Chemical characterization of PM 1.0 aerosol in Delhi and source apportionment using positive matrix factorization
- PDF / 7,288,788 Bytes
- 18 Pages / 595.276 x 790.866 pts Page_size
- 84 Downloads / 185 Views
RESEARCH ARTICLE
Chemical characterization of PM1.0 aerosol in Delhi and source apportionment using positive matrix factorization Jaiprakash 1 & Amrita Singhai 1 & Gazala Habib 1 Tarun Gupta 3
&
Ramya Sunder Raman 2 &
Received: 25 April 2016 / Accepted: 13 September 2016 # Springer-Verlag Berlin Heidelberg 2016
Abstract Fine aerosol fraction (particulate matter with aerodynamic diameter 2 were identified as strong species, whereas species with moderate S/N ratio between 0.2 and 2.0 were termed as weak. Further, the uncertainties of weak species were increased three times as suggested by Paatero and Hopke (2003). The S/N ratios for most of the species measured here were greater than 2.0 except Cr and Ca (Table 1).
Application of positive matrix factorization The model was run with input data for a number of factors from 2 to 10, and the solutions associated with the minimum value of objective function Q(E) as described in section A3 in the SI and shown in Fig. S2a in the SI were examined. The solution for which the modeled Q(E) was close to theoretical Q(E) value, for example, six factor solution in the present case was used as a starting point. In addition to this the Individual column Mean (IM), and Individual column Standard Deviation (IS) of scaled residual matrix showed a decrease in low rate or flattened after 6-factors solution (Fig. S2b in the SI). However, 5- and 7-factor solutions were also examined to confirm the physically interpretable solution. Therefore, in the present work, 5-, 6-, and 7-factor solutions were further analyzed by externally regressing the model apportioned factors contribution against measured PM1.0 mass to derive the scaling coefficients. The positive scaling coefficients were obtained suggesting the 5- or 6-, or 7-factor solution may lead to an acceptable solution. The regression coefficients were multiplied with source contribution matrix following literature (Kim and Hopke 2004a, b; Sunder Raman et al. 2010), and the sum was considered as model-predicted PM1.0 mass. The modelpredicted PM1.0 mass was not well co-related with the measured PM1.0 (r2 = 0.60). Thus, PMF-2 was run including missing mass (see Eq. (4)) in addition to the 15 chemical species, as model input. Again, the multiple linear regression after including missing mass as a model input yielded positive scaling coefficients when regressed externally with measured PM1.0 mass. Now, the predicted and measured PM1.0 were well corelated (slope 0.77, r2 = 0.73) for the 6-factor solution (Fig. S3, in the SI). For 5- and 7-factor solutions, the coefficient of determination (r2) between predicted and measured PM1.0 mass was moderate as 0.63 and 0.67, respectively. The factor profiles for 5-, 6-, and 7-factor solutions were also examined and the mixing of sources in the 5-factor solution (e.g. biomass burning and soil dust) and splitting of soil dust in the 7-factor solution were observed. However, the 6-factor solution was physically interpretable. Therefore, the 6-factor solution was chosen and further analyzed. Like all
Data Loading...
