Abstract

Ambient VOC measured at a 1996 spring campaign at Santiago, Chile, have been analyzed using the receptor models UNMIX and Positive Matrix Factorization (PMF). The ambient campaign took place at two sites: a downtown site, close to major traffic emissions, and a residential site, downwind of major industrial sources and highways. At the downtown site the following source apportionments estimates were obtained: fuel evaporation, 29.7±5.6%; gasoline exhaust, 22.0±3.4%; diesel exhaust, 18.1±2.9%; biogenic, LPG and evaporative emissions, 18.0±3.4%, unexplained: 12%. At the residential site, the following source apportionment was obtained: transported gasoline exhaust, 31.2±4.1%; local gasoline exhaust, 25.5±4.0%; evaporative losses, 11.7±2.8%; LPG losses, 11.0±2.5%; biogenic emissions, 7.7±1.7%; diesel exhaust, 6.2±1.5%; unexplained, 7.7%. Thus, near 70% of ambient VOC impacts at both sites are due to mobile sources. The receptor analyses produced source profiles that had distinctive, dominant compounds; in addition, source contributions exhibited diurnal profiles that were consistent with ambient temperature and wind speed data, and the expected activity patterns within the city. Typical errors in the source contributions vary between 15% for the larger sources - like gasoline exhaust - and 25% for the smaller sources - like biogenic emissions. It was found that the number of factors given by the UNMIX model was a good starting point to refine the solution using PMF. Both models showed good performance at discriminating between source profiles that had similar compositions in subsets of common species, but PMF was able to find better, cleaner source profiles that did UNMIX. At both monitoring sites LPG losses appear mixed in with other source profiles, and this feature could not be solved by adding more source profiles in the analyses; this was likely due to a lack of C3 measurements needed to better resolve an LPG source profile. © 2004 Elsevier Ltd. All rights reserved.