The use of positive matrix factorization (PMF) in the source apportionment of ambient aerosol in the Central Mediterranean

Abstract

Positive Matrix Factorization (PMF) is a widely used receptor modelling technique in order to determine the nature and contribution of the different aerosol sources modulating ambient levels of particulate matter at a receptor. This cumulative thesis together with the papers included within, reports the results of three source apportionment exercises: a) the isolation of the natural contribution to PM10 levels at a rural background site in Malta; b) the estimation of the contributions of the sources driving PM2.5 levels at a traffic hotspot in Malta and c) a methodological investigation of how PMF can be used on a smaller dataset using samples taken from an agricultural area in Apulia, South eastern Italy. The study on the magnitude of the natural contribution to PM10 involved a sampling campaign at a rural background station in Għarb, Gozo (one of the islands in the Maltese archipelago). This sampling campaign resulted in the collection 224 samples, which were subsequently characterised by inductively coupled plasma mass spectrometry (ICP – MS) and ion chromatography (IC) respectively for, their metallic and ionic content. The use of data resulting from this analysis with PMF resulted in the isolation of the two natural sources expected to be of relevance to Malta (marine aerosol and Saharan dust) as well as in the estimation of their apportionment. PMF also resolved three additional sources: a local crustal source, and two secondary inorganic aerosol components (one rich in nitrate and the other rich in sulfate). The natural sources jointly accounted for approximately 39% of the measured PM10, which is at the upper end of the 0.5 – 58% range determined by previous studies for natural contributions in Europe. A total of 180 membranes sampled throughout 2016 were used in the study on the sources of PM2.5 at a traffic site. These membranes were analysed for: elemental concentrations (using X-ray fluorescence spectroscopy, XRF); ionic content (using IC) and for black carbon – BC (using an optical method). The use of this chemical database with PMF resulted in the isolation of 7 aerosol sources, 4 of which were common to the exercise carried out on PM10 at the rural background site (all the sources except for the local crustal source). The additional three sources isolated at this site were traffic, shipping and fireworks. The isolation of the latter component is itself an interesting result, because it shows that a seasonal activity such as the letting of fireworks during the summer village feasts affects the annual levels of PM2.5. Additionally, this component probably has an effect on human health due to its chemical composition. This work will also provide evidence-based information to the policy makers on the emission reductions required in order for the PM2.5 levels to be compliant with the annual air quality guideline issued by the World Health Organization. Finally, a fundamental methodological investigation on how PMF can be used on a small dataset was carried out. This study is based on 29 PM10 and 33 PM2.5 samples collected from a rural area in Apulia, Italy. PMF did not work correctly when the datasets for the two different fractions were used separately. The datasets were therefore aggregated into a single chemical database of 62 samples and this was then used with PMF. A 5-factor model, which exhibited a fairly good rotational stability was the result of this modelling exercise. This was subsequently further improved through the imposition of constraints based on the chemical constitution of the aerosol sources affecting this receptor, which is made possible by the new features included in the United States Environment Protection Agency PMF version 5. Given the size of the dataset the, the uncertainties in the solution returned by PMF were fully characterised using all the error estimation methodologies included in this version of PMF. Additionally, the results of the PMF modelling were validated against those returned by two other models, Constrained Weighted Non-negative Matrix Factorization (CW – NMF) and Chemical Mass Balance (CMB) as well as through the use of other statistical parameters. These results essentially confirm the validity of the model returned by PMF and indicate that the latter model extracted all the information about the aerosol sources affecting the receptor from the speciation data.