Development and application of a protocol for source apportionment of ambient volatile organic compounds in urban/industrial areas

Resumen

Air pollution is a major environmental health problem and also an important issue forthe economy, the environment and the global climate. Reliable and quantitativeinformation on pollution sources is essential for the implementation and evaluation ofair quality abatement strategies.Receptor models are mathematical procedures widely used in air quality managementto identify sources of pollution and quantify their contribution to the ambientpollutant concentration measured at a given receptor location by solving a massbalance equation. However, there is still a lack of commonly accepted procedures andcriteria for the application of these kinds of models to source apportionment studiesand obtained results are noticeably influenced by expert subjectivity.In this work, a standard protocol to apportion and locate sources of ambient volatileorganic compounds (VOC) is developed and then tested, based on the application ofreceptor models. The aim is to ensure quality standards of the results obtained insource apportionment studies and to improve their comparability, as a critical steptowards enabling efficient pollution control strategies. The most widespread sourceapportionment methodologies and auxiliary techniques and their respective datarequirements are reviewed to define the main steps of the protocol: 1) ambient datacollection, 2) conceptual model formulation, 3) database analysis, 4) development ofreceptor model input dataset, 5) receptor model application and 6) modelperformance evaluation.Once the protocol has been developed, it is tested through its application to 64 hourlynonpolar VOC measured at two urban/industrial sites, Zorroza and Zubileta, from July2010 to July 2011 over different periods, for the most common situation in whichaccurate information on pollution sources and speciated composition of theiremissions is not available. In the study area there are a large and varied number of2VOC emitting sources (e.g., a crude oil refinery, a coke oven, busy highways, populatedurban areas, etc).VOC samples are taken using two comparable thermal desorption systems andanalysed using similar automated gas chromatographs with twin flame ionizationdetectors. Both systems are optimized for the automatic measurement of hourly 64nonpolar C2¿C10 nonmethane VOC in urban air, with sensitivities down to parts pertrillion in volume. An optimized validation methodology is followed to handle, processand evaluate the large volume of chromatographic data so as to obtain a reliable andconsistent VOC dataset. Additionally, hourly conventional pollutants andmeteorological parameter data are also collected to later assist in sourceidentification.Previously to apply receptor models, temporal patterns and correlations of VOCspecies concentration are analysed together with wind speed and directiondependence to get a preliminary explanation of possible sources that impact thereceptor sites, the timing of emissions and the meteorology that affects receptorconcentrations. From this analysis it becomes clear that meteorology stronglyinfluences observed VOC concentrations and that industrial emissions enriched inethane, ethene, benzene and naphthalene, most likely originating from the coke oven,have a great impact both at the Zorroza and Zubileta sites. Road traffic is alsoidentified as a probable major VOC source at both sites. Other minor sources may berelated to solvent and coating use and even the influence of the emissions from thecrude oil refinery are likely to be noticeable at the Zorroza station.Once a preliminary explanation of the sources is obtained, Unmix and PMF are appliedsequentially, in a combined manner. First, Unmix is applied in order to estimate aminimum number of factors and to obtain preliminary source profiles. This helps todefine a starting point for the PMF modelling, which can provide a more detailedsource apportionment. Then, the PMF model is executed several times until an optimalsolution is found. The resolved factors are identified by comparing their compositionwith known source emission profiles and by analysing the temporal patterns and the3bivariate polar plots of the modelled contributions. Finally, the obtained solution isvalidated by examining its stability and its rotational ambiguity.Applying this novel protocol, nine sources are extracted at the Zorroza site and fivesources at the Zubileta site. At both sites similar VOC sources are identified, related tocoke oven, road traffic, solvent/coating use and biogenic emissions, being the Unmixand PMF results quite analogous. For Zorroza data, 9 VOC sources are identified: rawcoke oven gas (23%), natural gas (17%), high aromatic events due to road traffic andsolvent/coating use (15%), coke oven combustion gases (12%), urban road traffic(10%), mixed industrial and regional traffic (9%), crude oil refinery (6%),solvent/coating use (5%) and isoprene emissions (3%). At Zubileta, 7 major sources areidentified: raw coke oven gas (41%), pollutant accumulation due to adversemeteorology (17%), combustion gases from coke oven and motorway traffic exhausts(17%), high aromatic events due to road traffic, coal pile evaporative emissions andsolvent/coating use (12%), fuel evaporation (10%) and isoprene emissions (3%). Thevariability and rotational analyses suggest that the identified sources accuratelyrepresent the ambient VOC data at the two monitoring sites.