataset

Abstract

The objective of this study was to investigate how a meaningful PMF solution can be obtained with smaller datasets using the features included in USEPA version 5. In all 29 PM10 and 33 PM2.5 samples were collected from a rural setup in Apulia. The filters were characterised chemically. Running the model with the PM10 and PM2.5 samples separately resulted in the model not working correctly. The datasets for the two PM fractions were therefore aggregated in a single dataset made up of 62 samples. PMF isolated 5 different factors. The solution returned by PMF was validated against the solutions returned by two other models CW-NMF and CMB. The daily samples reconstructed by the three models were compared to each other using Pearson correlation and the source contribution estimates returned by the three models were also compared with modelling the Total traffic factor in PM10 and PM2.5 and marine aerosol in PM2.5. Figure 1 shows that the source contribution estimates of PMF and CW-NMF are within the respective error bars, while it shows that CMB has some problems with modelling secondary aerosol factors, this is a well-known issue with CMB. The comparison between PMF and CMB is not always straight forward and alignment is normally obtained after the profiles inputted into CMB are tweaked using the profiles returned by PMF. In literature one does find cases in which the daily source contributions of these two models do not correlate well e.g. Rizzo and Scheff. We however believe these mismatches do not however undermine the model returned by PMF, therefore showing that it is possible to extract useful information using smaller datasets too.