Eight clusters were found to describe the full range of aerosol types observed at Villum: five that peak in the ultrafine (UF) mode range, and three that peak in the accumulation mode range. Their typical PNSDs and monthly occurrences are presented in Fig. 4. Their names are assigned, in order of increasing peak diameter, as Nucleation, Bursting, Nascent, Pristine, Ultrafine (UF) Bimodal, Accumulation (Acc.) Bimodal, Haze, and Aged. These are the same clusters found previously in the aforementioned studies by Dall´Osto et al. (2017; 2018; 2019) and Lange et al. (2018; 2019), so the naming scheme has been adopted. The Pristine cluster represents clean Arctic air, while Nucleation, Bursting and Nascent represent different stages of new particle formation (NPF) and growth. The UF Bimodal cluster peaks in both the ultrafine and accumulation mode ranges, owing to some influence from both processed Arctic marine aerosols and aerosols of anthropogenic origin (Dall'Osto et al., 2018). The Acc. Bimodal cluster also has both biogenic and anthropogenic sources with a proposed contribution from Arctic marine aerosols. Haze and Aged are known to be heavily anthropogenic, with Haze representing Arctic haze aerosols, with high concentrations of sulphates and BC, and Aged representing aged anthropogenic aerosols with a noticeable lack of smaller particles (Lange et al., 2018). Their properties are further described in these previous studies by Dall’Osto et al. (2018) and Lange et al. (2018). The eight cluster can also be grouped into those that are primarily Anthropogenic (Haze and Aged), those that are primarily Biogenic (Nucleation, Bursting and Nascent), and those that are Mixed (UF Bimodal and Acc. Bimodal). The Pristine cluster alone represents the Background aerosol.
The yearly relative occurrences of the eight clusters over the entire time series are shown in Fig. 5. Due to the extreme seasonal variation in cluster occurrence shown in Fig. 4, any data gaps may significantly affect the yearly occurrence of certain clusters. For example, data collection began in the summer of 2010 and SMPS data are lacking during a large part of winter 2010 and 2011. It is therefore unsurprising that the Aged cluster, primarily observed in midwinter, is almost absent from these years. The same applies to 2016 and 2017; SMPS data are sparse in December 2016 and from October to December 2017. To reduce this bias, Fig. 5 has been separated into two, motivated by the seasonality observed in Fig. 4: “summer” months (from May to October) and “winter” months (from November to April). Variability is still seen of course, and it should be noted that this will likely be a combination of the aforementioned seasonality in data availability, the true variability in the aerosol observed at Villum, and potentially long-term trends in source emissions and aerosol transport pathways to Villum. In a previous study, the long-term trends of physical aerosol properties were investigated, using the same k-means clustering method (Pernov et al., 2022). It was found in that study that the occurrence of the Nucleation cluster during summer is increasing, attributed to changing transport patterns into the Arctic. It was also found that the occurrence of the Pristine cluster during autumn is increasing, likely due to a combination of increased accumulated precipitation along the transport path and reduced time that air masses spent above the mixed layer.
The basis of this clustering method is that it tags each time step as belonging to a certain cluster. In this study, we use hourly resolution, and therefore each hour with available SMPS data is tagged as belonging to one of the eight clusters. Once this is done, essentially classifying the aerosol observed during that hour as one of eight aerosol types, one can begin building profiles of these aerosol types using concurrent measurements of meteorological parameters, chemical and physical aerosol properties, and so on. Figure 6 shows this for the most relevant parameters in this study: scattering coefficients from the nephelometer at 525 nm, absorption coefficients from the aethalometer at 520 nm, and total incoming solar irradiance. Each row represents data observed when the named cluster was tagged, thereby illustrating a profile of the cluster. The scattering and absorption coefficients directly relate to the physical properties of the aerosol; for example, it can be inferred that the Aged cluster will strongly scatter and absorb sunlight compared to the Nucleation cluster. In contrast, the total incoming solar irradiance shown in Fig. 6 is a meteorological parameter, as opposed to a physical property of the sampled aerosol. It is included to illustrate how much incoming solar radiation is typically present when each aerosol cluster is observed. For example, when the UF Bimodal cluster is tagged, the median irradiance measured is 194 W m− 2 (median absolute deviation: 120 W m− 2), and it can be seen from the histogram distribution that it is observed rather equally during periods of high and low irradiance. In comparison, when the Aged cluster is tagged, the median irradiance measured is 0.79 W m− 2. This reveals that when the Aged cluster is present, it is usually dark. This is an important consideration when assessing its direct radiative impact.
Looking further into the results in Fig. 6, one can infer relationships between the clusters’ physical properties. Scattering coefficients are observed to relate strongly to the aerosol size, which is as expected from Mie theory. The anthropogenic clusters Haze and Aged typically scatter significantly more light than the clusters representing aerosols with lower diameters. This is also compatible with knowledge of the chemical composition: Haze and Aged are both known to be anthropogenic, and typically contain high concentrations of sulphate aerosols, which are highly scattering. With respect to absorption, the two anthropogenic clusters contrast even more strongly against the other six, dominating the highly absorbing aerosol sampled in the aethalometer. This is compatible with the knowledge of high BC concentrations observed in both the Haze and Aged clusters; BC is the primary absorbing aerosol at this site. The Acc. Bimodal cluster is seen to have low absorption coefficients, more similar to the other five clusters with lower peak diameters. This is in contrast to what was found by Lange et al. (2018; 2019) regarding equivalent black carbon (eBC) concentrations (which correlate directly with absorption coefficients), who found that this cluster typically had higher eBC concentrations compared to all UF clusters. The eBC concentration measurements in this previous study were taken in 2011 and 2012, so this could potentially reflect a long-term decreasing trend. However, with the current dataset, this cannot be confirmed with any significance. The irradiance histograms shown in Fig. 6, unlike the scattering and absorption coefficients, do not directly relate to a physical property of the aerosol cluster. Instead, these data reflect the time of year that these aerosols are observed at Villum. The typical amount of incoming radiation observed when the aerosol is present is a vital quantity when assessing its direct radiative forcing potential.
The typical amount of light that an aerosol interacts with in a month can be calculated using the mean optical properties of each aerosol cluster along with its monthly occurrence. This will demonstrate for which months there are highly scattering or absorbing aerosols in the atmosphere at Villum. These results are shown in Fig. 7, where the aerosol clusters are regrouped as previously described into the Anthropogenic, Mixed, Biogenic and Background clusters. It is clear that the anthropogenic aerosols at Villum represent the majority of scattering and absorbing material in the atmosphere, and this is even more pronounced for absorbing than for scattering. For scattering, there is a substantial contribution during the summer from the Mixed and Biogenic cluster groups, though these aerosols are not observed to absorb light as readily.
While Fig. 7 illustrates the scattering and absorbing coefficients of atmospheric aerosols throughout the year at Villum, it does not represent when sunlight is actually being scattered or absorbed. This is because it does not take into account the incoming solar irradiance, which varies immensely over the year at these latitudes. The PDRF is therefore finally calculated using Eq. 2, describing interaction of sunlight with specific aerosol types. This quantity was calculated as a monthly mean for each cluster in each month, separately for scattering and absorption, and the results are shown in Fig. 8. It demonstrates that aerosols have the highest potential to directly force the radiative balance of the Arctic in late spring and early summer. Specifically, the months of April, May, June and July experience the highest values of PDRF due to both the scattering and absorption of incoming sunlight, with the peak for both processes occurring in May. However, PDRF due to scattering is seen at relatively similar levels during these four months, whereas the PDRF due to absorption is seen to be significantly higher in the months of April and May compared to June and July. This illustrates the effect of the aerosol SSA changing throughout the year, and therefore the relative strengths of the direct warming and cooling effect from aerosols can also be expected to have a seasonal variation. It can be seen by comparing the scattering and absorbing axes in Fig. 8 that aerosols at Villum tend to scatter much more sunlight than they absorb throughout the year, such that monthly mean SSAs calculated from these data vary only between 0.92 and 0.96 throughout the year. However, this variation is observed in a clear seasonal pattern: the monthly mean SSA is between 0.92 and 0.93 between December and April, then increasing to 0.96 between June and August, before falling again in the autumn. Assessing the direct warming or cooling nature solely from measurements of scattering and absorption is complex, but it can be predicted that, broadly speaking, the direct radiative effect from aerosols at Villum is more cooling in the summer months.
Table 1 shows the PDRF results for each cluster and cluster group, shown as mean values across all months. These are separated into the PDRF from scattering and absorption of sunlight. For example, the Haze cluster has a PDRF of 0.062 (0.039) mW m− 3 from scattering, which is a relative contribution of 31%, and a PDRF of 0.0018 (0.0009) mW m− 3 from absorption, accounting for 43%. The Haze cluster therefore represents the largest contribution in both cases, though notably larger for absorption. The Aged cluster has a lesser effect, but important especially in the case of absorption. The Mixed clusters, which have been proposed to both have Arctic marine and anthropogenic sources (Dall'Osto et al., 2018; Lange et al., 2019), by comparison contribute significantly more to the scattering of light than the absorption, and the Biogenic clusters contribute more evenly to both processes. Overall, this shows that anthropogenic aerosols do have a larger direct radiative effect than biogenic aerosols at Villum, but the difference between their impacts is not as large as initially expected. This is primarily due to the anthropogenic aerosols being abundant most often during the winter, when the incoming solar radiation is low.
Table 1
Mean contributions to PDRF across all months from scattering and absorption at ground level from each of the eight clusters. Summed contributions from the cluster groups Background (BG), Biogenic, Mixed and Anthropogenic are also shown.
Cluster name
|
Pri.
|
Nuc.
|
Bur.
|
Nas.
|
UFB.
|
Acc.B.
|
Haze
|
Aged
|
Scattering / mW m-3
|
0.022
|
0.011
|
0.013
|
0.017
|
0.027
|
0.036
|
0.062
|
0.015
|
|
± 0.010
|
± 0.005
|
± 0.006
|
± 0.009
|
± 0.015
|
± 0.019
|
± 0.039
|
± 0.008
|
Absorption / mW m-3
|
0.0010
|
0.0007
|
0.0006
|
0.0006
|
0.0008
|
0.0016
|
0.0054
|
0.0018
|
|
± 0.0005
|
± 0.0003
|
± 0.0003
|
± 0.0003
|
± 0.0004
|
± 0.0008
|
± 0.0034
|
± 0.0009
|
Group name
|
BG
|
Biogenic
|
Mixed
|
Anthropogenic
|
Scattering / mW m-3
|
0.022
|
0.041
|
0.063
|
0.077
|
|
± 0.010
|
± 0.004
|
± 0.012
|
± 0.020
|
Absorption / mW m-3
|
0.0010
|
0.0019
|
0.0024
|
0.0072
|
|
± 0.0005
|
± 0.0002
|
± 0.0005
|
± 0.0017
|
The effect of modifying these values for ambient temperature, pressure and RH has also been broadly assessed. The results can be converted from standard temperature and pressure (STP) to ambient temperature and pressure simply, and for Villum, these PDRF correction factors were found to lie in the range from 1.05 (during summer) to a maximum of 1.20 (in spring). RH at Villum tends to be lower and more stable from January to April, with a mean of 70.0% and standard deviation (SD) of 7.4%. It becomes higher and more variable from May to August, with a mean of 77.5% and SD of 13.8%. A study by Zieger et al. (2010) presented measured scattering enhancement factor f(RH) values at Ny-Ålesund, Svalbard, both in the case of an air mass that was heavily influenced by sea salt, and an air mass that was not. A preliminary estimation of f(RH) at Villum can be made using these results, adopting the same parameterisation of f(RH) that only uses the RH and a single parameter γ. Here, γ was taken to equal 0.60, as was found by Zieger et al. in the case of a non-sea salt air mass with high RH. These calculations are very sensitive when RH is above 95%, as the calculated f(RH) goes to infinity as RH goes to 100%. The monthly means of f(RH) at Villum from these calculations, noting the sensitivity to high RH, are 2.1 in March and April, rising to 4.0 in July and falling to 3.1 in September. Median monthly f(RH) values all lie between 2.1 and 3.2. These calculations indicate that the enhancement from RH would increase scattering PDRF of the aerosols present in summer more than those in spring, therefore amplifying the effect of the biogenic and mixed clusters (Nucleation, Bursting, Nascent, UF Bimodal and Acc. Bimodal) compared to the anthropogenic clusters (Haze and Aged). The conversion to ambient temperature and pressure will work to counteract this slightly, by increasing the PDRF values more so in spring than in summer. Currently, exact calculations of f(RH) are not available, but from this qualitative assessment, it can be inferred that once ambient conditions are accounted for, the relative total effect of anthropogenic aerosols (compared to biogenic aerosols) will be reduced.
The method of using k-means clustering to classify aerosol types in the Arctic is useful because it provides a complete bottom-up method of investigating the impacts of aerosols on climate. In essence, the method investigates the aerosols directly as they appear at Villum, describing the total aerosol population and distinguishing the aerosol present by its particle number size distribution, which is a physical parameter that influences direct radiative forcing. This is a more direct method than analysing concentrations of various known atmospheric pollutants divided into chemical species. For example, BC is a species with a direct warming effect, but it is often co-emitted or co-occurring with many scattering aerosol species, resulting in internally mixed aerosol particles. The cooling effect of these mixed species makes the quantification of overall radiative forcing from BC emissions more nuanced (Bond et al., 2013). It has been shown that BC concentrations at Villum are positively correlated (R2 = 0.729) with sulphate concentrations (Massling et al., 2015), and the co-transported sulphate will scatter sunlight efficiently, leading to a cooling effect (IPCC, 2013). The polluted air masses containing elevated BC and sulphate concentrations will therefore more likely lead to a direct cooling effect. This is seen intuitively in the PDRF results for the anthropogenic clusters, in which they are seen to have much stronger scattering PDRFs. This implies a direct net cooling effect of the Arctic haze aerosol, which has been confirmed in previous studies that show the anthropogenic aerosol radiative forcing effect to be negative in the Arctic troposphere due to strong negative radiative forcing from sulphates (Breider et al., 2017).
In Pernov et al. (2022), it was found that the occurrence of the Nucleation cluster has been increasing during the summer months of the past decade, most likely owing to changing transport patterns into Villum (Pernov et al., 2022), and consequently, increased time that the air masses spent over open ocean before arriving at Villum. Following the results from our PDRF study, this trend may have a measurable direct radiative effect, despite the small aerosol particle size that is not normally associated with a strong direct atmospheric forcing. It is therefore complex to fully illustrate how even direct forcing may change in the future, and how much of this change can be attributed to anthropogenic aerosol emissions. Arctic warming leads to a greater area of open ocean, from which aerosol precursors such as DMS are emitted, leading to new particle formation. These aerosols would be biogenic in nature, but amplified by anthropogenic warming, making the picture more nuanced. It is important that as the Arctic continues to change in the coming years, we do not limit ourselves definitions of anthropogenic aerosols relying solely on their chemical composition, and instead strive to understand the full extent to which anthropogenic activity will influence Arctic climate change.
It was also demonstrated by Pernov et al. (2022) that the occurrence of the Pristine cluster has been increasing in the autumn months, seen alongside an increase in accumulated precipitation along the trajectory path and a decrease in time the air masses spent above the mixed layer. This increase in the Pristine cluster occurrence is not likely to have a significant direct radiative effect based on the PDRF results displayed here, but if we continue to see changes in the amount of accumulated precipitation, this will disproportionately affect the larger clusters. Since wet removal is more effective in the accumulation mode range relative to other size ranges, increased precipitation may lead to a decrease in the occurrence of the anthropogenic aerosol clusters, though the impact would depend on the season that these precipitation changes occur. Additionally, removal from the atmosphere by precipitation does not mean removal from the climate system, and aerosols deposited onto snow and ice surfaces can have significant forcing effects in the Arctic (Masson-Delmotte et al., 2021; Quinn et al., 2008; Sand et al., 2016). It is therefore necessary to perform further studies on the effects of direct forcing due to aerosol deposition in the Arctic.