Interannual and seasonal variations in aerosol optical depth of the atmosphere in two regions of Spitsbergen Archipelago (2002-2018)

In this work hourly averaged sun photometer data from the sites Barentsburg and Ny-Ålesund, both located in Spitsbergen in the European Arctic, are compared. Our data set comprises the years 2002 to 2018 with overlapping measurements at both sites during 2011 to 2018. We found for more turbid periods (aerosol optical depth τ0.5 > 0.1) that typically Barentsburg is more polluted than Ny-Ålesund, especially in the short wave spectrum. However, the diurnal variation of AOD is highly correlated. Next, τ was divided into a fine and coarse mode. It was found that generally the fine mode aerosol optical depth dominates and also shows a larger interannual as seasonal variation. The fine mode optical depth is in fact largest in spring during the Arctic Haze period. Overall the aerosol optical depth seems to decrease (at 500nm the fine mode optical depth decreased by 0.016 within 10 years), although this is hardly statistically significant.

stations occur in periods of Arctic haze and outflows of smokes from forest fires, which are differently manifested in these two regions. However, a comparison of monthly averages may introduce a bias due to different data availability at the involved sites.
In contrast to the above-mentioned works, we compared quasi-synchronous (nearly time coincident) AOD measurements in the neighbouring regions in 2011-2017 (see [Kabanov et al., 2018] for more details). The data of the SP1A (Ny-Ålesund) and SPМ (Barentsburg) photometer observations were used to calculate the hourly average AOD values. Then, the datasets from the two regions were compared provided that the times of the AOD measurements differed by no more than one hour.
Comparison of measurements with the two photometers showed a large dispersion of the data under the conditions of strong atmospheric turbidities, namely, during outflow of smoke plumes from forest fires and in the Arctic Haze situations. Due to large spatial inhomogeneity of these structures, AOD, measured in two regions, may strongly differ, making the comparison incorrect. Therefore, further analysis was performed for usual situations, when  а (0.5 µm) < 0.2. Figure 1 illustrates the regression relation between  а (0.5 µm) measurements in the neighbouring regions of Spitsbergen.
A comparison of the statistical characteristics showed that the average AOD values are a little larger in Barentsburg than in Ny-Ålesund [Kabanov et al., 2018]. The maximum difference in AOD is observed in the shortwave part of the spectrum (0.38 µm), ∆ =  а (SPM) - а (SP1A) = 0.024; while in near-IR range (0.87µm) the difference decreases to ∆ = 0.005. This feature is real despite the decreasing AOD at longer wavelengths and indicates that fine aerosol is more abundant in the atmosphere of Barentsburg. At the same time, we note that the AOD differences are minor (comparable with uncertainty of determining AOD -about 0.01-0.02 [Kabanov et al. 2009;Sakerin et al., 2013]), and the interdiurnal AOD variations in the two regions are coordinated in character (correlation coefficients are 0.83-0.89). Comparison of quasi-synchronous AOD measurements in Barentsburg and Hornsund gave close results [Kabanov et al., 2018]: ∆ = 0.004 -0.024, the correlation coefficients are 0.71-0.81.
Hence, observations in the neighbouring regions on Spitsbergen are quite compatible and identically reflect the specific features of the AOD variations.
The joint use of results from AOD monitoring in the neighbouring regions makes it possible to control the reliability of information, as well as to identify the specific features of AOD variations not only for a specific site, but for the region as a whole. The results of the observations in each of the regions have their own advantages.
The advantage of the data from Barentsburg (SPM / SP-9 photometers) is a wider range (0.34-2.14 µm) of the spectral measurements and the possibility to separate the contributions from two AOD components, using an empirical method (see subsection 2.3).
The valuable feature of the data from Ny-Ålesund is a longer AOD observation time series. However, different errors have been accumulated in these data for the long period of measurements. A simple exclusion of all suspect AOD measurements was undesirable because for analysis of multiyear variations it was necessary to keep the observation time series as long as possible. Taking this circumstance into account, the multiyear observation time series was prepared for sorting out or correction of suspect AOD values [Kabanov et al., 2019a]. The initial dataset was processed to remove the data in which short-term bursts or rapid AOD variations were observed, as well as the distortions to smoothness of the wavelength dependences  а (). Owing to a certain redundancy of the number of spectral channels, we could identify false measurements and select most reliable data.

Fine and coarse AOD components
The attenuation of radiation by atmospheric aerosol varies as a function of wavelength, depending on sizes and refractive index of aerosol particles. To characterize the AOD, measured at different wavelengths, the Ångström formula is widely used: where β and α are the approximation parameters of the spectral dependence of AOD; β is the turbidity coefficient, which is close in value to AOD at the wavelength of 1 µm; and α is the selectivity exponent (power-law decay).
Numerous studies in different regions and atmospheric conditions showed that the formula (1) does describe well the wavelength dependence  а () in the main range (0.34 -1 m) of AOD measurements. [Angstrom, 1964;Shifrin, 1995;Eck et al., 1999;Cachorro et al., 2000;Schuster et al., 2006]. At the same time, the use of this formula has limitations and disadvantages, requiring an explanation.
First, the Ångström formula becomes unsuitable for describing the wavelength behavior of AOD in the atmospheric "transparency windows" in the wavelength range of 1 -4 m. This is because the power-law dependence (1) stems from the combined action of fine and coarse aerosol fractions, which have different spectral properties. Extinction of radiation by small particles (2r/ < 1) is dominant in the visible region of spectrum; however, it rapidly decays with the growing wavelength and becomes insignificant in the region beyond of 1 m.
Extinction of radiation by coarse aerosol barely changes with the wavelength and becomes predominant in the near-IR range. Mie calculations and experimental data [Sakerin and Kabanov, 2007a;Sakerin et al., 2008b] confirm that the power-law AOD decay gradually goes over into almost neutral dependence. Therefore,  а () in a wider wavelength range is better to represent by a sum of two components: where  с is the constant (wavelength independent) coarse AOD component;  f () is the selective fine component; m and n are the approximation parameters of the spectral dependence of τ f () (they are similar to the parameters  and  of the Ångström formula)  and  are themselves correlated.
Thus, the use of the Ångström parameters in the analysis of AOD variations is ambiguous and may lead to erroneous conclusions. It is more preferable to consider the specific features of variations in two AOD components:  f () and  с . In addition to different sizes and spectral properties, fine and coarse aerosol fractions differ in the origins of particles and their transformation in the atmosphere. Fine (sulfate, organic, etc.) aerosol is formed in the atmosphere as a result of various photochemical and microphysical processes [Kondratyev et al., 2006]. The lifetime of fine aerosol in the troposphere is a few days; therefore, it can be transported long distances (hundreds and thousands of kilometers) away. The main source of coarse (marine and dust) aerosol is the underlying surface.
Because of its short lifetime and small transport distance, coarse aerosol is more local in character and pertains to a specific terrain. The only exceptions are powerful dust outflows in tropical latitudes.

Methods for determining fine and coarse AOD components
As was already indicated above, in the near-IR range, the effect of fine aerosol becomes insignificant, and AOD is determined only by the coarse component. Therefore,  с can be determined by an empirical method (ЕМ), i.e., from minimal or average AOD values, measured in the range of 1.24-2.14 m [Sakerin and Kabanov, 2007b;Sakerin et al., 2008b]. Then, the second (fine) component is found as a residual of the total AOD. Usually, it is estimated for the wavelength of 0.5 m: However, most sun photometers (and in particular the SP1A in Ny-Ålesund) operate in the wavelength range up to 1.05 m, making empirical method inapplicable. In this case,  с and can be estimated using calculation methods. For instance, in the AERONET system (http://aeronet.gsfc.nasa.gov), is calculated using the spectral deconvolution algorithm [O'Neill et al., 2003], based on the relationship of spectral AOD, measured in the shortwave part of the spectrum 0.38-1.02 m.
In the work  we suggested simpler methods for separating the contributions from and  с , based on the regression interrelations with the parameters of Ångström formula. In the first regression method (RM1),  с is estimated using its interrelation with the parameter β (see formula (3) below). In the second method (RM2), the regression dependence of on the parameters α and β (see formula (4)) is used. Comparison of different methods of (or  с ) estimation for the conditions of the marine and continental (Tomsk) atmosphere showed close results: the average difference of  с from data of base empirical method (EM) does not exceed 0.007 for the standard deviation from 0.006 to 0.024.
For the conditions of Arctic region (Spitsbergen), we performed an additional study [Kabanov et al., 2019a], concerning the selection of an optimal method of (or  с ) estimation. Different methods were tested using SPM photometer measurements of AOD in Barentsburg. The error of the methods was estimated by comparing the calculated values of or  с with the data from base empirical method. Figure 3 illustrates the results of testing two regression methods (RM1 and RM2), based on the interrelations (а) between  с and the parameter , and (b) between and the parameters , . For the conditions of Spitsbergen, we obtained the following regression equations: RM2: = (-0.829 + 1.0660.5 - ) (4) Table 2 presents the standard deviations  and the correlation coefficients R between the calculated (RM1, RM2) and empirical (ЕМ)  с values. These results suggest the regression methods make it possible to estimate  c with an identical error of 0.007.
The disadvantage of the regression methods is that they require a preliminary data accumulation under the conditions of a specific region for determining optimal regression coefficients in equations (3) and (4)  . Of course, the error of the regression methods may increase if aerosol characteristics strongly differ from those typical for the region and do not correspond to the selected regression coefficients.
Therefore, in addition to the regression methods, we considered the applicability of another two methods of estimation, based on the results of solving the inverse problem, namely: retrieval of particle sizes from measurements of spectral AOD. The inversion method 1 (IM1) is based on the interrelation between and volume or cross section of particles of fine aerosol. This method is implemented in the following steps: 1st step: Based on any known method of solving the inverse problem (for a specified refractive index, type of the particle distribution function, and grid of radius ranges), the spectral AOD values are used to calculate the particle distribution function (dV/dr) or (dS/dr).
2nd step: In the distribution (dV/dr) thus obtained we select its part referring to the fine fraction, and, for it, calculate the total particle volume (V f ) through integration.
3rd step: We consider the regression interrelation between particle volumes in the fine fraction (V f ) and the  f values, calculated by empirical method (EM). The interrelation thus obtained ( Fig. 4а) is used to select the parameters of a linear regression equation which makes it possible to calculate the component  f according to the particle volumes V f .
The inversion method 2 (IM2) is implemented by solving first the inverse problem, and then the direct problem of the aerosol optics: (a) as in IM1, the spectral AOD values are used to retrieve the distribution functions (dS/dr); and (b) based on the (dS/dr) values, is calculated for the size range of fine aerosol.
The inverse problem on retrieving the distribution functions (dS/dr) was solved using iteration algorithm of M.A. Sviridenkov [Sviridenkov, 2001], modified from Twitty algorithm [Twitty, 1975]. The particle distribution was assumed to be lognormal, and the refractive index was assumed to have the real part of 1.5 and the imaginary part of 0. In the calculations we used the following radius grid: 0.09-0. (dV/dr); and (c) for different radius boundaries of particles of fine fraction (0.1-0.5 m and 0.1-0.45 m). Figure 4 presents examples of interrelations: (a) between (EM) and calculated values of particle volume V f ; and (b)  between values, determined using base (EM) and inversion (IM1) methods. The calculations in this case were performed for the wavelength range of AOD 0.38-1.02 m and particle radius range of 0.1-0.5 m.
Analysis of application of IM1 and IM2 [Kabanov et al., 2019a] showed that the determination error decreases by about a factor of 1.5 when the AOD is used in a wide (0.34-2.14 m) wavelength range. However, for the narrower wavelength range of the SP1A photometer (0.38-1.02 m) the calculation error is comparable with results from regression methods (see Table 2). That is, the relative errors of the  с and calculations for the mean conditions of Barentsburg ( a 5 . 0  = 0.086 [Sakerin et al., 2018a]) are 30% and 11% respectively.
The IM1 method was chosen for a subsequent use. Despite a more complicated procedure of its calculations, IM1 is more sensitive to aerosol variations, which is indicated by the highest correlation coefficient between f and data from base (ЕМ).
There may be a question as to why after retrieval of aerosol size distribution we nonetheless consider the seasonal and interannual variations in optical characteristics:  с and f 5 . 0  ? Analysis of disperse composition of aerosol is a more complex and non-unique problem because it is necessary to consider the transformation of two aerosol fractions, which are described by a few parameters: shapes and widths of distributions for each fraction, separation boundary, and effective particle radii. Moreover, an uncertainty remains about the values of these parameters because of the priori specified aerosol refractive index.
In this work, we pursued a simpler task: to determine the character and magnitude of variations in aerosol optical characteristics. In this case, instead of many microstructure parameters, it is sufficient to analyze their more compact optical image in the form of two components,  f and  с .

Discussion of the results
Current climate change and environmental transformation influence the regularities of variations in aerosol characteristics to some degree. Because of the deficit of its own aerosol sources in the Arctic zone, an important role in AOD variations is played by outflows of smoke, anthropogenic and volcanic aerosol from midlatitudes.
The frequency of these outflows in particular months and years determines the specific features of seasonal dynamics of AOD in Arctic regions and magnitude of interannual oscillations.

Interannual variations
The highest atmospheric turbidities in the region of Spitsbergen were observed on July 10,  Pakszys and Zielinski, 2017;Markowicz et al., 2016]. We also considered in detail the second anomalous situation (in May 2006) [e.g. Myhre et al., 2007;Stohl et al., 2007], associated with the outflow of smoke aerosol from agricultural fires in the Eastern Europe.
Episodes with high atmospheric turbidities were also observed in June 2003, March and August 2008, in April and August 2009, and in April 2011 The AOD values in these periods of time had already been analyzed by many authors [Toledano et al., 2012;Glantz et al., 2014;Tomasi et al., 2015;Chen et al., 2016;Pakszys and Zielinski, 2017]. Independent of the causes for these short-term turbidities (Arctic haze, outflows of smoke or volcanic aerosol), they enhance not only the monthly, but also annual AOD values.
The above-mentioned high-turbidity episodes (2006,2008,2009,2015) were reflected partly in annual AOD oscillations (Fig. 6). Moreover, a maximum appeared in the interannual variations in 2011-2012. This maximum was due not to extreme 1-3-day AOD bursts, but to stronger turbidities as compared to the neighbouring years.
The annual AOD maxima occur with the average periodicity of about three years. When high-turbidity episodes are eliminated (see dashed line in Fig. 6), certain maxima disappear; however, the general character of the AOD oscillations remains unchanged. Among these maxima, the highest AOD value in 2003 seems suspect. This annual AOD value cannot be considered as representative because of short period of observations (4 days in March and 5 days in May-June) in that year.
In addition to oscillations, a tendency for a minor AOD decrease can be discerned in the multiyear variations ( Fig.   6). On average, the decline of  a 0.5 is 0.016 in ten years, but the significance of this trend is 0.062, which is close but exceeds the statistically significant level of 0.05. This is also indicated by a comparison of AOD characteristics The interannual oscillations in the Ångström exponent can be considered as minor: the variation coefficients for  are 13-15%. The average  and the total variability range (from 1 to 1.7) are in the regions of values characteristic for the continental midlatitude atmosphere and are larger than in the marine atmosphere [Sakerin et al., 2008b;2018b]. These values of the Ångström exponent are because the ratio ( / с = 2.6-2.9) and the relative contribution of fine component ( . 0  = 0.73-0.75) are close to continental values. As an example, we present multiyear data in boreal zone of Siberia in spring (smoke-free) period [Kabanov et al., 2019b]. The average AOD From Figs. 6 and 7 it is clearly seen that AOD in Barentsburg is almost always higher than in Ny-Ålesund. The average excess of annual AOD is (see rows 2 and 3 in Table 3

Specific features of seasonal variations
The most common regularity of the seasonal AOD behavior at midlatitudes is the spring (and sometimes also summer) maximum and fall minimum [e.g., Sakerin et al., 2015;Chubarova et al., 2014;Holben et al., 2001]. The  primary causes for this AOD behavior are the annual cycle in the Sun's declination meaning a return of sunlight and possibly a longer aerosol life-time over the frozen ocean. The springtime increases in insolation and temperature trigger a few processes: (а) snow cover evaporates and melts; (b) the atmosphere is enriched by different deposition products, accumulated over the winter, (c) primary (marine, soil) aerosol starts to come from the underlying surface; and (d) photochemical processes of production of in situ aerosol in the atmosphere and emission of organic aerosol are activated [e.g., Kondratyev et al., 2006].
The seasonal AOD dynamics in the Arctic zone is analogous to midlatitudes: springtime maximum and a decay toward fall [e.g., Toledano et al., 2012;Tomasi et al., 2015;Sakerin et al., 2018]. This AOD behavior is because of similar annual rhythms of both own aerosol sources in the Arctic and long-range aerosol transports from midlatitudes. Seasonal AOD variations in Barentsburg are characterized by an additional summer maximum in July-August.
Despite this difference, common factors in AOD variations in the two regions are nonetheless predominant.
Analysis of interrelation between AOD values, measured in Ny-Ålesund and Barentsburg ( Fig. 9), showed quite a high (0.90) correlation coefficient. Hence, the synoptic, seasonal, and interannual AOD oscillations are largely coordinated in character.
Observation time series in the two regions were compared to clarify the causes for the summertime AOD maximum in Barentsburg. The increased AOD values in July and August were found to be due to the situations with smoke outflows (and, in particular, on July 10, 2015 [Sakerin et al., 2018a]). Of the total number of measurements, percentage of smoke-contaminated measurements turned out to be larger in Barentsburg than in Ny-Ålesund. A few rare, but powerful outflows of smoke aerosol have led to an increase in the Monthly

Conclusions
We present brief results of our study.
1. It is noted that, to identify the specific features of seasonal and multiyear variations in atmospheric AOD, it is important to analyze separately fine and coarse AOD components, having different spectral properties, origins, and lifetimes. As applied to AOD measurements in Ny-Ålesund, we considered a few methods for estimating the contributions of fine and coarse components, and one of the methods (IM1) is selected for a subsequent use. A comparison with data from the base (ЕМ) showed that the standard deviation of the  с and calculations is 0.007, and the relative errors are 30% and 11% respectively.
2. Outflows of fine aerosol of different types from the Eurasian and North American midlatitudes affect appreciably the monthly (and even annual) AOD in the Arctic atmosphere. Outflows of smokes from massive  Abstract. In this work hourly averaged sun photometer data from the sites Barentsburg and Ny-Ålesund, both located in Spitsbergen in the European Arctic, are compared. Our data set comprises the years 2002 to 2018 with overlapping measurements at both sites during 2011 to 2018. We found for more turbid periods (aerosol optical depth τ0.5 > 0.1) that typically Barentsburg is more polluted than Ny-Ålesund, especially in the short wave spectrum. However, the diurnal variation of AOD is highly correlated. Next, τ was divided into a fine and coarse mode. It was found that generally the fine mode aerosol optical depth dominates and also shows a larger interannual as seasonal variation. The fine mode optical depth is in fact largest in spring during the Arctic Haze period. Overall the aerosol optical depth seems to decrease (at 500nm the fine mode optical depth decreased by 0.016 within 10 years), although this is hardly statistically significant.

Introduction
The studies of the character and causes of variations in all components of climate system, including the aerosol composition of the atmosphere, became more urgent with regard to climate change [IPCC, 2013]. Atmospheric aerosol plays an important role in the processes of solar radiative transfer and exchange by different substances (and, in particular, pollutants) between the continents and ocean [e.g. Kondratyev et al., 2006]. As compared to gases, aerosol is characterized by a complex physicochemical composition and stronger variation of concentration and radiative impact.
Of the various aerosol characteristics, the observations of aerosol optical depth (AOD) of the atmosphere are most widespread and carried out at the international and national networks of stations using sun photometers (see, e.g., [WMO, 2005;Holben et al., 1998]). The AOD represents the extinction of radiation, integrated over atmospheric column, and can be considered as an optical equivalent of the total aerosol content.
One of the main aerosol climatology problems is to determine the specific features of interannual and seasonal variations in different regions. However, under the conditions of changing climate system, a time series 10 years (or even 20 years) long may turn out to be insufficient to identify correctly the tendencies or periodicities in variations of aerosol characteristics.
First observations of spectral AOD in the Arctic zone were carried out about 40 years ago [Shaw, 1982;Freund, 1983;Radionov and Marshunova, 1992]; however, they become regular in character only in the early 2000s, after the development of photometric observations at continental stations. A comprehensive overview of atmospheric AOD in the Polar Regions has been presented by C. Tomasi [Tomasi et al., 2012;.
Studies by various authors showed that the Arctic atmosphere is affected appreciably by outflows of aerosols of different types (smoke, industrial, sulfate, organic) from Eurasia and North America. The most powerful effect is due to smoke from forest fires that cover large areas of boreal zone [e.g. Chubarova et al., 2012;Sitnov et al., 2013;Zhuravleva et al., 2017]. Long-range transports of smoke plumes lead to a considerable pollution of the Arctic atmosphere [Stohl et al., 2006;Stone et al., 2008;Eck et al., 2009;Vinogradova et al., 2015]. These episodes are short-term (1-3 days) and rare because they depend on the product of probabilities of two independent events: (а) the fire itself in any area of boreal zone, and (b) the fact that the trajectory of air transport from a fire center arrived exactly at a given region of the Arctic.
In addition to smokes, anthropogenic and other types of fine aerosol also outflow from midlatitudes. In contrast to forest fires, the sources of these aerosols operate almost all the time and are distributed over the entire territory of human life activity. A somewhat larger concentration of anthropogenic aerosol in densely populated regions of Europe was observed in the past century; however, recently the industrial emissions have been stabilized or reduced in this area [Tørseth et al., 2012;Li et al., 2014;Zhdanova et al., 2019].
An AOD increase may also be associated with volcanic activity. In the period of time considered here, there were no large eruptions (like Pinatubo volcano in 1991), having a global effect. The effects of less powerful volcanic eruptions on the Arctic atmosphere are short-term to mid-term (some weeks in duration) and comparable to those due to smokes from forest fires. For instance, AOD increase on Spitsbergen was observed after the eruptions of volcanoes Kasatochi (August 2008) and Sarychev (July 2009) [Hoffmann et al., 2010;Toledano et al., 2012].
The effects of pollutant outflows on the Arctic atmosphere intensify in late winter -early spring. The temperature inversions in this season lead to the formation and accumulation of aerosol in separate layers of the troposphere (the Arctic Haze Phenomenon) [e.g. Shaw, 1995;Quinn et al., 2007;Tomasi et al., 2015].
In the recent decade increasingly more work was published analyzing the multiyear AOD variations in different regions, either based on spectral [Zhdanova et al., 2019;Chubarova et al., 2016;Putaud et al., 2014;Li et al., 2014;Xia, 2011;Michalsky et al., 2010;Sakerin et al., 2008a;Weller et al., 1998]  For Spitsbergen an analysis of aerosol properties for separate periods was already performed before [Herber et al., 2002;Glantz et al., 2014;Chen et al., 2016;Pakszys and Zielinski, 2017]. In this work, we discuss the AOD  [Herber et al., 2002]. Here, we analyze the AOD variations in a later period when measurements became regular and homogeneous in character. The main characteristics of sun photometers (SP1A and SP2H), used in measurements, are presented in Table 1.
The sun photometer in Ny-Ålesund is located just south of the settlement in about 10m asl on the BSRN radiation field. The temporal resolution of the data is 1 minute. The instruments are regularly calibrated in Izaña /Tenerife. The air masses for aerosol and for ozone were considered according to the formulas by Kasten and Young (1989) and the WMO No 183 report (2008) [Anderson et al., 1986]. Water vapor absorption is accounted for using real water vapor contents, measured in the wavelength channel of 0.94 m.
Total amount of data (hours and days of measurements), which were used in the AOD analysis in two regions, is presented in Table 1

Data comparison and preparation of observation time series
A comparison of observations using two photometers may be of interest for intercalibration of the instruments, i.e., estimating instrumental-methodic AOD determination errors (for the Polar Regions e.g. Mazzola et al. 2011).
If the measurements are separated in space (as in the given case), the difference in the data makes it possible to estimate the local AOD inhomogeneities (Sakerin et al. 2010), caused by anthropogenic or natural factors: local weather conditions, type and state of the underlying surface, orography, and the effect of industrial or other sources of aerosol.
It should be noted that we already compared earlier the AOD measurements at the neighbouring stations on Spitsbergen Archipelago, i.e., Hornsund and Ny-Ålesund [Toledano et al., 2012;Pakszys and Zielinski, 2017].
Comparison of time independent measurements showed that the average difference in the annual and seasonal AOD values at the wavelength of 0.55 µm does not exceed 0.01-0.02. Large differences in AOD between these stations occur in periods of Arctic haze and outflows of smokes from forest fires, which are differently manifested in these two regions. However, a comparison of monthly averages may introduce a bias due to different data availability at the involved sites.
In contrast to the above-mentioned works, we compared quasi-synchronous (nearly time coincident) AOD measurements in the neighbouring regions in 2011-2017 (see [Kabanov et al., 2018] for more details). The data of the SP1A (Ny-Ålesund) and SPМ (Barentsburg) photometer observations were used to calculate the hourly average AOD values. Then, the datasets from the two regions were compared provided that the times of the AOD measurements differed by no more than one hour.
Comparison of measurements with the two photometers showed a large dispersion of the data under the conditions of strong atmospheric turbidities, namely, during outflow of smoke plumes from forest fires and in the Arctic Haze situations. Due to large spatial inhomogeneity of these structures, AOD, measured in two regions, may strongly differ, making the comparison incorrect. Therefore, further analysis was performed for usual situations, when  а (0.5 µm) < 0.2. Figure  The joint use of results from AOD monitoring in the neighbouring regions makes it possible to control the reliability of information, as well as to identify the specific features of AOD variations not only for a specific site, but for the region as a whole. The results of the observations in each of the regions have their own advantages.
The advantage of the data from Barentsburg (SPM / SP-9 photometers) is a wider range (0.34-2.14 µm) of the spectral measurements and the possibility to separate the contributions from two AOD components, using an empirical method (see subsection 2.3).
The valuable feature of the data from Ny-Ålesund is a longer AOD observation time series. However, different errors have been accumulated in these data for the long period of measurements. A simple exclusion of all suspect AOD measurements was undesirable because for analysis of multiyear variations it was necessary to keep the observation time series as long as possible. Taking this circumstance into account, the multiyear observation time series was prepared for sorting out or correction of suspect AOD values [Kabanov et al., 2019a]. The initial dataset was processed to remove the data in which short-term bursts or rapid AOD variations were observed, as well as the distortions to smoothness of the wavelength dependences  а (). Owing to a certain redundancy of the number of spectral channels, we could identify false measurements and select most reliable data.

Fine and coarse AOD components
The attenuation of radiation by atmospheric aerosol varies as a function of wavelength, depending on sizes and refractive index of aerosol particles. To characterize the AOD, measured at different wavelengths, the Ångström formula is widely used: where β and α are the approximation parameters of the spectral dependence of AOD; β is the turbidity coefficient, which is close in value to AOD at the wavelength of 1 µm; and α is the selectivity exponent (power-law decay).
Numerous studies in different regions and atmospheric conditions showed that the formula (1) does describe well the wavelength dependence  а () in the main range (0.34 -1 m) of AOD measurements. [Angstrom, 1964;Shifrin, 1995;Eck et al., 1999;Cachorro et al., 2000;Schuster et al., 2006]. At the same time, the use of this formula has limitations and disadvantages, requiring an explanation.
First, the Ångström formula becomes unsuitable for describing the wavelength behavior of AOD in the atmospheric "transparency windows" in the wavelength range of 1 -4 m. This is because the power-law dependence (1) stems from the combined action of fine and coarse aerosol fractions, which have different spectral properties. Extinction of radiation by small particles (2r/ < 1) is dominant in the visible region of spectrum; however, it rapidly decays with the growing wavelength and becomes insignificant in the region beyond of 1 m.
Extinction of radiation by coarse aerosol barely changes with the wavelength and becomes predominant in the near-IR range. Mie calculations and experimental data [Sakerin and Kabanov, 2007a;Sakerin et al., 2008b] confirm that the power-law AOD decay gradually goes over into almost neutral dependence. Therefore,  а () in a wider wavelength range is better to represent by a sum of two components: We also note that the component  с , which influences the exponent , is tightly related and has close values to the second Ångström parameter [Sakerin and Kabanov, 2007a, b]:  с  . A consequence of this is that the parameters  and  are themselves correlated.
Thus, the use of the Ångström parameters in the analysis of AOD variations is ambiguous and may lead to erroneous conclusions. It is more preferable to consider the specific features of variations in two AOD components:  f () and  с . In addition to different sizes and spectral properties, fine and coarse aerosol fractions differ in the origins of particles and their transformation in the atmosphere. Fine (sulfate, organic, etc.) aerosol is formed in the atmosphere as a result of various photochemical and microphysical processes [Kondratyev et al., 2006]. The lifetime of fine aerosol in the troposphere is a few days; therefore, it can be transported long distances (hundreds and thousands of kilometers) away. The main source of coarse (marine and dust) aerosol is the underlying surface.
Because of its short lifetime and small transport distance, coarse aerosol is more local in character and pertains to a specific terrain. The only exceptions are powerful dust outflows in tropical latitudes.

Methods for determining fine and coarse AOD components
As was already indicated above, in the near-IR range, the effect of fine aerosol becomes insignificant, and AOD is determined only by the coarse component. Therefore,  с can be determined by an empirical method (ЕМ), i.e., from minimal or average AOD values, measured in the range of 1.24-2.14 m [Sakerin and Kabanov, 2007b;Sakerin et al., 2008b]. Then, the second (fine) component is found as a residual of the total AOD. Usually, it is estimated for the wavelength of 0.5 m: However, most sun photometers (and in particular the SP1A in Ny-Ålesund) operate in the wavelength range up to 1.05 m, making empirical method inapplicable. In this case,  с and can be estimated using calculation methods. For instance, in the AERONET system (http://aeronet.gsfc.nasa.gov), is calculated using the spectral deconvolution algorithm [O'Neill et al., 2003], based on the relationship of spectral AOD, measured in the shortwave part of the spectrum 0.38-1.02 m.
In the work  we suggested simpler methods for separating the contributions from and  с , based on the regression interrelations with the parameters of Ångström formula. In the first regression method (RM1),  с is estimated using its interrelation with the parameter β (see formula (3) below). In the second method (RM2), the regression dependence of on the parameters α and β (see formula (4)) is used. Comparison of different methods of (or  с ) estimation for the conditions of the marine and continental (Tomsk) atmosphere showed close results: the average difference of  с from data of base empirical method (EM) does not exceed 0.007 for the standard deviation from 0.006 to 0.024.
For the conditions of Arctic region (Spitsbergen), we performed an additional study [Kabanov et al., 2019a], concerning the selection of an optimal method of (or  с ) estimation. Different methods were tested using SPM photometer measurements of AOD in Barentsburg. The error of the methods was estimated by comparing the calculated values of or  с with the data from base empirical method. Figure 3 illustrates the results of testing two regression methods (RM1 and RM2), based on the interrelations (а) between  с and the parameter , and (b) between and the parameters , . For the conditions of Spitsbergen, we obtained the following regression equations: RM2: = (-0.829 + 1.0660.5 - ) (4) Table 2 presents the standard deviations  and the correlation coefficients R between the calculated (RM1, RM2) and empirical (ЕМ)  с values. These results suggest the regression methods make it possible to estimate  c with an identical error of 0.007.
The disadvantage of the regression methods is that they require a preliminary data accumulation under the conditions of a specific region for determining optimal regression coefficients in equations (3) and (4)  . Of course, the error of the regression methods may increase if aerosol characteristics strongly differ from those typical for the region and do not correspond to the selected regression coefficients.
Therefore, in addition to the regression methods, we considered the applicability of another two methods of estimation, based on the results of solving the inverse problem, namely: retrieval of particle sizes from measurements of spectral AOD. The inversion method 1 (IM1) is based on the interrelation between and volume or cross section of particles of fine aerosol. This method is implemented in the following steps: 1st step: Based on any known method of solving the inverse problem (for a specified refractive index, type of the particle distribution function, and grid of radius ranges), the spectral AOD values are used to calculate the particle distribution function (dV/dr) or (dS/dr).
2nd step: In the distribution (dV/dr) thus obtained we select its part referring to the fine fraction, and, for it, calculate the total particle volume (V f ) through integration. The inverse problem on retrieving the distribution functions (dS/dr) was solved using iteration algorithm of M.A. Sviridenkov [Sviridenkov, 2001], modified from Twitty algorithm [Twitty, 1975]. The particle distribution was assumed to be lognormal, and the refractive index was assumed to have the real part of 1.  Figure 4 presents examples of interrelations: (a) between (EM) and calculated values of particle volume V f ; and (b)  between values, determined using base (EM) and inversion (IM1) methods. The calculations in this case were performed for the wavelength range of AOD 0.38-1.02 m and particle radius range of 0.1-0.5 m.
Analysis of application of IM1 and IM2 [Kabanov et al., 2019a] showed that the determination error decreases by about a factor of 1.5 when the AOD is used in a wide (0.34-2.14 m) wavelength range. However, for the narrower wavelength range of the SP1A photometer (0.38-1.02 m) the calculation error is comparable with results from regression methods (see Table 2). That is, the relative errors of the  с and calculations for the mean conditions of Barentsburg ( a 5 . 0  = 0.086 [Sakerin et al., 2018a]) are 30% and 11% respectively.
The IM1 method was chosen for a subsequent use. Despite a more complicated procedure of its calculations, IM1 is more sensitive to aerosol variations, which is indicated by the highest correlation coefficient between f and data from base (ЕМ).
There may be a question as to why after retrieval of aerosol size distribution we nonetheless consider the seasonal and interannual variations in optical characteristics:  с and f 5 . 0  ? Analysis of disperse composition of aerosol is a more complex and non-unique problem because it is necessary to consider the transformation of two aerosol fractions, which are described by a few parameters: shapes and widths of distributions for each fraction, separation boundary, and effective particle radii. Moreover, an uncertainty remains about the values of these parameters because of the priori specified aerosol refractive index.
In this work, we pursued a simpler task: to determine the character and magnitude of variations in aerosol optical characteristics. In this case, instead of many microstructure parameters, it is sufficient to analyze their more compact optical image in the form of two components,  f and  с .

Discussion of the results
Current climate change and environmental transformation influence the regularities of variations in aerosol characteristics to some degree. Because of the deficit of its own aerosol sources in the Arctic zone, an important role in AOD variations is played by outflows of smoke, anthropogenic and volcanic aerosol from midlatitudes.
The frequency of these outflows in particular months and years determines the specific features of seasonal dynamics of AOD in Arctic regions and magnitude of interannual oscillations.

Interannual variations
The highest atmospheric turbidities in the region of Spitsbergen were observed on July 10,  Pakszys and Zielinski, 2017;Markowicz et al., 2016]. We also considered in detail the second anomalous situation (in May 2006) [e.g. Myhre et al., 2007Stohl et al., 2007], associated with the outflow of smoke aerosol from agricultural fires in the Eastern Europe.


Episodes with high atmospheric turbidities were also observed in June 2003, March and August 2008, in April and August 2009, and in April 2011 The AOD values in these periods of time had already been analyzed by many authors [Toledano et al., 2012;Glantz et al., 2014;Tomasi et al., 2015;Chen et al., 2016;Pakszys and Zielinski, 2017]. Independent of the causes for these short-term turbidities (Arctic haze, outflows of smoke or volcanic aerosol), they enhance not only the monthly, but also annual AOD values.
The above-mentioned high-turbidity episodes (2006,2008,2009,2015) were reflected partly in annual AOD oscillations (Fig. 6). Moreover, a maximum appeared in the interannual variations in 2011-2012. This maximum was due not to extreme 1-3-day AOD bursts, but to stronger turbidities as compared to the neighbouring years.
The annual AOD maxima occur with the average periodicity of about three years. When high-turbidity episodes are eliminated (see dashed line in Fig. 6), certain maxima disappear; however, the general character of the AOD oscillations remains unchanged. Among these maxima, the highest AOD value in 2003 seems suspect. This annual AOD value cannot be considered as representative because of short period of observations (4 days in March and 5 days in May-June) in that year.
In addition to oscillations, a tendency for a minor AOD decrease can be discerned in the multiyear variations (Fig.   6). On average, the decline of  a 0.5 is 0.016 in ten years, but the significance of this trend is 0.062, which is close but exceeds the statistically significant level of 0.05. This is also indicated by a comparison of AOD characteristics From the statistical characteristics (Table 3)  The interannual oscillations in the Ångström exponent can be considered as minor: the variation coefficients for  are 13-15%. The average  and the total variability range (from 1 to 1.7) are in the regions of values characteristic for the continental midlatitude atmosphere and are larger than in the marine atmosphere [Sakerin et al., 2008b;2018b]. These values of the Ångström exponent are because the ratio ( / с = 2.6-2.9) and the relative contribution of fine component ( . 0  = 0.73-0.75) are close to continental values. As an example, we present multiyear data in boreal zone of Siberia in spring (smoke-free) period [Kabanov et al., 2019b]. The average AOD From Figs. 6 and 7 it is clearly seen that AOD in Barentsburg is almost always higher than in Ny-Ålesund. The average excess of annual AOD is (see rows 2 and 3 in Table 3 Fig. 7). Different behaviors of may be because observation time series are inhomogeneous in each region due to clouds or because AOD are measured at different times.

Specific features of seasonal variations
The most common regularity of the seasonal AOD behavior at midlatitudes is the spring (and sometimes also summer) maximum and fall minimum [e.g., Sakerin et al., 2015;Chubarova et al., 2014;Holben et al., 2001]. The  primary causes for this AOD behavior are the annual cycle in the Sun's declination meaning a return of sunlight and possibly a longer aerosol life-time over the frozen ocean. The springtime increases in insolation and temperature trigger a few processes: (а) snow cover evaporates and melts; (b) the atmosphere is enriched by different deposition products, accumulated over the winter, (c) primary (marine, soil) aerosol starts to come from the underlying surface; and (d) photochemical processes of production of in situ aerosol in the atmosphere and emission of organic aerosol are activated [e.g., Kondratyev et al., 2006].
The seasonal AOD dynamics in the Arctic zone is analogous to midlatitudes: springtime maximum and a decay toward fall [e.g., Toledano et al., 2012;Tomasi et al., 2015;Sakerin et al., 2018]. This AOD behavior is because of similar annual rhythms of both own aerosol sources in the Arctic and long-range aerosol transports from midlatitudes. Analysis of interrelation between AOD values, measured in Ny-Ålesund and Barentsburg (Fig. 9), showed quite a high (0.90) correlation coefficient. Hence, the synoptic, seasonal, and interannual AOD oscillations are largely coordinated in character.
Observation time series in the two regions were compared to clarify the causes for the summertime AOD maximum in Barentsburg. The increased AOD values in July and August were found to be due to the situations with smoke outflows (and, in particular, on July 10, 2015 [Sakerin et al., 2018a]). Of the total number of measurements, percentage of smoke-contaminated measurements turned out to be larger in Barentsburg than in Ny-Ålesund. A few rare, but powerful outflows of smoke aerosol have led to an increase in the Monthly  Table 4 and in Fig. 11.   values vary in a similar way (Fig. 12а). The

Conclusions
We present brief results of our study.
1. It is noted that, to identify the specific features of seasonal and multiyear variations in atmospheric AOD, it is important to analyze separately fine and coarse AOD components, having different spectral properties, origins, and lifetimes. As applied to AOD measurements in Ny-Ålesund, we considered a few methods for estimating the contributions of fine and coarse components, and one of the methods (IM1) is selected for a subsequent use. A comparison with data from the base (ЕМ) showed that the standard deviation of the  с and calculations is 0.007, and the relative errors are 30% and 11% respectively.
2. Outflows of fine aerosol of different types from the Eurasian and North American midlatitudes affect appreciably the monthly (and even annual) AOD in the Arctic atmosphere. Outflows of smokes from massive