An iterative method for evaluating the inter-comparability between chemical mass balance and multivariate receptor models

• A comparative study between CMB and PMF was conducted.
• Reactive species were apportioned by allowing CMB source profiles to vary (profile iteration).
• A process named Factor Mapping was developed for intercomparison between RMs.
• Profile iteration substantially improved the ability of CMB to reconstruct ambient data.
• Factor Mapping significantly assisted in the interpretation of PMF factors.

A comparative study between Chemical Mass Balance (CMB) and multivariate receptor modeling techniques was conducted. The study involved common application of Robotic Chemical Mass Balance (RCMB) and 2-dimensional Positive Matrix Factorization (PMF). A two-fold methodology was developed for Source Apportionment (SA) of reactive species, in order to address previous limitations of CMB models. The developed methodology (a) uses a detailed set of theoretical source profiles, taking into account secondary reactions that were not considered in CMB modelling until now, and (b) implements a Least Squares (LS) fitting method that iteratively readjusts the values of independent variables in the CMB fit, providing (for the first time, to our knowledge) CMB source profiles as output data, in which secondary transformations may be reflected. A straight-forward computational procedure named Factor Mapping (FM) was developed as well, for intercomparison between RCMB and PMF. The distinctive feature of FM is that similarity measures are used not just to compare the results of the two models, as in previous intercomparison exercises, but, moreover, to actively assist in the physical interpretation of PMF factors, thus minimizing user interference. The intercomparison between RCMB and PMF also involved an independent evaluation of each model's performance in reproducing ambient concentrations of particulate matter (PM) and associated chemical constituents. Overall, it was shown that the employed series of computational steps substantially improve qualitative as well as quantitative agreement between the two models.