Sources of Humic-Like Substances in the Pearl River Delta, China: Positive matrix 1 factorization analysis of PM 2.5 major components and source markers

12 HUmic-LIke Substances (HULIS), the hydrophobic part of water soluble organic carbon 13 (WSOC), account for a significant fraction of PM 2.5 mass. Their source studies are so far 14 largely qualitative. In this study, HULIS and WSOC were determined in 100 PM 2.5 samples 15 collected in 2009 at an urban site (Guangzhou) and a suburban site (Nansha) in the Pearl 16 River Delta in South China. The annual average concentration of HULIS was 4.83 and 4.71 17 μ g m -3 , constituting 8.5% and 10.2% of the PM 2.5 mass, while HULIS-C (the carbon 18 component of HULIS) contributed 48% and 57% of WSOC at the two sites, respectively. 19 HULIS was found to correlate with biomass burning (BB) tracers (i.e., levoglucosan and K) 20 and secondary species (e.g., SO 42- and NH 4+ ), suggesting its association with BB emissions 21 and secondary formation processes. Sources of HULIS were investigated using positive 22 matrix factorization analysis of PM 2.5 chemical composition data, including major 23 components and source markers. In addition to secondary formation process and BB 24 emissions, residual oil combustion related to shipping was identified for the first time as a 25 significant source of HULIS. Secondary formation process contributed the most, accounting 26 for 49-82% of ambient HULIS at the two sites


210
The concentrations of OC and WSOC were both higher at GZ than NS (Table 1). Annual The variation of WSOC_h and WISOC among different samples will be discussed later in this 228 paper (Sect 3.2.4).

229
Unlike OC and WSOC that exhibit a concentration gradient between GZ and NS, the 230 concentrations of HULIS were similar at both sites (Table 1) galactosan, were also higher on that day than all the other samples (Fig. S1). In addition, respectively) than "continental days" (222 and 179 ng m -3 in GZ and NS, respectively).

278
The obvious correlations of these three species (R 2 >0.80, Fig. S1) confirm similar sources of 279 the three isomers. HULIS, levoglucosan and sulfate (Fig. 3). The temporal variation trend of HULIS is roughly 290 similar to, but not exactly the same as, that of levoglucosan (Fig. 3). In winter, the trends of 291 levoglucosan and HULIS were similar; when levoglucosan increased, HULIS also increased,

302
The PMF analysis was based on the combined data set of 100 samples at GZ and NS. The where e ij is the residual of the concentration of j th species in the i th sample and s ij is the input 310 uncertainty of the j th species' concentration of the i th sample. IM indicates the least fit species.

311
If IM drops dramatically when the number of factors is increased by 1, it indicates that the 312 larger number of factors is more appropriate. For our data set, IM dropped dramatically when 313 the number of factors increased from 5 to 6, and dropped slightly when the factor number was 314 further increased from 6 to 9 (Fig. S2). Thus, the more suitable number of factors should be 315 higher than 5.

316
The interpretability of the source profile and explained variation (EV) was another  Table   336 S3. The modeled average concentrations of these major species deviate less than 7% from the

345
HULIS is present in three of the six fractors resolved by PMF, that is, secondary process, 346 biomass burning, and ship emissions and sea salt aerosols. The other three factors did not 347 contribute to HULIS. Table 2 shows the average factor contributions of HULIS-C. Figure 6   348 shows the spatial and temporal variation of individual factor contributions to HULIS-C.

349
Overall, secondary formation process was the most important source of HULIS  (Table S2). between HULIS-C and Na + was very weak (R 2 = 0.16) (Fig. 8). We note that the HULIS-C vs.

406
V correlation was nearly zero when the whole data set was considered, as contribution of  of different air masses (Fig. 7).

467
WISOC was apportioned to all factors resolved by PMF. The dust factor was a very minor 468 contributor (<3%). The contributions from the other five factors were roughly comparable on 469 'continental'/'transitional' days while more varied on 'marine' days ( Fig. 7). WISOC had 470 moderate correlations with EC, with R 2 = 0.51 at GZ and 0.74 at NS (Fig. S4), suggesting 471 primary combustion sources as the main suppliers of WISOC in PM 2.5 . We note that a sizable 472 portion of WISOC was apportioned to the Cland NO 3 dominated factor. We are unclear 473 about the underlying source or formation processes.

504
Stephen Griffith for editing the paper.