A Factor and Trends Analysis of Multidecadal Lower Tropospheric Observations of Arctic Aerosol Composition, Black Carbon, Ozone, and Mercury at Alert, Canada

Observations from 1980 to 2013 of 20 aerosol constituents, ozone and mercury at Alert, Canada (82.50°N, 62.35°W), were analyzed for trends and dominant factors of the Arctic haze during winter and spring. Trends reflect changing emissions in Eurasia, the main source region for surface pollution in the high Arctic. SO42−, H+, NH4+, K+, Cu, Ni, Pb, Zn, nonsoil V, nonsoil Mn, and equivalent black carbon decreased between 23% and 80% as emissions declined rapidly in northern Eurasia during the early 1990s. NO3− increased by 20% as aerosol acidity declined. Metals were linked to emissions from smelting and fossil fuel combustion. In winter, ozone increased by 5% over 23 years, consistent with other observations and global modeling. Twelve PMF factors emerged for the dark period (November to February) and 13 for the light period (March to May). Eleven PMF factors are common to both dark and light, a twelfth factor was associated with sulfate in the dark and nitrate in the light, and the thirteenth (light period) was related to ozone and gaseous mercury depletion near Alert. IODINE and NITRATE factors, important for Arctic chemistry, changed with sunlight. In the light, 50% of all NO3− was on the NITRATE factor, while in the dark, most was associated with MODIFIED SEA SALT and equivalent black carbon. In the dark (light), 90% (28%) of iodine were found on the factor IODINE and 58% associated with SEA‐SALT and MODIFIED SEA‐SALT. These results help in understanding the role of atmospheric chemistry in weather and climate processes.


Introduction
Anthropogenic aerosol constituents and gases in the Arctic lower troposphere are most abundant in the colder half of the year (November to May). Lower wet and dry removal in winter/spring relative to summer results in longer aerosol lifetimes (Barrie, 1986;Garrett et al., 2011;Quinn et al., 2007;Shaw, 1995). Arctic air pollution near the surface originates mainly from northern Europe and northern Asia Christensen, 1997;Stohl, 2006;Hole et al., 2009).
Since 1980, Alert observations have been a continuous thread supporting, complementing, and connecting many Arctic air chemistry process studies. In the 1980s, analysis of lead isotopes in Alert aerosols showed that Eurasian pollution was the main source of Arctic haze (Sturges & Barrie, 1989) and confirmed atmospheric chemistry transport modeling apportionment of lead to sources . The discovery that particulate bromine and iodine peaked in April to May just after polar sunrise (Sturges & Barrie, 1986) linked lower tropospheric ozone depletion events observed at Alert and at Utqiagvik, Alaska (Oltmans & Komhyr, 1986), with heterogeneous snow/air photochemistry involving aerosol bromine (Barrie et al., 1988). This led to research on heterogeneous chemistry involving release of reactive atmospheric gases from salt-laden surface snow in sunlight. Photochemically induced heterogeneous reactions in the surface snowpack that release gases such as NO to the atmosphere were then discovered on the Greenland ice cap at Summit Greenland (Honrath et al., 1999) and at elevation on the Antarctic ice sheet near South Pole station (Davis et al., 2001). It sparked a broad area of international research on the role of ice surfaces in atmospheric and environmental chemistry (Grannas et al., 2007).
In the late 1980s and 1990s, ozone depletion/aerosol Br production events were connected with production of oxygenated organic aerosols (Kawamura et al., 1995(Kawamura et al., , 1996 and with depletion of atmospheric gaseous elemental mercury (GEM) (Lu et al., 2001;Schroeder et al., 1998bSchroeder et al., , 1998c. At the end of the 1990s, sufficient observations were accumulated to define the trends of 18 Arctic aerosol constituents over a 15-year period and to identify 10 distinct aerosol factors . This revealed that in winter/spring, high Arctic sulfate aerosols were over 86% anthropogenic in origin with sea salt and marine biogenic sulfur sources contributing the rest. This was confirmed independently by an analysis of sulfur isotopes in sulfate aerosols (Norman et al., 1999). Alert sulfate data played a central role in an international comparison of climate/chemistry models simulating the transport of aerosols to the high Arctic from midlatitudinal sources . Sharma et al. (2004Sharma et al. ( , 2006Sharma et al. ( , 2013Sharma et al. ( , 2017 observed and investigated the cause of a 50% decline in equivalent black carbon (EBC) at Arctic surface stations since 1990. Recently, black carbon was isotopically apportioned to anthropogenic and biomass burning sources (Winiger et al., 2019).
Other studies expanded the existing knowledge of the summertime Arctic aerosol and the role of natural marine sources of particles in climate with a particular focus on organic components of the aerosol (e.g., Abbatt et al., 2018;Croft et al., 2016;Leaitch et al., 2013;Leaitch et al., 2018;Mungall et al., 2017;Sharma et al., 2012). Alert measurements were a vital part of Arctic Council assessments of impacts of black carbon on Arctic climate in support of policy development (AMAP, 2015) and Canadian Network on Climate and Aerosols: Addressing Key Uncertainties in Remote Canadian Environments . This paper focused on 34 years of aerosol chemistry and gas measurements conducted at Alert, Nunavut, Canada (82.50°N, 62.35°W), began in 1980 (Barrie et al., 1981). In the 1990s, atmospheric ozone, aerosol absorption (i.e., black carbon) and mercury measurements were added. More recently, the observations were supplemented with submicrometer filter sampling and aerosol physical measurements (Leaitch et al., 2013Schmeisser et al., 2017).
Here we examined the trends and sources of 19 aerosol constituents plus EBC, ozone, and GEM over 34 years in the most polluted period of the year. For the period of 1995-2013, positive matrix factor analysis (PMF) was applied to reveal sources of particles during the dark winter (November-February) and the sunlit early spring period (March-May). The present study complements and extends the PMF analysis of Sirois and Barrie (1999) that only focused on winter/spring data from 1980-1995. The present study also gave additional insights into atmospheric composition during the dark and light periods.

Materials and Methodology
Details of the aerosol chemistry filter collection, measurement, and data analysis methodologies were given in Sirois and Barrie (1999). They did not change over the decades although chemical analysis instrumentation improved in sophistication, sensitivity, and ease of use. A summary of methodology is given in this section. in the light intensity corresponded to a noise level of 2 ng·m −3 and a detection limit of 4 ng·m −3 for 1-hr integration time (Backman et al., 2017).

Ozone
Surface ozone by Environment and Climate Change was sampled through a filtered inlet at a height of 5 m and measured using the technique of ultraviolet absorption using Thermo models 49 (1992-2003) and 49C (2003-2013). All measurements were referenced to a NIST primary reference photometer. Similar to filter pack times, weekly averages of ozone were calculated and used for the trend analysis.

Methodology for Time Series Analysis
The detailed description of the analysis (Magnusson, 2016) is summarized in this section. The time series model was used: where C t is the observed weekly concentration; M t , the long-term trend factor; S t , the seasonal factor; and Y t , a noise factor (Brockwell & Davis, 2002). Many variables in the observational data set were lognormally distributed or the variance grows increasingly with the amplitude of the data. In these cases, a temporary transformation of the concentration (log (Ct) or √ (Ct)) was made in order to have fluctuations not dependent on the level of the concentration series. Following Brockwell and Davis (2002), first, the long-term trend was identified and removed using cubic smoothing splines. If no long-term trend was found, the mean of the concentration series was removed in order to center it. Then, a seasonal trend was removed to obtain weakly stationary residuals before fitting an autoregressive moving average model to Y t .
Two types of data were analyzed from the data set: (i) all-year data and (ii) a data subset for the winter/spring part of the year when air pollution aerosols are most abundant and the Arctic is best connected meteorologically to the south.
Generally, missing values were not a large problem since they were spread out so that not more than a couple of weeks were missing in a row. This does not affect the long-term trends notably, with the exceptions of H + , methane sulfonic acid (MSA), I, and gaseous O 3 measurements, which had larger gaps. Retrieving trends for the missing years was not possible, and the spline fit was done with the missing values removed.
For all-year data, log (C t ) transformations were needed for all constituents except for ozone and mercury to obtain normally distributed residuals. After the long-term trends M t were removed from the observed respective constituent time series, we considered the seasonal variation S t . There exists strong seasonality in the observed aerosol constituents. Using the weekly long-term detrended data, it was possible to estimate seasonality variations using smoothing splines similar to those used for the long-term trends where we set the degrees of freedom to adjust the amount of smoothing. Ten degrees of freedom of the spline was used for all seasonal trends.
As all-year time series analysis (data not shown here) included a large number of below detection limit values, especially during summer, strong seasonality and varying variance motivated the decision to reduce the data set and only look at observations for the more polluted winter half of the year. January to March data were used for all variables. Excess Vanadium (xV) and excess Manganese (xMn) were the V and Mn associated with nonsoil (in excess of average crustal composition), calculated as xV ¼ V−0:0013*Al; xMn ¼ Mn−0:0065*Al: 15 years of data from November to May. In contrast, this extended winter data set required log transformations for all metals (Zn, Pb, Fe, and Ni) and the alkali metal K + . For xV and xMn the transformation of √(C t ) was made in order to obtain normally distributed residuals.

Seasonal Variation
Seasonal cycles for the 22 variables, in Figure 1, showed maximum for all variables except for Al, MSA, I, and GEM during the Arctic haze period, from November to May, when strong transport from the south occurred at a time of minimal wet and dry removal from the Arctic air mass (Barrie, 1986;Garrett et al., 2011). Excluding Al, MSA, I, and GEM, the timings of the spring-to-summer and summer-to-fall transitions were not identical. For example, similar to organic material , the maximum in NO 3 − was in May, while the spring maximum in SO 4 −2 occurred in April. Also, the summer-to-fall increase in SO 4 −2 began in August, about 1 month before NO 3 − started to increase. The seasonal variation in natural aerosol constituents (MSA, I, and Al) with two seasonal peaks would be discussed in detail in section 3.2.5.
Ozone and GEM showed different seasonal trends relative to the other components. GEM and ozone depletion events occurred at Alert concurrently after polar sunrise (beginning of March at Alert) until early May (Lu et al., 2001;Schroeder et al., 1998b;Steffen et al., 2005Steffen et al., , 2014. Depletion was driven by halogens Br and Cl mobilized from sea salt in the frozen Arctic snowpack after polar sunrise. It involved complete destruction of ozone in the surface-based inversion layer on the Arctic Ocean near Alert within a few days to a week (Hopper et al., 1998). Complete conversion of GEM to reactive gas-phase mercury and particulate mercury took place at the same time. In the absence of halogens, GEM had a lifetime in the free troposphere of 6 to 12 months (Steffen et al., 2008).  Figure 2n) that ended in 1997. The trends for EBC, GEM, and ozone (Figures 2r, 2s, and 2t, respectively) covered 1990 to 2013. The differences in January-March and January-April concentrations between the first 5 years and the last 5 years, given in the top right corner, are significant at p < 0.01. Trends of MSA and I, which had two peaks annually are discussed in section 3.2.6.
Constituents were divisible into two groups based on the differences. The first group (SO 4 2− , H + , NH 4 + , K + , Br − , Cu, Ni, Pb, Zn, xV, xMn, Ca, Mg, Al, GEM, and EBC) decreased over the observational period between 23% and 80%, while the second group (NO 3 − , Na + , Cl − , and O 3 ) increased between 5% and 43%. Recent studies suggested that these largely reflected changes in Northern Hemispheric emissions to the atmosphere rather than changes in transport during the January to March period . Declines in the anthropogenic variables are consistent with shifting economies and emission technologies driven by political changes that took place in Europe and the Former Soviet Union (FSU) during the 1990s. The marked drop in the early 1990s of atmospheric EBC, SO 4 2− , NH 4 + , xV, and Pb at Alert was likely due to a decrease in industrial activity in Russia and Eastern Europe. Emissions from smelters and coal combustion at Norilsk and the Kola Peninsula would explain declining trends in metals. Other studies also connected trends in anthropogenic aerosols to shifting emissions in the Eurasian source regions (Hole et al., 2009;Sharma et al., 2004Sharma et al., , 2006Sharma et al., , 2013. Higher sea salt contributions from January to March (i.e., Na + and Cl − ) can originate from sea salt spray from stormy northern seas or from wind-driven resuspension of snowpack sea salt or open leads in the frozen Arctic Ocean. Based on chemical-transport-model simulations, Huang and Jaeglé (2017) suggested that wind-driven resuspension of sea salt deposited on surface snow was the dominant source of wintertime sea salt at Alert, more important than frost flowers. A minor role for frost flowers was supported by the sulfur isotope analysis of Seguin et al. (2014), although it had been suggested otherwise (Xu et al., 2016). Xie et al. (1999) suggested sea spray transported from the North Atlantic Ocean and Bering Sea as the explanation for the higher sea salt aerosol concentrations at Alert. Their analysis, based on potential source contribution analysis using air parcel back trajectories, also found the dominant source of methane sulfonic acid (MSA) at Alert to be the North Atlantic Ocean, consistent with a similar approach that associated oceanic biological activity with MSA (Li et al., 1993;Sharma et al., 2012). As Willis et al. (2018) pointed out that the relative importance of open ocean seaspray sources versus windblown snow sources in determining aerosol sea salt requires further investigation. Kirpes et al. (2019) used scanning electron microscopy with energy dispersive X-ray spectroscopy to analyze 1691 individual particles in size ranges 0.32-0.56 and 1.0-1.8 μm aerodynamic diameter collected on the north slope of Alaska at Utqiagvikon January 24-27 and February 26-27, 2014. Winds were > 4 m s −1 and originated from the nearby Arctic Ocean. Sea salt particles produced locally from open sea ice fractures (leads) were dominant in these samples. The sea salt aerosol composition was not consistent with frost flowers or surface snow above sea ice. This suggests that there is a need for inclusion of lead-based sea salt production in modelling of Arctic atmospheric aerosols.
From 1990 to 2013, there was a 5% (2% per decade) increase (p < 0.001) in lower tropospheric ozone during dark winter (Figure 2t) at Alert. On average, it ranged from a 5-year mean of 32 ± 0.49 at the beginning to 35 ± 0.55 ppbv at the end. Beryllium isotope observations indicated that surface ozone at Alert was mainly tropospheric rather than stratospheric in origin (Dibb et al., 1994;Willis et al., 2018). The changes in ozone at Alert were consistent with a general increase of Northern Hemispheric free tropospheric ozone over the twentieth century due to increasing anthropogenic emissions of ozone precursors (e.g., NO x and volatile organic compounds) in North America and East Asia (Zhao et al., 2013). The influence of East Asia on surface O 3 at Alert might be via higher altitude transport followed by subsidence (Willis et al., 2018(Willis et al., , 2019. The lifetime of free tropospheric ozone in the dark Arctic winter was longer than an ensemble ) for the peak polluted winter months January-March. Aerosol concentrations are plotted on a logarithmic scale and long-term curves (in red) are splines fit to the observations after removal of the seasonal variation. The percentage shown is the change in the first 5 years relative to the last 5 years. mean of 22.3 days derived from global air chemistry models (Young et al., 2013). Surface ozone trends in the Northern Hemisphere varied by region (Cooper et al., 2014), but in the region of 40°N to 90°N mean ozone levels from the surface to 700 hPa (about 3 km) were generally increasing. An ensemble of global air chemistry models showed a 5-to 10-ppbv increase in O 3 from 1850 to 2000, and from 1980 to 2000 a change of less than 2 ppbv on an annual mean of 29 ppbv (Young et al., 2018). That change, of less than 3% per decade, was close to the change observed at Alert from 1990 to 2014 of 2% per decade.
GEM at Alert from January to March decreased by 16% from 1995 to 2014. During winter at Alert, there was an absence of local photochemical conversion of GEM to shorter-lived reactive mercury compounds, and thus, regional photochemistry could not account for this decline. This trend was most likely due to changes in emissions from sources in the Northern Hemisphere . Mercury global emissions peaked in the 1970s (Pacyna et al., 2016). While North American and European emissions have decreased, Asian emissions have continued to increase (Pacyna et al., 2010). The decrease in European emissions and to a lesser degree North American emissions combined with the increase in Asian emissions might explain the net change in GEM at Alert from 1995 to 2014.

Trends in Alert Anthropogenic Aerosols in Relation to
Emissions of SO x , BC, NH 4 , and NO x Aerosol SO 4 2− and EBC concentrations peaked in the winter and spring months (Figure 1), and they declined significantly during the mid-1990s ( Figure 2). Trends of surface-based SO 4 2− and EBC at Alert, Barrow, and Ny-Ålesund were mainly due to changes in source emissions from the FSU and northern EU region (Quinn et al., 2007;Sharma et al., 2004;Xie et al., 1999). Normalized concentrations of SO 4 2− and EBC, averaged over January to March, declined by 50% from 1990 values ( Figure 3a). The changes were associated with declines in SO x and BC emissions from Eurasia (EMEP for SO x emissions for 1990-2015; BC emissions for 1990-2005 from Sharma et al., 2006) largely due to emission reduction in the FSU region between 1990 and 1995 ( Figure 3b) rather than changed transport .
Since 1990, NO x emissions from Eurasia declined by about 40% (Figure 3c), compared with decreases in emissions of SO x and BC of roughly 80%, most of which occurred before 2000. In contrast, NO 3 − concentrations at Alert increased since 1990 by about 20% (Figure 3a) and EMEP observations for Europe and Russia (Fagerli & Aas, 2008) showed little change between 1980 and 2003 in total atmospheric nitrate (gas-and particle-phase nitrate or NO 3 − ). Using modeling and observations, these authors demonstrated that the lack of trends for NO 3 − in air can be attributed at least partly to a shift in the equilibrium between nitric acid and ammonium, which controls the level of ammonium nitrate. Considering that Eurasian emissions showed a relative increase in NH 3 to SO x (Figure 3c), the associated reduction in aerosol acidity at Alert (Figure 2d) likely shifted the partitioning of atmospheric nitrate from gaseous nitric acid to aerosol nitrate (e.g., Guo et al., 2016). Similar trends in nitrate were also found in Greenland ice cores (Iizuka et al., 2018).  Figure 3c). In other words, the Alert aerosol showed an increase in neutralization associated with a shift from sulfuric acid to ammonium sulfate particles, due to a decline in SO 2 emissions relative to NH 3 emissions. One potential consequence of increased neutralization in the Arctic would be an increase in ice nuclei formation, which would ameliorate the infrared cooling effect of winter ice clouds associated with acidic arctic haze sulfates acting through the dehydration-greenhouse feedback effect (  ozone (ppb), and GEM for the dark mid-winter months (November to February) and for the polar sunrise light months (March to May). In the upper right-hand side, the percentage change between the first 5 years and the last 5 years means is shown.

Arctic Dark Versus Sunlit Period
Trends in photochemically produced constituents SO 4 −2 , O 3 , Br, and GEM in the dark period November to February and in the sunlit polar sunrise period March to May are compared in Figure 4. For all variables, but particularly ozone and GEM, there was more scatter in the data during the sunlit period than in the dark period.
From 1980 to 2013, there was a net decrease of aerosol sulfate by 47% and 34% for dark and light periods, respectively, consistent with 20-45% less oxidation of SO x associated with the dark (Barrie & Hoff, 1984;Barrie, Li, et al., 1994a). Differences in transport and removal may also occur between dark and light but would generally work to cause decreasing aerosol pollution going from dark to light. As observed by Barrie et al. (1994a), aerosol fine particle V was well correlated with SO x throughout January to April of 1992. The seasonal variation of nonsoil V ( Figure 1p) showed V levels generally declining from dark to light period suggesting weaker transport and/or stronger removal from dark to light. However, the levels of sulfate peaked in April ( Figure 1a). The only difference between V and SO 4 −2 is that SO 4 −2 had a gaseous precursor and V did not. Thus, it can be concluded that the spring peak in SO 4 −2 was driven mainly by oxidation of SO 2 in the Arctic.
On average, aerosol Br concentrations were much higher in the light period than in the dark period (58%). Greater photochemical release of surface sea salt Br expected during the sunlit period into the atmosphere as Br gases and aerosol Br (Pratt et al., 2013). Also, after polar sunrise, aerosols became highly enriched in bromine because of the transfer of bromide from snowpack sea salt into the air as gaseous bromine compounds that then deposited on existing aerosol particles. In the sunlit period (Figure 4d), there was a marked decrease in Br by 52%, most occurring after 2002, suggested that there was a decrease in the photochemically induced release of bromine from snow (snowpack or blowing snow) into the air before it arrived at Alert. This warrants further detailed investigation with an atmospheric chemical transport model that includes realistic air-snow chemical/physical interactions and their dependence on meteorological factors such as air temperature and vertical mixing of ozone from the free troposphere into the surface-based inversion layer over the Arctic Ocean where the bromine explosion chemistry takes place.
The average O 3 levels (Figures 4e and 4f) were lower during the sunlight conditions (averaging 18 to 25 ppb) than in the dark period (averaging 35 to 40 ppb). Variations in ozone during the dark were small compared with the light period. In the dark, ozone increased slightly by 3%, while in the light, ozone changed little except for a marked decrease in 2012 and 2013. The variance of GEM concentrations ( Figure 4h) increased from dark to light reflecting enhanced GEM depletion by halogens after polar sunrise. In the dark when GEM was likely influenced mostly by distant sources, concentrations decreased by 16% with most occurring from 2008 to 2013. In the light, average GEM was constant until the last 2 years 2012 and 2013 when it decreased by 11%. Since O 3 and GEM were very well correlated at Alert after polar sunrise due to bromine-related chemistry, the minima in 2012 and 2013 in these two constituents indicated a common photochemical cause. This further needs investigation.

Trends in Natural Aerosol Constituents With Two Seasonal Peaks
Three natural aerosol constituents (MSA, I, and soil) have marked double peaks in their annual cycles, not associated with the Arctic haze components and the trends in their peak concentrations shown in Figures 5-7.
MSA, an aerosol product of oxidation of dimethyl sulfide gas of marine biogenic origin, could be considered as a biogenic indicator aerosol. At Alert, it had a primary peak in late spring (April to May) and a secondary peak in summer (July to August). The spring peak was connected with distant transport from open water areas of the North Atlantic Ocean, while the summer peak was associated with open water areas in the Arctic Ocean exposed by retreating ice (Sharma et al., 2012;Li & Barrie, 1993;Li et al., 1993). MSA concentrations showed more variance in the spring than in summer (

Journal of Geophysical Research: Atmospheres
July and August, which paralleled an increase at Barrow of 12% per year (Quinn et al., 2009). This increase was reported as a possible consequence of sea ice reduction (Quinn et al., 2009;Sharma et al., 2012). However, sea ice reduction did not explain the higher MSA prior to 1990.
Between 1980 and 2005, aerosol iodine ( Figure 6) had a primary peak in spring from March to May and secondary peak from August to September. Previous studies suggested that the spring peak was a mixture of marine biogenic iodine and iodine involved in photochemical reactions at polar sunrise, while the late summer peak was a product of marine biogenic production and transport (Allan et al., 2015;Raso et al., 2017;. More discussion on aerosol iodine could be found in section 4.2.9 since it appeared as a distinct factor in the PMF analysis. The variance in the spring peak was greater than in the summer peak. The spring iodine levels reached a maximum in 1997, while changes in the late summer I concentrations were much less across 1980 to 2005. Contributions from soil to the aerosol at Alert were represented by Al and Ca (see also section 4.2.7). Figure 7 showed the highest concentrations of Al found in late summer/early autumn (15 August to end of October) and a smaller but still significant concentration peak occurred in spring from March to May. Sirois and Barrie (1999) deduced that the spring peak was related to long-range transport of windblown dust from Asian deserts, while the other peak had more of a local Arctic wind-driven soil factor. This was supported by a recent Arctic dust modeling study by Groot Zwaaftink et al. (2016) who concluded that (i) Arctic dust deposition from local sources peaked in autumn, while dust deposition from remote sources occurred mainly in spring and (ii) contributions of local dust sources were more important in the Arctic than previously thought, particularly with respect to surface concentrations and dust deposition. The variance of the late summer/early autumn peak of Al was greater than that of the spring peak. While the late summer/early autumn levels decreased by 27%, the spring levels increased slightly by 5%. Changes in aerosol Ca +2 were similar. Investigation using a comprehensive wind-blown aeolian dust transport model that took

Positive Matrix Factorization and Factor Analysis Methodology
Positive matrix factorization (PMF), a multivariate technique was used, in general, to lower the dimensionality of a data set and, specifically in atmospheric science, to explore the source factors of trace constituents in the atmosphere. One of the key features in atmospheric aerosol and gas observational data sets is the lack of negative variables. This natural physical constraint was the motivation for the development of PMF. We used the program PMF2 from Paatero (1997). It was developed, described, and applied initially by Paatero and Tapper (1994) (Antilla et al., 1995;Juntto & Paatero, 1994;Paatero, 1997;Xie et al., 1999). We adhered to the methodology described in Sirois and Barrie (1999) in analyzing the first 15 years of aerosol observations data at Alert. A detailed description of the methodology can be found online in a published thesis (Magnusson, 2016). There are several ways of normalizing the output matrices G and F from PMF2; the normalization was made to properly compare the solutions. Seven different options (Paatero, 2004) can be chosen such as normalizing by the maximum or mean value of each column of G or F. In this analysis, the option used was normalizing so that the mean value of each column in G equaled 1 as to compare with results from Sirois and Barrie (1999), where the same normalization method was used. The choice of number of factors was based on a combination of statistical and physical criteria. The tests used included the minimization of unexplained normalized variance factor Q in PMF. When this offered multiple solutions, physical criteria were used similar to those described in Sirois and Barrie (1999). For the dark period, the lowest Q was between 10 and 13 with a minimum at 12 PMF and for the light period between 13 and 14. So based on this and the results looking physically realistic, the choice was made to use 12 PMF factors for the dark period and 13 PMF factors for the light period.
The results of Sirois and Barrie (1999) were reproduced first for aerosol observations from 1980 to 1995 for the cold polluted period from November to May, which combined both dark and light periods. Then, we proceeded to analyze the data set of aerosols and gases listed in Table 1 for the period 1995 to 2013 because EBC and GEM were only available from 1990 and 1995 onward. Two PMF analyses were conducted, one for the dark period (November to February) and one for the light period (March to May) in order to examine the effects of changing meteorological and light conditions.

Results of PMF Analysis
PMF analysis for 22 variables during the dark with 12 PMF factors is shown in Table 2 and during the light with 13 PMF factors are shown in Table 3. With only one exception, the first 12 PMF factors in each period were the same. The exception was that a NITRATE factor in the light period replaced ANTHRO in the dark. The extra thirteenth factor in the light PHOTO O 3 -GEM appeared due to increased variability during the polar sunrise period of March to May (see Figures 1u, 1v, and 4).
For dark and light periods, respectively, Tables 2 and 3 list (a) the apportionment of average variable mass to each factor as well as, in brackets, the percentage of total variance that was explained by the PMF and (b) the ion balance of each factor (the ionic balance of each factor is represented by the sum of cation and anion equivalents, and their ratio is given in the last row). Each entry has a standard deviation. Bold highlighting is used to flag table entries that were significant at two standard deviations. Unexplained variance was always less than 19% and often less than 10%.
From Tables 2a and 3a, one can easily extract information on the quantitative questions, "what PMF factors are significant in explaining the total average mass of a particular constituent (e.g., SO 4 −2 )?" and "what fraction (i.e., percentage) of the total variance in a variable such as SO 4 −2 is explained by each PMF factor?" For example, from the first row of Table 3a, total SO 4 −2 mass in the LIGHT period was significantly loaded (bold values in the table that are greater than at least 2 times the standard deviation) on 9 of 13 PMF factors highlighted in bold. The percentage of total variance explained by each factor was given in brackets in the second line of the row for SO 4 −2 .
From Tables 2b and 3b one could easily extract information on the quantitative questions "what fraction of the total ion equivalents for a PMF factor is contributed by a particular aerosol constituent?" and "do the 10.1029/2019JD030844 ). The estimated standard deviation is given for each loading. Loadings for which the interval value plus or minus two times the standard deviations does not include zero 0 as indicated in bold. Percent (%) in brackets are is variance explained by the factor of that variable. Note.

Journal of Geophysical Research: Atmospheres
The bold and highlighted numbers are values that are significant at two standard deviations. The ratio of total cation to anion equivalents on the bottom row indicate the ion balance in the analyzed data. ). The estimated standard deviation is given for each loading. Loadings for which the interval value plus or minus two times the standard deviations does not include zero 0 as indicated in bold. Percent (%) in brackets are is variance explained by the factor of that variable.

Note.
The bold and highlighted numbers are values that are significant at two standard deviations. The ratio of total cation to anion equivalents on the bottom row indicate the ion balance in the analyzed data.
In mass unit ng m -3 . The estimated standard deviation is given for each loading. Loadings for which the interval value plus or minus two times the standard deviations does not include zero as indicated in bold.
% in brackets are variance explained by the factor of that variable.

10.1029/2019JD030844
Journal of Geophysical Research: Atmospheres negative cations balance the positive anions?" Equivalent mass is defined in any standard physical chemistry text. If charge balance is achieved, then total cation equivalent mass equals total anion equivalent mass in the aerosol sample and the ratio CATION/ANION was not significantly different than 1. If it is not balanced, important anions such as carbonate ion or organic acid ions have been missed in the analysis.

Factor 1 ACID PHOTO-S
This factor was dominated by H + and SO 4 −2 with an equivalence ratio of H + /SO 4 −2 of 1.67 and 1.21 in dark and light periods, respectively (Tables 2b and 3b), and only small amounts of NH 4 + . The equivalence ratio increased from 0.92 during 1980 to 1.67 for the 1995-2013 period. There is long-standing evidence that this factor represents an anthropogenic aerosol that was formed by in situ gas-to-particle conversion of anthropogenic SO 2 at polar sunrise in the Arctic (Barrie & Hoff, 1984;Barrie, Li, et al., 1994a, 1994c. The average concentration of SO 4 −2 was greater in the light period (179 ng·m −3 ) than in the dark period (118 ng·m −3 ). This factor had an excess of cations to anions (Tables 2b and 3b) for both dark and light periods (1.4 and 1.5, respectively, increased from a value of 1.1 in . This is consistent with the absence of organic anions in the measurements. Carbonate ion, which was not measured, would not play a role in such an acidic environment.

Factor 2 BROMIDE
This factor hosted most of the aerosol Br in both dark and light periods (71% and 67% respectively, as compared to 92% in . Additionally, it was significantly loaded by SO 4 −2 , Na + , K + , and Pb in the dark and by only SO 4 −2 and Na + in the light. The enrichment factor with respect to bulk seawater of Br − to Na + was [Br − /Na + ] aerosol /[Br − /Na + ] seawater . The EF ss Br − was 37 and 173, respectively, indicating that Br − was not from primary sea salt but rather some enrichment process that also involved gas-toparticle conversion in snow containing marine sodium and non-sea salt sulfate. We believed that processes in snow in the atmosphere or in the snowpack on the ground were involved. Indeed, Toom-Sauntry and Barrie (2002) observed that fresh snowfall at Alert had a sea salt Br − enrichment factor that was twice the enrichment found in atmospheric aerosol. The BROMIDE factor in the dark is associated with the extra three elements (i.e., SO 4 −2 , K + , and anthropogenic Pb) but in the light, only with SO 4 −2 .
This suggests that in the dark, gas-to-particle conversion of Br compounds and SO 2 was contributing extra Br − and SO 4 −2 to snow containing anthropogenic K + and Pb, while in the light, the conversion contributed extra Br − and SO 4 −2 to snow containing elevated levels of sulfates but less K + and Pb.
This explanation is consistent with anthropogenic aerosol K + and Pb primary concentrations higher in the dark than in the light period (Figures 2h and 2m).
After polar sunrise, Br − can be produced incrementally by each ozone depletion event that occurred as the air pathway encountered the atmospheric boundary layer and surface wind blown snow on its way to Alert. As an aerosol, its lifetime was much longer (Barrie, 1986) than one ozone/GEM depletion event (Gong et al., 1997). Thus, aerosol Br − was the outcome of many ozone depletion events and therefore not strongly correlated with O 3 and GEM (see section 4.2.12 for more explanation).

Factors 3 and 4 MODIFIED SEA SALT and SEA SALT
These factors clearly emerged as having ratios of Cl − , Ca +2 , and K + close to that of bulk seawater (supporting information Table S1, where the sea salt enrichment factors [EF ss ] are shown). These two factors emerged in the PMF analysis of Sirois and Barrie (1999) for the 1980 to 1995 period for November to May. Mg was also available and in the same ratio to Na + as in seawater.
There were two notable differences between the MODIFIED SEA-SALT and SEA-SALT factors. First, Cl − was depleted and SO 4 −2 enriched in the modified factor. It was well documented that sulfuric acid produced in the Arctic atmosphere from SO 2 oxidation, especially as sunlight increases, could titrate Cl − from sea salt aerosol. This produced volatile HCl gas that leaves the aerosol and could be scavenged by other more alkaline aerosols, at the surface or by precipitation. Indeed, Toom-Sauntry and Barrie (2002) showed that snowfall in the winter at Alert is slightly enriched in Cl − relative to Na + in bulk seawater. One source of this enrichment could be scavenging by snow of HCl gas from sea salt aerosol modification.
Second, while nitrate significantly loaded on both sea salt factors, NH 4 + only loaded on the modified sea salt factor. This likely reflected the close association of NH 4 + to sulfuric acid, which was largely responsible for modifying the sea salt (see section 3.2.4).
A notable difference between dark and light periods was that SO 4 −2 enrichment in the MODIFIED SEA-SALT factor increased from an EF ss of 13.7 to 22 (supporting information Table S1) likely because of increased photo production and availability of sulfuric acid in the sunnier Arctic atmosphere.
The loading of O 3 and GEM on this factor is interesting. First, note that in Figures 1u and 1v the variance of ozone and GEM in the dark was much less than in the light. In the dark Arctic troposphere, the lifetime of ozone in the atmosphere is much greater than the average transport time of air to Alert from t midlatitudes (i.e. months versus weeks). Similarly, GEM is long-lived with an average tropospheric lifetime of 0.5 to 1 year unless it encounters halogens from the snowpack after polar sunrise. The vertical profiles of ozone and GEM in the troposphere tend to increase with altitude. Any variance in these variables in the dark is likely due to sources of air at Alert arriving from different altitudes (e.g., free tropospheric versus boundary layer). Thus, we speculate that the modified sea salt factor is regulated by altitudinal dependent processes.

Factors 5 ANTHRO (Dark) and NITRATE (Light)
The ANTHRO factor appearing in the dark period had the largest number of variables loading on it (13 out of 22). The major anion (   with additional contributions by Na + and K + . It is not acidic suggesting an aged and anthropogenic mixed aerosol. This is consistent with significant loading by many heavy metals (Pb, V, Zn, Cu, Mn, and Ni) explaining 60% in the variance of Pb and between 14% and 22% of variance of the other heavy metals as well as I (11%) and Br (8%). ANTHRO also appeared in the analysis of Sirois and Barrie (1999) where its factor score was a maximum mostly in the dark from December to March (Figure 7 of . The ratio of Ni/Pb in ANTHRO was 0.046 ± 0.15, within a factor of 2 of iron and steel production emissions to the Arctic from Northern Europe and Russia in the year 2000 (Pacyna et al., 2006) having a ratio of 0.075. It was much lower than the ratio from nonferrous metal production (0.33) and from "other sources and cement production" (0.862). It was higher than in gasoline emissions which had no nickel relative to Pb. Thus, iron and steel production likely had a major influence in this factor.
In the light period (March to May), NITRATE replaced ANTHRO. It was significantly loaded by NO 3 − and SO 4 −2 , which were reasonably well balanced by NH 4 + and Ca +2 (Table 3b). It had relatively less acidity. Bottenheim et al. (1993) and Muthuramu et al. (1994) found that a considerable fraction of total atmospheric nitrogen oxide mass was in the gas phase in the form of organic nitrate gases during April 1992. Organic nitrate gas was, possibly, converted to inorganic particle nitrate to yield a particulate NITRATE factor through photochemical reactions after polar sunrise (Bottenheim et al., 1993;Jaeschke et al., 1999).
The distribution of NO 3 − among PMF factors is shown in Figure 8a. The average total NO 3 − concentration in the dark and light were similar at 81 and 103 ng·m −3 , respectively (Tables 2a and 3a). However, the distribution among aerosol factors was very different. In light, 50% of the total NO 3 − was on the NITRATE factor, while in dark, most was associated with MODIFIED SEA-SALT and BC.

Factor 6 ZINC
In both dark and light periods, the Zn loadings were mostly split between ZINC and SMELTER factors. Additionally, there was a smaller contribution associated with the dark-period ANTHRO factor (Figure 8b). Other variables that showed significant loadings associated with dark-period ZINC factor (Table 2a)  , K + , Pb, and Mn were found in the ZINC factor. However, the fractional masses of these other variables explained by ZINC was less than 10% (second line of each row in the table) in all cases except for Pb and Mn in the light period (34% and 15%, respectively). Following the global emissions estimates of Pacyna and Pacyna (2001), it was likely that most ZINC factor was related to stationary fossil fuel combustion sources and ferrous smelting.

Factor 7 OIL COMBUSTION
This factor was distinguished in both dark and light by atmospheric aerosol V (dark 67%, light 56%) and Ni (dark 19%, light 40%). Oil combustion mainly from stationary sources and shipping was the major source of atmospheric emissions of Ni and V globally Viana et al., 2014). The emissions ratio of V/Ni in 1995 was 2.52 globally and 2.80 for Europe including Russia . For our OIL COMBUSTION factor, the average ratio of V to Ni in the dark and light period factors were 5.3 ± 2.0 and 1.5 ± 0.3. Viana et al. (2014) reported that in Spain, PM 10 and PM 2.5 aerosols have ratios of V/Ni = 4 ± 1 for oil burning shipping. They also noted that oil combustion from ship sources had V/EBC <2. In contrast, characteristic ratios obtained from land sources for their study area were V/Ni = 12 and V/EBC > 8. In our dark period OIL COMBUSTION SOURCE factor, the ratio of V/EBC was 0.065 (i.e., <<2).
Considerable V, Ni, and EBC were loaded on the dark period ANTHRO factor with the dark period ANTHRO factor. Values of V/Ni equal to 0.98 ± 0.63 and V/EBC equal to at 0.07 ± 0.30 are consistent with emissions from iron and steel production sources (i.e., V/Ni = 1) from northern industrial complexes such as Norilsk.

Factor 8 SMELTER
This factor appeared in both dark and light periods. It was marked by a loading of 80% to 90% by Cu in both dark and light periods but only 19% and 47% Zn in the dark and light periods (respectively Figure 8b and 8c). Sirois and Barrie (1999) found that the mass ratio of Zn/Cu in SMELTER was 1.17 ± 0.34. The ratio in the present study was 0.77 ± 0.20 and 0.77 ± 0.16 for dark and light periods, respectively. Within the uncertainty, our results did not differ statistically from Sirois and Barrie (1999) and the averages were closer to the ratio in atmospheric emissions of 0.77 from the NILU database reported by Pacyna (1995). Most of the smelting activity in the European and Russian north was reported to be in the Kola Peninsula and around Norilsk. A snowpack deposition study around Monchegorsk smelter (Jaffe et al., 1994) found that only 25% of Cu  Tables 2a and 3a. emissions were deposited locally leaving the remaining fine particles exported from the region. It was this finer particulate Cu observed in the high Arctic.
Note that there was an insignificant SO 4 −2 loading on the smelter factor in dark or light period (Tables 2a   and 3a). As argued by Sirois and Barrie (1999), this was consistent with northern smelter emissions having very low primary sulfate. Most of the contribution of smelters to Alert sulfate aerosol comes from SO 2 emissions that were converted during transport to sulfate and likely appeared in the ACID PHOTO-S, MODIFIED SEA-SALT, and ANTHRO factors. 4.2.8. Factor 8 SOIL SOIL explained considerable variance in Al (dark 76%, light 62%), Ca +2 (dark 58%, light 60%), and Mn (dark 43%, light 34%). Also, SOIL was significantly loaded by NO 3 − but explained only 4% and 7% of the variance.
Sirois and  concluded that the ratios of Ca +2 , Mn, and V to Al in aerosols at Alert were closer to that of local soil than the average composition of crustal rock (Mason, 1966). This study suggested that relative to Mason's average crustal rock, Alert atmospheric SOIL aerosols were 42% lower in Mn relative to Al and a factor of 2 higher in Ca relative to Al (supporting information Figure S2). For another discussion of SOIL see section 3.2.6 and Figure 7.

Factor 10 IODINE
The distribution of total aerosol I among factors was very different between the dark and the light periods ( Figure 8d). In the dark, 90% of iodine was found on the factor IODINE and the rest on ANTHRO. IODINE in this period was significantly loaded by small amounts of Br and GEM. However, in the light period only 28% of the total I is on the factor IODINE and 58% is associated with SEA-SALT and MODIFIED SEA-SALT. There is also small but significant loading on ACID PHOTO-S and NITRATE.
The IODINE factor in the light period is significantly loaded by soil-associated elements Al (27% variance explained), Mn (18%), and V (13%). However, V and Mn had soil enrichment factors EF soil of 9 and 17, respectively (Ratio of EF soil to Mason Average, Figure S2). Thus, soil dust had been modified by anthropogenic V and Mn. This makes some sense as long-range transported windblown dust peaks in the Arctic at this time was likely coming from the Asian deserts via polluted northern China. See also the discussion of I and Al in section 3.2.6.
The photochemistry associated with polar sunrise production of aerosol iodine likely involved gaseous production of involatile iodide compounds from chemistry in the atmosphere or surface snowpack with subsequent deposition on available and accommodating aerosol surface areas (Raso et al., 2017). For instance, involatile iodide compounds were likely to be accommodated on aerosols with low acidity. The two factors SEA SALT and IODINE with soil were not acidic.

Factor 11 BLACK CARBON
BLACK CARBON explained 60% and 59% of the variance in EBC in dark and light periods, respectively. The fraction of total EBC contributing to this factor was the same for both dark and light periods at 76-77% (Figure 8e). , NO 3 − , Na + , K + , Pb, and O 3 were also contributed to the BLACK CARBON factor. Of total nitrate, 23% and 11% were loaded on this factor in the dark and light periods, respectively (Figure 8a). The loading of SO 4 −2 on BLACK CARBON in the light but not the dark is consistent with black carbon hosting major condensation of secondary sulfate from oxidation of SO 2 in the Arctic as the light increases in spring.
K + apportioned significantly to BLACK CARBON in the light period but not the dark period (Figure 8f). Leaitch et al. (2018) found nss-K + correlated with EC in both the light and dark periods (r 2 = 0.62, 0.63). K + of nonsoil, non-sea salt origin existed in submicrometer aerosols associated with combustion of biomass rather than of fossil fuels (Andreae & Raemdonck, 1983). Warneke et al. (2010) reported that biomass burning aerosols were occasionally observed throughout the Arctic and that due to a lack of long-term data and strong interannual variability, it was not clear if a trend in fire emissions in Russia existed and it was unclear how much of Arctic warming could be attributed to biomass burning emissions. Long-term trends in nss-K + in the dark ( Figure 2e) showed a steady decline totaling 44%, which was comparable to a decline in EBC of 52%. Between the beginning and the end of the observation period, the average fraction of total K + that was nss-K + declined from 65% to 43%. There was 19% more aerosol Na + at the end than in the beginning of the record (Figure 2g).
Thus, in the winter polluted high Arctic lower troposphere, EBC was largely independent of other variables and loaded on a single factor BLACK CARBON that was connected with secondary SO 4 −2 and NO 3 − aerosols and, in the spring, with K + from biomass burning. Recent isotopic analysis of Alert black carbon showed that in the period 2011 to 2015 BC was 60% fossil fuel and 40% biomass burning throughout the year (Winiger et al., 2019). EBC decreased by 52%, mostly during the early 1990s (Figure 2r). This decrease was likely due to a decline in emissions from anthropogenic fossil fuel and to a lesser extent biomass burning sources. 4.2.11. Factor 12 MSA Over 90% of total MSA loaded on this factor for both the dark and light periods (Tables 2a and 3a). The average concentration was much higher in the light (8.12 ng·m −3 ) than in the dark (1.43 ng·m −3 ). This was due to the seasonal variation (Figures 1c and 5), which had a peak in April/May and a secondary peak in July/August. In the dark, the MSA factor was significantly loaded with Al, O 3 , and Hg and in the light only SO 4 −2 . See the discussion in section 3.2.6.
Our analysis yielded a mean molar ratio of MSA/biogenic SO 4 −2 of 0.045 ± 0.027 and 0.135 ± 0.060 in the dark and light, respectively. These ratios were generated by using the MSA and sulfate loadings on the MSA factor for the dark and light periods (Tables 2a and 3a). In comparison to our results, Sirois and Barrie (1999) found an MSA factor for Alert data from 1980 to 1995 for the winter months (November-May) that had a mean molar ratio of MSA/biogenic SO 4 −2 of 0.08 ± 0.05. These results are very consistent with independent determinations of this ratio by using isotopes of sulfur in sulfate by Li and Barrie (1993) where the molar ratio of MSA/biogenic SO 4 −2 was <0.01 in mid-winter (our dark period) and increased through May to a mid-summer peak of 0.2 to 0.6.

Factor 13 PHOTO O 3 -GEM
This factor only appeared in the light period and likely reflected the added variance in the O 3 and Hg observations caused by lower tropospheric ozone and GEM destruction occurring after polar sunrise on 5 March at Alert and before springtime melt in late June in the Arctic (Figures 1u and 1v). At Alert, there was a strong correlation of O 3 and GEM depletion (Schroeder et al., 1998b(Schroeder et al., , 1998c and indeed throughout the Arctic (Lu et al., 2001). This correlation was driven by meteorological modulation of the air transport to Alert. The PHOTO O 3 -GEM explains 59% and 55% of the variance in the O 3 and GEM, respectively.
The fact that bromide in aerosol in the light period loaded on the BROMIDE factor and not the PHOTO O 3 -GEM is not inconsistent with the well-documented peaks in Br − during polar sunrise ozone and GEM depletion events at Alert. It rather suggested that long-lived Br − aerosols produced like this throughout the Arctic atmosphere could arrive at Alert in situations where ozone and mercury were not depleted locally. In other words, Br − production could be the net result of many bromine explosions as air encounters the arctic boundary layer on its long trajectory to Alert. Barrie et al. (1994c) observed using daily observations (rather than weekly mean) from January to April that Br − was anticorrelated with ozone but the relationship was very nonlinear. The highest Br − concentrations were only observed when ozone was completely depleted. Above ozone levels of 5 ppb, there was much less correlation with O 3 . Thus, given that we dealt with weekly samples, it was not surprising that Br − loads mostly on the BROMIDE factor and only a little on PHOTO Ozone-GEM (see also section 4.2.2).

Conclusion
For the peak period of winter (January-March), the concentrations of many anthropogenic constituents of Arctic aerosols (SO 4 −2 , H + , NH 4 + , K + , Cu, Ni, Pb, Zn, nonsoil V, nonsoil Mn, and EBC) decreased over the observational period (1980-2013 except for metals that were analyzed only until 2006) by amounts between 23% and 80%. The decline is best explained by changes in emissions rather than in transport/deposition processes . Much of the decrease took place in the early 1990s as European and Northern Asian emissions decreased rapidly. An exception was anthropogenic aerosol nitrate, which increased by 20% at Alert despite a marked decrease of approximately 40% in emissions of nitrogen oxides from Eurasia. This is consistent with the observed decline of aerosol acidity from an H + /(SO 4 −2 + NO 3 − ) ratio of 0.40 to 0.15 as total aerosol SO 4 −2 dropped by 52%. This acidity change would have shifted the partitioning of NO 3 − from the gas phase as HNO 3 to aerosols, thus offsetting the effect of declining NO x emissions.

Journal of Geophysical Research: Atmospheres
Sulfate, a dominant constituent of the Arctic aerosol, plays an important role in indirect climate forcing in the high Arctic winter atmosphere. Over the 34-year observation period, the aerosol neutralization at Alert increased due to a shift from formation of sulfuric acid to ammonium sulfate aerosol resulting from declined SO 2 emissions relative to NH 3 emissions in the source regions. Aerosol acidity concentrations decreased by 76%, while sulfate decreased less by 52%. We suggest that, through the dehydration-greenhouse feedback effect in Arctic winter ice clouds involving ice nucleation Blanchet & Girard, 1994;Fisher et al., 2011;Girard et al., 2005), the observed increase in aerosol neutralization enhances ice formation (Abbatt et al., 2006) and ameliorates the infrared cooling caused by acidic sulfates at their peak in the 1980s.
The large drop in the metals Cu, Ni, Pb, non-soil V, non-soil Mn, gas-phase elemental Hg, and Zn is clearly linked to changes in emissions from smelting and fossil fuel combustion activities.
Sea salt derived Na + (+19%) and Cl − (+43%) increased at Alert by 19% and 43% between 1980 and 2013. These peaked in abundance from November to March at a time when wind-driven sea spray production in open northern oceans as well as resuspension of sea salt from the Arctic Ocean surface is at a maximum.
We employed PMF analysis following the same methodology as Sirois and Barrie (1999) who studied winter/spring observations from 1980 to 1995. Our study dealt with subsequent observations from 1995-2006 with four added variables (Ni, EBC, O 3 , and GEM) and with two contrasting data sets for dark winter (November-February) and for light spring (March-May). Alert aerosols were divisible into 12 major factors (in contrast to 10 for Sirois and Barrie) that could be clearly identified with sources or atmospheric chemical processes in both dark (November to February) and light periods (March to May) for 1995-2006. In addition, in the light period, a thirteenth factor emerged related to atmospheric ozone and gas-phase elemental mercury depletion. The 12 PMF factors consisted of one photochemical sulfate aerosols; two sea salt types (one for standard sea salt and one for chemically modified); two halogen types (Br and I); five anthropogenic aerosol types (black carbon, zinc, oil combustion, smelter, and a general anthropogenic type related to primary emissions); one soil type; and one biogenic marine sulfur type. The 22 observed individual aerosol and gas constituents were quantitatively apportioned to these factors for dark and light periods.
For some compounds (e.g., iodine and nitrate), there were marked changes associated with the appearance of sunlight. In the light, 50% of the total NO 3 − was on the NITRATE factor, while in the dark, most was associated with MODIFIED SEA-SALT and BC. In the dark, 90% of iodine was found on the factor IODINE, while in the light only 28% was on the factor IODINE and 58% associated with SEA-SALT and MODIFIED SEA-SALT.
The SMELTER factor was marked by Zn and Cu in a ratio consistent with estimated emissions of these metals from northern smelter complexes in Eurasia. Ni and V marked a distinct oil combustion factor that was low in black carbon. A BROMIDE factor was separate from an O 3 -Hg depletion factor. This is consistent with the growing consensus that aerosol bromide production can be the net result of many "bromine explosions" as air encounters the arctic boundary layer, sunlight, and sea salt halogens on its long pathway to Alert.
These cost-effective unique measurements at Alert, using a simple robust technique of collecting large weekly samples of aerosol archived and later analyzed by multiple techniques, were extremely valuable. Over more than 30 years, the observations have provided a continuous thread of information connecting many short-term campaign research studies as well as unique information on seasonal and long-term trends of natural and anthropogenic aerosol factors and their sources. They formed the basis for understanding the simultaneous long-term climate observations of aerosol physical and radiative properties and of greenhouse gases made at Alert since the late 1980s. The multidecadal trends and quantitative apportionment to source related factors of 22 aerosol and gas-phase atmospheric constituents could be used for many valuable purposes, for instance, validating climate and chemical transport models for accurate emissions, transport, transformation, and removal of atmospheric aerosol constituents over the weeks they were transient from midlatitudes to the high Arctic. A recent test of chemical transport models using Arctic data shows that only a few of many models captured accurately the transport and seasonality sulphate and black carbon (Eckhardt et al., 2015).