Week-long samples of airborne particulate matter were obtained at Alert, Northwest Territories, Canada, between 1980 and 1991. The concentrations of 24 particulate constituents have some strong, persistent seasonal variations that depend on the transport from their sources. In order to explore the nature of the cyclical variation of the different processes that give rise to the measured concentrations, the observations were arranged into both a two-way matrix and a three-way data array. For the latter the three modes consist of chemical constituents, weeks within a year, and years. The two-way bilinear model and a three-way trilinear model were used to fit the data and a new data analysis technique, positive matrix factorization (PMF), has been used to obtain the solutions. PMF utilizes the error estimates of the observations to provide an optimal pointwise scaling data array far weighting, which enables it to handle missing data, a common occurrence in environmental measurements. It can also apply nonnegative constraints to the factors. Five factors have been obtained that reproduce the data quite well for both two-way and three-way analyses. Each factor represents a probable source with a compositional profile and distinctive seasonal variations. Specifically, there are (i) an acid photochemical factor typified by Br-, H+, and SO42- and characterized by a concentration maximum around April, or shortly after polar sunrise; (ii) a soil factor representing by Si, Al, and Ca and having its main seasonal maximum in September and October; (iii) an anthropogenic factor dominated by SO42- together with metallic species like Pb, Zn, V, As, Sb, Se, In, etc., peaking from December to April; (iv) a sea salt factor consisting mainly of Cl, Na, and K with maximum concentrations during the period from October to April: and (v) a biogenic factor characterized by methane-sulfonate and having a primary maximum at May and a secondary maximum in August. The results obtained by both two-way and three-way PMF analyses are generally consistent with one another. However, there are differences because of additional constraints on the solution imposed by the three-way analysis. The results also help to confirm the hypotheses regarding the origins of the Arctic aerosol.
Samples of airborne particulate matter were collected over a continuous sequence of 1 week intervals at Alert, Canada beginning in 1980 and analyzed for a number of chemical species, It was found that the measured weekly average concentrations display strong, persistent seasonal variations. In another recent study, the measured concentration of 24 constituents were arranged into both 2-way and 3-way data arrays and bilinear and trilinear models were used to fit the data using a new mathematical technique, positive matrix factorization. Five factors were found to explain the data for both 2-way and 3-way modeling with each factor representing a likely particle source. In the 2-way modeling, the yearly cyclical seasonal variations were not directly retrieved since the whole 11 yr of data was regarded as a single mode in the fitting. In the 3-way analysis, assuming the week-to-week patterns of the source contributions recur from year to year imposed fixed seasonality on the solutions. The resulting fit becomes worse if the year-to-year pattern of variation is not identical for any given source. These results suggested that a mixed model containing both 2-way and 3-way components might provide the best representation of the data. The methodology to calculate such a mixed model has just been developed. The multilinear engine is introduced in this study to estimate a mixed 2-way/3-way model for the Alert aerosol data. Five 2-way and two S-way factors have been found to provide the best fit and interpretation of the data. Each factor represented probable source with a distinctive compositional profile and seasonal variations. The five 2-way factors are (i) winter Arctic haze dominated by SO42- including metallic species with highest concentrations from December to April, (ii) soil represented by Si, Al, Ca, (iii) sea salt, (iv) sulfate with high acidity peaking in late March and April and (v) iodine representing most of the observed I with two maximal one around September and October and another around March and April. The two 3-way factors are (i) bromine characterized by a maximum in the spring around March and April; and (ii) biogenic sulfur which includes sulfate and methanesulfonate with maxims in May and August. The acidic sulfate, bromine, and iodine factors have a common maximum around March/April, just after polar sunrise, suggesting the influence of increased photochemistry at that time of year. The strength of the year-to-year biogenic sulfur factor showed a moderate correlation (r(2) = 0.5) with the yearly average Northern Hemisphere Temperature Anomaly suggesting a relationship of temperature with biogenic sulfur production. The results obtained are consistent with those obtained in the previous study and agree with the current understanding of the Arctic aerosol. (C) 1999 Elsevier Science Ltd. All rights reserved.
The chemical composition of particles collected at Alert, Northwest Territories, Canada, show strong, persistent seasonal variations. In a previous study, a 2-way/3-way mixed factor model was performed on the weekly average concentrations of 24 aerosol components measured over the period from 1980 to 1991. The Multilinear Engine (ME), a new mathematical technique, was used to obtain the solution. The two modes of the 2-way model consist of the source composition profiles and mass contributions over the 11 yr, while for the three modes of the 3-way model, source profiles, mass contributions variations over the weeks within a year, and the year-to-year variation over the 11 yr within the measurement period. Five 2-way and two 3-way factors were found to provide a good fit to the data and were easily interpreted. In this investigation, potential source contribution function (PSCF) analysis was applied to the source contributions derived from the ME analysis by incorporating meteorological information in the form of 5-d air parcel back trajectories. The potential locations and/or the preferred pathways of these possible sources were then determined by the PSCF analysis. (C) 1999 Elsevier Science Ltd. All rights reserved.