Sun-photometers measure direct sun spectral irradiance, which is used to derive the AOD at different wavelengths. AOD is an indicator of the amount of aerosol in the atmospheric column. In addition, the Ångström exponent (AE) can be calculated from the spectral dependence of the AOD for the range of wavelengths available. AE is related to the aerosol particle size . A mean value of AE equal to 1.3 is observed for the average continental aerosol , while values close to 0 indicate coarse particles.

Different models of sun-photometers are distributed around Svalbard. A precision filter radiometer PFR; and a sun photometer 1 automatic SP1A; are located at the Sverdrup and the Zeppelin Observatory respectively. In Hornsund, there is a CE318-T (Cimel Electronique) belonging to the Aerosol Robotic Network AERONET;. CE318-T instruments can also measure sky radiances at various wavelengths, which are then used together with simultaneous AOD measurements to retrieve microphysical and optical aerosol properties using an inversion algorithm . Finally, a Microtops II manual photometer onboard the R/V OCEANIA was used to complement the fixed location measurements with AODs from different locations around the Svalbard Archipelago. This instrument is part of the maritime aerosol network from AERONET , and is used to provide ship-borne AOD measurements. The Microtops II sun photometer technical parameters as well as calibration techniques can be consulted in and . During the 2019 study, each measurement slot lasted for about 1 min. In order to reduce potential human errors (e.g. sun pointing error), 5 scan slots with each measurement and the scan with the lowest standard deviation was further regarded for the processing analyses. For this study only data recorded when the vessel was 50 km distance or less from Ny-Ålesund have been used.

The sun-photometer dataset used in this study from includes the AOD at several wavelengths (depending on the instrument) and the AE derived in the range 380–870 nm. The comparability of the AOD from the different sun-photometers was already assessed within the mentioned projects. For this analysis, hourly and instantaneous values of AOD at 500 nm and AE in the 380 and 870 nm range have been used. In addition, particle volume size distribution (PVSD) and single scattering albedo (SSA) have been directly obtained from the AERONET website (https://aeronet.gsfc.nasa.gov/, last accessed: 12 January 2026). All AERONET data used correspond to level 1.5 products (version 3). These retrieved properties have been filtered by the residual obtained by the inversion algorithm; inversions with residuals bigger than 10 % have been rejected, which ensures the quality of the retrievals. For the period of study the PVSD and SSA are only available at Hornsund.

In-situ aerosol optical and microphysical properties are measured continuously at ZEP and GAL. At these two sites the aerosol absorption coefficient (βabs) and the aerosol scattering coefficient (βsca) both at 530 nm, and the particle number size distribution (PNSD) between 10–700 nm are recorded. In addition, the single scattering albedo at the same wavelength (SSA530) can be calculated as the ratio between the scattering coefficient and the scattering coefficient plus the absorption coefficient.

At ZEP, βabs was originally measured at 637 nm using a multi-angle absorption photometer (MAAP, Thermo Fisher Scientific Inc., Germany, model 5012), and after converted to 530 nm (assuming an AE of 1). βsca was recorded using a nephelometer (Ecotech Pty Ltd., Australia, model Aurora 3000), measuring at 450, 550 and 700 nm, and converted to 530 nm using the scattering Ångström exponent (SAE). The PNSD was measured using a twin differential mobility particle sizer (DMPS) system, consisting of two custom built Hauke-type differential mobility analysers (DMAs) each connected to a condensational particle counter (CPCs a TSI 3010 CPC and TSI 3772) used to count the size discriminated aerosol particles. All samplings were conducted using a whole-air inlet in accordance with sampling procedures laid out by the World Meteorological Organization / Global Atmosphere Watch (WMO/GAW) guidelines.

At GAL, βabs was recorded with a Particle Soot Absorption Photometer (PSAP, Radiance Research) at three wavelengths (467, 530, 660 nm), while βsca was measured at 530 nm with a Nephelometer (Radiance Research). Raw data were corrected, averaged over 1 h, and converted to standard temperature and pressure conditions as described by . The aerosol PNSD was continuously measured at GAL using a combination of a Scanning Mobility Particle Sizer (SMPS) model TSI 3034 and an Aerodynamic Particle Sizer (APS) model TSI 3321, both from TSI seeand references within.

The lidar measurements were performed by the Koldewey Aerosol Raman Lidar (KARL) at AWIPEV. This system has a Spectra 290/50 Nd:YAG laser, which emits a laser pulse at 355, 532 and 1064 nm with 50 Hz and 200 mJ per colour vertically into the atmosphere. A telescope with 70 cm diameter collects backscattered photons with a field of view of about 2 mrad. Data recording is done via Hamamatsu photomultipliers and Licel transient recorders with a raw resolution of 7.5 m and 82 s. Both, analogue and photo counting signals, are recorded. Full overlap is reached at about 700 m altitude. In addition, KARL receiver can separate the backscattered signal into parallel and perpendicular components at 532 nm, which allows the calculation of the volume aerosol depolarization (δaer) as the ratio between the perpendicular and the parallel signal. This magnitude is useful to distinguish between spherical and non-spherical particles. A more detailed description of KARL can be consulted in .

The aerosol backscatter profiles at the three available wavelengths have been calculated with 60 m and 600 s resolution according to Klett with clear sky approximation (aerosol backscatter small against molecular backscatter at altitudes >22 km) and the choice of a prescribed lidar ratio of 70, 45 and 45 sr for the wavelengths of 355, 532 and 1064 nm, respectively. The lidar ratios for 355 and 532 nm have been verified by backscatter values in the clear troposphere. An uncertainty of ±10 sr for the lidar ratio has been estimated, giving rise to about 10 % uncertainty in the derived aerosol backscatter. For the 1064 nm wavelength the uncertainty is dominated by the assumed backscatter >22 km as a boundary condition, such that also 10 % uncertainty at this wavelength is realistic. Data points in time and altitude which were covered by clouds have been removed to not bias aerosol properties.

The backscatter profiles have been used to calculate the colour ratio (CR), for each (λ1,λ2) combination of the three laser wavelengths as Eq. (): 1CR(λ1,λ2)=βaer(λ1)βaer(λ2) with λ1<λ2 To adjust for measurement uncertainties, an uncertainty of ±5 % of the CR has been considered. The CR can also be theoretically estimated using the Mie calculator from Library for Radiative Transfer – libRadtran (http://libradtran.org/doku.php, last access: 21 November 2024) . For this calculus we have assumed a log-normal particle size distribution with a width of σ=1.1 and a refractive index of m=1.5+0.05i; which is a typical refractive index for Arctic aerosol . According to the previous work from the choice of the refractive index will probably not critically affect the solution. The best fit between the theoretical and observed colour ratios for each combination of the available wavelengths 355, 532 and 1064 nm has been used to estimate the height-dependent effective radius of the aerosol. This way, vertical profiles of the aerosol effective radius have also been obtained.

The set-up of surface radiation observations established at AWIPEV includes direct normal irradiance (DNI) by a CHP1 pyrheliometer, diffuse (DIF), global and reflected shortwave horizontal solar irradiance by CMP22 pyranometers, as well as up and downward longwave radiation by CGR4 and Eppley PIR pyrgeometers. All instruments follow the World Radiometric Reference (WRR) standards for solar radiation data and the World Infrared Standard group (WISG) guidelines for longwave radiation. Additionally, the data contributes to the Global Climate Observing System (GCOS) Reference Upper-Air Network (GRUAN). The surface radiation measurement setup, data retrieval, and quality control follow Baseline Surface Radiation Network (BSRN) standards. More information is given by and , among others. All measurements were performed automatically with a 1 min resolution and can be found in (10.1594/PANGAEA.914927). These data are available since 2006.

The hourly anomaly (ΔDNI) for the hourly DNI values has been calculated with respect to a reference value (DNIref), as shown in Eq. (): 2ΔDNI=DNI-DNIref The DNIref values have been calculated for each month and each hour as the mean of all DNI hourly values recorded at that specific hour during that month from the 2006 to 2020 reference period. As a result, 24 reference values are obtained for each month. This approach is used because, due to the high frequency of cloudy conditions in the region, the amount of available data to construct a reliable reference is limited.

In addition, the diffuse ratio DifR; see has been calculated following Eq. (): 3DifR=DIFDIF+DNI⋅cos⁡(SZA) This ratio can take values from zero to unity, with small values related to cloud-free cases. Only data showing an hourly DifR <0.3 have been used, trying to reject the influence of clouds on the results.

CALIOP (Cloud-Aerosol LIdar with Orthogonal Polarization) is the main instrument onboard CALIPSO satellite. CALIPSO was launched in April 2006 and after 17 years of operation, it ended its mission on 1 August 2023.

CALIOP consists of a two-wavelength lidar that uses a Nd:YAG laser to emit pulses at 1064 and 532 nm simultaneously. The laser pulse repetition frequency is 20.16 Hz, which allows a horizontal resolution of 335 m at the ground. The outgoing 532 nm pulses are linearly polarized. A polarization beam splitter separates the components of the 532 nm backscatter signal polarized parallel and perpendicular to the plane of the outgoing beam, which are collected by two detectors. The instrument operates continuously, providing observations during both day and night.

In this work it has been utilized the level 2 (L2) profile data products (LID_L2_05kmAPro-Standard-V4-51) for aerosol extinction at 532 nm, which is given at a spatial resolution of 60 m vertically and horizontal averaging length of 5 km along the satellite track. The L2 layer data product (LID_L2_05kmALay-Standard-V4-51) includes the tropospheric and stratospheric AOD at 532 nm, as well as the L2 vertical feature mask (LID_L2_VFM-Standard-V4-51) for aerosol and cloud classifications. The classification (read out from the vertical feature masks of the profile) and the lidar ratios, which are used by the CALIOP retrieval teams for conversion of the measured backscatter into the estimated extinction have been included in Table S1 in the Supplement.

CALIPSO data have been downloaded from the ICARE Data and Services Center (http://www.icare.univ-lille1.fr/, last access: 23 August 2023). For this analysis, data within 50 km distance from Ny-Ålesund have been extracted, and the closest extinction profile has been utilized (see Fig. S1 in the Supplement).

To investigate the possible origin of BC (black carbon) aerosol air-masses reaching the Arctic, the Lagrangian particle dispersion model FLEXPART version 10.4 has been used . The model has been driven by hourly reanalysis meteorological fields (ERA5) from the European Centre for Medium-Range Weather Forecasts (ECMWF) with 137 vertical levels and a horizontal resolution of 0.5° × 0.5° . Computational particles have been released in FLEXPART at various heights (0–100, 100–500, 500–4000, 4000–6000, 6000–10 000 m) from the receptor (considered the Zeppelin Observatory) and tracked backward in time in FLEXPART's “retroplume” mode. Simulations have been run 30 d back in time, that is a sufficient time to include most BC sources arriving at the station, given a typical BC lifetime of 1 week .

Gravitational sedimentation for spherical particles is included and differs between trajectory models due to its ability to simulate dry and wet deposition of aerosols , turbulence , unresolved mesoscale motions , and convection . In the simulations of this work it has been assumed that BC has a density of 1500 kgm-3 and follows a logarithmic size distribution with an aerodynamic mean diameter of 0.25µm and a logarithmic standard deviation of 0.3 .

Anthropogenic emissions have been adopted from the latest version (v6b) of the ECLIPSE (Evaluating the CLimate and Air Quality ImPacts of ShortlivEd Pollutants) emission inventory, an upgraded version of the previous version . This inventory includes industrial combustion emissions, emissions from the energy production sector, residential and commercial emissions, emissions from waste treatment and disposal sector, transportation, shipping activities and gas flaring emissions. Biomass burning emissions have been adopted from the Copernicus Global Fire Assimilated System (CAMS GFAS) because this product provides an estimation of the injection altitude of the fire emissions. Smoke particles consist mainly of organic carbon (OC) and a small amount of BC . These components can absorb radiation, warming the surrounding air and inducing upward motion that lifts the aerosol ; i.e., the so-called self-lofting mechanism. GFAS uses two different models to calculate the injection height, based on satellite observed fire radiative power and ECMWF forecasts of key atmospheric parameters . Radiative self-lofting in global models, such as FLEXPART, is not considered yet, but the scientific basis now exists with the ECMWF radiation scheme (ECRAD) that computes shortwave heating rates of an imposed smoke layer . However, online implementation of this module in global models might be demanding, due to the need of remote sensing data as input parameters (e.g., CALIOP aerosol observations, MODIS aerosol optical depth retrievals etc.). A more detailed discussion of the potential mechanisms responsible for self-lofting is provided in the FLEXPART results section (see Sect. ).

FLEXPART simulations have been performed every 3 h during the studied period. The FLEXPART output consists of a footprint emission sensitivity, which results in a modelled concentration at the receptor (Zeppelin Observatory), when coupled with gridded emissions from an emission inventory (like the ones described before). The emission sensitivity expresses the probability of any release occurring in each grid-cell to reach the receptor. The modelled concentrations can be displayed as a function of the time elapsed since the emission has occurred (i.e., “age”), which can be shown as an “age spectrum”. Detailed source contribution maps have been calculated, showing which regions contributed to the simulated concentration.

The source contributions to receptor BC have been derived by combining each gridded emission sector (e.g. gas flaring, transportation, waste management, etc.) from an emission inventory with the footprint emission sensitivity. Calculations for anthropogenic sources and open biomass burning have been performed separately. This has enabled the identification of the exact origin of BC and allowed for quantification of its source contribution.

Figure  shows the hourly AOD values at 500 nm (AOD500) recorded in and around Svalbard from June to September 2019. The mean daily AOD500 values calculated considering all the available period (2002–2020) are included as reference (see dashed line in Fig. ). It can be seen that most of the summer of 2019 (considered from June to September for this study, if nothing else is specified) presented consistently higher values of AOD with respect to past and following years (2002 to 2020 period). AOD500 in 2019 was similar to the reference values during June, but since 6 July, when an unprecedented increase in AOD occurred, it remained high for the rest of the summer. The median AOD500 from July to September for the 2002–2020 reference period is 0.048, but this value is 0.162 when it is calculated only with 2019 data; it means that AOD500 in this period was more than three times higher than during the reference. Regarding the AE, which is related to the size of the aerosol, it did not show a significant difference with respect to the reference period as can be observed in Fig. S2 in the Supplement. The days on which the AOD500 values were higher than the summer average plus three times the standard deviation have been identified as days when an aerosol intrusion was occurring in Svalbard. This way, three different aerosol events have been identified during the summer of 2019 in the following periods: 6 to 10 July, 25 to 28 July and 6 to 17 August, which have been named C1, C2 and C3 respectively because they are observed in the atmospheric column (C). During these three events the mean AOD500 values were 0.28, 0.19 and 0.23 respectively. Moreover, the periods between C1 and C2, and between C2 and C3 presented mean AOD500 values of 0.15 and 0.17 respectively; this mean value was 0.15 from the end of event C3 (17 August) to the end of August. Finally, during September the mean AOD500 was 0.11.

The columnar PVSD and SSA values retrieved for the analyzed period are shown in Fig. , highlightening those corresponding to the identified events. Table  summarizes the mean AOD and mean parameters of the PVSD, observed during the periods identified with aerosol event, and also in the adjacent periods during summer 2019. Figure a–d show a high aerosol concentration in fine mode (Accumulation mode) during July and August; this also happens for September, when no aerosol events have been identified. Aitken mode and coarse mode particles are almost absent; the highest variability in PVSD during this summer is observed in the Accumulation mode, when the maximum volume concentrations of the PVSD oscillate from 0.01 to 0.04 µm3µm-2 in all the period.

In July, the mean effective radius of the fine mode (〈ReffF〉) was 0.145 µm for the days with no aerosol event, and 0.156 and 0.155 µm in events C1 and C2, respectively. However, while C1 showed a slim fine mode distribution with one maximum, C2 generally showed a main maximum at radius about 0.1 µm and a second maximum at about 0.3 µm. This might indicate the presence of particles of different origin during C2.

In August, the fine mode distribution was wider and with a gaussian shape, presenting in general lower concentrations at each radius. Event C3 presented the highest 〈ReffF〉 (0.182 µm) of the three identified events. This might either point to different aerosol source or be an aging effect, such that the mean time between the aerosol emission and arrival over Spitsbergen might had been longer in August compared to July. These larger radii made AE values lower in event C3 (see Fig. S2) than in the other events even when all were dominated by fine particles . Nevertheless, days not identified as aerosol intrusion showed a distribution similar to the ones observed during event C2, with a mean 〈ReffF〉 of 0.156µm.

In September only two inversions are available (on the 5^th). These presented a 〈ReffF〉 of 0.142 µm and, again, a similar shape, but with a lower concentration at each radius than the ones observed in event C2 (July). However, the coarse mode seems slightly enhanced and these particles, larger than 2 µm, should sediment down quickly. As similar observations are made throughout July, August and September, this might indicate a steady supply of new aerosol (because of the life-time of these coarse particles). Alternatively, due to the low concentration of the coarse mode, it could just indicate the presence of a steady stratospheric aerosol starting from event C2, which was combined with new aerosol with the arrival of new aerosol events. Therefore it can be hypothesized that we are facing a combination of different aerosol sources.

The SSA, shown in Fig. e–h, presents values close to 1 for all wavelengths, indicating a low aerosol absorption even during the aerosol events. Only a few background cases showed values lower below 0.96 in the infrared, but these cases do not show up for the identified events, indicating less variability. The high SSA and no spectral dependency during the events could be caused by the presence of BB aged after long transport or might have been emitted under low combustion efficiencies . This kind of aerosol agrees well with the predominant fine mode with an effective radius larger than usual .

Figure a and b report, respectively, the time series of the aerosol absorption (βabs) and the scattering (βsca) coefficients at 530 nm measured in-situ at GAL and ZEP in summer 2019. During the full period, βabs and βsca showed the following mean values: 0.12±0.16 and 1.83±2.04 Mm⁻¹ at GAL and 0.10±0.21 and 2.43±2.97 Mm⁻¹ at ZEP, respectively. The average SSA₅₃₀ was 0.92±0.06 and 0.96±0.06 at GAL and ZEP, respectively. The seasonal averages for the summers of 2018 and 2020 together with the ones from 2019 are included in Table . It shows that 2019 did not show significant differences at ground level in these type of measurements.

The Babs530 at GAL and ZEP correlate with a Pearson coefficient (R) of 0.71. A lower correlation is observed for Bsca530, with an R of 0.34, although increasing and decreasing trends have been generally observed at the same time at the two sites. This low R value points to a more local origin for scattering aerosol. While absorption is predominantly associated with long-range transport, the scattering in summer appears to have a strong marine influence. Consequently, when wind directions differ between the two sites, the stations may perceive this local source differently. However, whenever a strong event comes, the scattering increases in both sites. Therefore, it is difficult to use only one station for comparison with climate models.

Aerosol events at the surface, associated with relatively high aerosol loading at the ground, have been identified when both Babs530 and Bsca530 exceeded, at least in one station, the average values recorded over three summers (2018–2020, from June to September), plus three times the standard deviation (horizontal lines in Fig. ). This way, three events of high aerosol load on the surface (S) have been identified on 5 to 9 July, 30 August, and 12 September. The first event (5 to 9 July) overlaps in time with event C1, therefore we will refer to it as CS1 instead of C1 from now on. The last two events have been named events S1 and S2 respectively. No data from remote sensing observations were available during these two events due to cloudy conditions.

The increase in absorption coefficient observed during the three surface events points to the impact of air masses affected by combustion emissions. The SSA₅₃₀ observed during events S1 and S2 were 0.89±0.04 and 0.92±0.05, respectively, reaching values as low as 0.81 (event S1) and 0.76 (event S2). For event CS1 the mean SSA₅₃₀ was 0.95±0.01 with lowest instantaneous values equal to 0.92. The slightly lower SSA₅₃₀ values observed during the events S1 and S2 might indicate that the aerosol particles emitted by combustion traveled a shorter distance before reaching Svalbard compared to event CS1, although the differences are within the standard deviations.

Figure  reports the averaged particle number size distribution during the surface events and for all available measurements in the summer of 2019. The PNSD values observed at the two sites align with the multi-year average distributions reported by for the summer months. Regarding GAL observations, the seasonal average of PNSD exhibits a bimodal pattern, with peaks occurring at approximately 30 nm (Aitken mode) and 140 nm (Accumulation mode). The Aitken mode predominates over the Accumulation mode, primarily due to new particle formation events that increase the number of ultrafine particles . Additionally, wet scavenging removes effectively Accumulation mode particles in summer both in the Arctic and during their transport to the Arctic . The surface events were still characterized by a bimodal distribution, but in these cases the particle number of the Accumulation mode was comparable or higher than the Aitken mode population, indicating that the site was reached by more aged aerosol populations. The increase in Accumulation mode particles agrees with the rise in scattering coefficients during the three events compared to the average. The PNSD at ZEP does not show a clear bimodal structure for either of the events. In these cases, the peak of the distribution lies between the Aitken and Accumulation modes, generally showing higher concentrations than GAL during events S1 and S2. Therefore, the surface events were perceived slightly differently at the two sites.

A time series with the CALIOP aerosol extinction profiles (at 532 nm) available in summer 2019 is shown in Fig. . The equivalent Fig. S3 in the Supplement indicates the classification of the layer aerosol type. The tropospheric and stratospheric AODs retrieved from CALIPSO for each profile are also included in the upper panel. For event CS1 a strong extinction is observed between 3 and 5 km a.g.l. Starting from 15 July a stratospheric aerosol layer at about 12 km a.g.l. is observed, with the corresponding increase in the stratospheric AOD. An aerosol layer between 10 and 15 km a.g.l. is also observed in the following extinction profiles, with variable intensity. During C2 and C3, no clear signature of aerosols is observed in the troposphere, maybe related to the presence of tropospheric clouds, which would mask the aerosol when observing from a satellite.

Unfortunately, KARL measurements are only available for 4 d in the summer of 2019 due to cloudy conditions and safety regulations of the instrument. The layers observed with KARL were temporally quite constant on each day (see Fig. S4 in the Supplement), therefore, the daily averaged backscatter profiles have been calculated. These are shown in Fig. . It is observed an increased backscatter between 10 and 16 km a.g.l. with several layers through August, but on 17 September the backscatter slowly decreases and becomes more homogenous with height. Only one of these days corresponds to a day identified with aerosol event, 11 August (Event C3). During this day, a high backscatter coefficient at 532 nm up to about 0.8 Mm⁻¹, with several layers, is observed throughout the entire troposphere, as well as in the stratosphere up to nearly 16 km a.g.l. Particularly, the layer just around 10 km a.g.l. observed with KARL correlates very well in altitude with the increased backscatter profile measured by CALIOP on the same date. With CALIOP it is also observed some extinction around 14 km a.g.l., which correlates with the stratospheric layers observed with KARL. Since the vertical and temporal resolution from both instruments is very different, we do not expect a closer agreement.

The daily averaged backscatter profiles have been used to estimate the height-dependent effective radius of the aerosol for each day, as described in Sect. : an a priori refractive index and a one-modal log-norm distribution are considered. The calculated effective radii for the four observation days are shown in Fig.  with a height resolution of 60 m. Due to the incomplete lidar overlap, the plot starts at 1 km altitude. This figure shows that for all days and the entire troposphere the background aerosol had radii values at around 0.1 µm, while the pronounced layers reveal coarser aerosols with radii from 0.5 to 0.8 µm. Furthermore, it can be seen that the effective radius for August was very similar, while on 17 September, between 7 and 9 km a.g.l., the effective radius was about 0.2 µm smaller. This suggests that the largest particles were removed from this layer, possibly due to sedimentation. Depolarization measurements at 532 nm (δaer<3%) were lower than 3 % for the four available days. These low values are typical for a spherical shape. Hence, we conclude that the large particles in the troposphere were likely droplets. Hygroscopically grown particles act directly as Cloud Condensation Nuclei (CCNs) and grow in the water phase, but do not serve as Ice Nucleating Particles (INPs). This can be seen in the low depolarization throughout the entire studied atmosphere. Furthermore, this process happened quickly, because otherwise the gap in effective radius in the troposphere between 0.2 to 0.5 µm would not exist. At an altitude of about 1 km, a distinct aerosol type appeared on 21 August. Only on this day at that height, the depolarization reached its maximum with 2.5 %, indicating slightly less spherical shape than in the layers above.

In the stratosphere (from 9–10 km) a different behavior is observed. A distinct aerosol signature appeared, characterized by an effective radius of approximately 0.2 µm, although dispersed values ranging from 0.2 to 0.4 µm are also observed, particularly between 12 and 17 km a.g.l. When comparing the results from KARL with the retrievals from photometer observations (see Sect. ), the effective radii are similar. It must be noted that generally the aerosol backscatter and extinction decreases with altitude (not shown here). Hence, in Fig.  the lower altitudes needs to be “weighted” more than high altitudes for a comparison with photometers. By comparison of Figs.  and it is obvious that the predominant mode around 0.1 µm (Accumulation) is tropospheric, while the second maximum (in Fig.  weaker) around 0.2 µm is stratospheric. Particles >0.5 µm seem to be less frequent in the lidar, compared to photometer. This is likely due to the selection effect that the lidar only operates during clear sky conditions. Hence, clouds are underrepresented in the lidar data. Still, some hygroscopically grown aerosol might be visible in the lidar data which we relate to the range of effective radii between 0.55 to 0.8 µm. Further, the weak coarse mode with particles >1 µm is missing in the lidar, as for this instrument we only assumed a one-modal aerosol distribution.

The footprint emission sensitivity together with the corresponding source contribution to BC and the continental contribution to BC at each model altitude for events CS1, C2 and C3 are depicted in Fig. . These parameters have been integrated, for each event, by the temporal duration determined in the previous section. The source contribution panels in the middle column of Fig.  indicate high concentrations of biomass burning BC during these events. Event CS1 (Fig. , top row) is the one that presents the highest concentration, with maximum values of 140 ng m⁻³ at 500 m a.g.l., but also a high concentration at 100 and 4000 m a.g.l. For this event the footprint emission sensitivity indicates that the BC was coming mainly from wildfires in central Siberia. The use of a continental mask indicated that the contribution of BC from fires in Russian territory was more than 95 % at heights between 0 and 4 km a.g.l. (Fig. ).

Similarly, the highest BC concentrations during event C2 (Fig. , middle row) were obtained at 4000 m a.g.l. However, modelled concentrations reached a maximum of only 40 ng m⁻³. The source contribution to column BC shows that BC had two branches of origin, one from Siberia and another from North America. Wildfires contributed over 70 % of the BC mass arriving at the station at altitudes up to 4 km, and above 97 % of the mass was coming from Russia.

Finally, for event C3 (Fig. , bottom row), the modelled BC concentrations at different heights reached a maximum at 4000 and 500 m a.g.l., with almost 70 and 60 ng m⁻³ respectively. While the footprint emission sensitivity shows impact from Canada, the majority of the air masses arriving at the Zeppelin Observatory originated from central and western Siberia at 4000 m a.g.l. North American fires dominated only at lower altitudes of up to 500 m a.g.l. This last event was the second most intense observed during this summer, showing a good correlation with the AOD values observed in Sect. . In addition, both C2 and C3 showed an important contribution from North America. This agrees well with the hypothesis of the long range transport considered after observing a wide and uniform aerosol size distribution in Sect.  in these events, specially for C3.

As mentioned in the introduction, during the MOSAiC expedition a persistent 10 km deep aerosol layer in the UTLS, roughly from 7–8 km up to 17–18 km over the central Arctic, with clear a sign of smoke was observed. A layer around 10–15 km has also been observed in the data for summer 2019 analyzed here. Therefore, some lifting of the smoke must have taken place. The air in July–August 2019 originated from ongoing large wildfires over Siberia and low-wind and stagnant conditions allowed air to accumulate. The lack of evidence of strong pyrocumulonimbus (pyroCb) activity over these fires during the key period in combination with CALIPSO smoke detections at 10 km, led to invoke that self-lofting might be a possible mechanism resulting in the persistent UTLS smoke layer. In a more recent publication, explicitly treat self-lofting as a credible alternative to pyroCb convection for raising large smoke masses from 2–6 km to the tropopause and cites the 2019 Siberian case and MOSAiC results as key evidence. In addition, tried to answer whether a dry firestorm plume (an intense conflagration that creates and sustains its own updraught wind system) can on its own reach the stratosphere. By using plume models (with and without entrainment), direct numerical simulations (DNS) and large-eddy simulations (LES) of idealized urban firestorms, they found that a dry plume starting at around 1 km (top of PBL) needs a temperature anomaly of about 60 K to stay positively buoyant up to a 15 km tropical tropopause. When they included entrainment, they found that for 1 km plume radius, mixing doubles temperature anomaly in the poles and sextuples it in the tropics. They conclude that narrow and dry plumes need to be unrealistically hot to reach stratospheric heights. Then, they used DNS and LES to simulate realistic dry firestorms and found that they never get hot enough to reach the stratosphere staying at around 5 km, at maximum. When relative humidity in the plume increased above 50 %, pyroCb-like convection developed, which lifted fire plumes to tropopause or even to stratosphere. They conclude that even moderately moist environments allowed latent heating to push firestorm plumes to the stratosphere. Overall, whether the lifting of smoke in summer 2019 was due to pyroCb-like latent heating (moist convection) or due to radiative heating (self-lofting) requires further research. While plume-rise parametrizations with moist thermodynamics and pyro-convection are already in use by many global models , they are not relevant here, as FLEXPART used emissions from CAMS GFAS.

During the periods between the aerosol events described above, relatively high AOD values have been also observed (AOD at 500 nm around 0.15–0.20 in July and August, dropping to 0.10–0.15 in September), which corresponded to fine aerosol with low absorption (see Sect. ). Modelling BC levels calculated using the described emissions inventory did not show a clear source of the aerosol. At around that period, an unexpected series of blasts from a remote volcano in the Kuril Islands (Raikoke volcano) sent ash and volcanic gases streaming high over the North Pacific Ocean . The dormant period ended around 18:00 UTC on 21 June 2019, when a vast plume of ash and volcanic gases shot up from its 700 m-wide crater. Several satellites observed as a thick plume rose and then streamed eastwards . observed that afterwards, the plume separated into an ash-dominated component in the south and a SO2-dominated component in the north. While the ash plume rapidly diluted and could not be further followed, the SO2 plume persisted. On 23–27 June 2019 parts of the dispersed SO2 cloud, which were observed with the Microwave Limb Sounder (MLS) satellite data, presented heights between 11 and 18 km, with a peak concentration at 14 km .

Therefore, we further simulated the Raikoke eruption using a new inventory for the volcanic SO2 emissions . The vertical cross-sections of the modelled volcanic SO2 arriving at Ny-Ålesund is illustrated in Fig. . This figure shows that the volcanic aerosol reaches Ny-Ålesund specially from 14 July, when it was also observed an increase in the AOD with respect to previous AOD values. This aerosol extended in a thick layer between 7 and 15 km a.g.l. for the first two weeks since its arrival, and afterwards it was mainly present between 10 and 15 km a.g.l., which is consistent with the observations made with CALIOP and KARL in Fig. .

Whether the aerosols observed at such altitudes were ash or soot was under strong debate. reported that the aerosol was smoke misclassified as volcanic sulfate focusing on lidar measurements in Leipzig (Germany). The reason for the misclassification was the aerosol identification algorithm that assumes non-spherical smoke particles in the stratosphere. They reported that self-lofting particles reached the tropopause within 2–7 d after emission and finally entered the lower stratosphere as aged spherical smoke particles. These spherical particles were misclassified as liquid sulfate particles. As explained in the previous subsection, the same group hypothesized that the layer detected during the MOSAiC expedition mainly originated from extraordinarily intense and long-lasting wildfires in central and eastern Siberia in July and August 2019 which may have reached the tropopause layer by self-lofting .

On the other hand, concluded that the stratospheric aerosol in the second half of 2019 over the Arctic consisted of Raikoke sulfate aerosol. This study was only based on observations of stratospheric infrared absorption spectra (in the framework of the satellite-based Atmospheric Chemistry Experiment mission). The authors found no indication of the presence of smoke. reported a high SO₂ concentration and aerosol clouds using data from the Ozone Mapping and Profiler Suite sensors on the Suomi National Polar-orbiting Partnership satellite. However, a recent comment on this publication made by states that the AOD derived from the measurements used in this analysis was four times higher than the corresponding AOD for the predicted SO₂ concentration from the Raikoke volcano. In this review, it is concluded that the Arctic UTLS aerosol was likely dominated by smoke in the lower part and by sulfate aerosol in the upper part. combined both observations and model simulations of the Raikoke eruption and found that a few days after the eruption, the volcanic plume was entrained in the Aleutian cyclone, and a month after, it circled the Earth (it reached Europe around 1 July). Accordingly, stratospheric AOD (sAOD) was as high as 0.045 (at 449 nm) at higher north-hemispheric latitudes, with an average value of 0.025 at longer wavelength (visible, 675 nm), while the background AOD was still enhanced in the Northern Hemisphere one year after the eruption. Finally, used a Raman lidar in the UK and found a thin layer at 14 km on 3 July, with the first detection of the main aerosol cloud on 13 July, which agrees very well with our own observations in Ny-Ålesund. At that period the aerosol was confined below 16 km extending to 20 km. The authors further reported a sustained period of clearly enhanced AOD from early August, with a maximum value (at 355 nm) around 0.05 in mid-August and remaining above 0.02 until early November.

Finally it has been investigated the origin of the surface events S1 and S2. The footprint emission sensitivity computed for the day of each event and the corresponding temporal evolution of the source and continental contributions to BC for the days of the events and days around have been plotted in Fig. . The footprint emission for events S1 and S2 (Fig. , left column), indicates that S1 had its source in Central Europe while, in S2, the BC is coming from Eastern Europe. This is corroborated by the continental contribution (Fig. , right column), which shows a main contribution from Europe together with a smaller contribution from Russia, more evident in S2. Both events had an anthropogenic source, mainly due to transportation (TRA) and commercial and domestic (DOM) emissions (see Fig. , middle column). The products for the age of the aerosol reaching Svalbard indicate that, in general, aerosol in these events traveled faster compared to CS1 (see Fig. S5 in the Supplement), which agrees with the slightly lower SSA₅₃₀ values observed.

By the late summer and early autumn, aerosol travel from lower latitudes to the Arctic is not as efficiently removed along the pathways as earlier, leading to a more efficient transport of polluted air masses and therefore an increase of the aerosol scattering and absorption coefficients measured . In addition, showed an evident change of atmospheric transport regime in the summer of 2019, from North America during the summer period to North Europe and Russia during autumn and early-winter, which supports the anthropogenic origin of the aerosol measured during these events.