Aerosol climatology in an Alpine valley



[1] Aerosol optical depth (AOD) is measured with a Precision Filter Radiometer in Innsbruck, Austria, from 2007 to 2011. The 5-year time series is analyzed with respect to the Ångström parameterα and quadratic coefficient γ. Information on the aerosol size distribution is obtained by adopting a graphical framework showing the particle fine mode fraction as well as the effective fine mode radius. The aerosol conditions in Innsbruck reveal a typical midlatitudinal pattern for continental urban sites with low pollution. Small particles originating from traffic pollution as well as natural sources dominate leading to an overall mean α of 1.53 +/− 0.28 and γ of −0.54 +/− 0.25. Low AOD at 500 nm is observed in winter with multiannual monthly means as low as 0.08 and higher AOD in spring and summer with multiannual monthly means up to 0.18. The maximum daily mean AOD at 500 nm always remains below 0.5. Special events such as Saharan dust events or presence of volcanic ash are detected and discussed.

1. Introduction

[2] Aerosols play an important role in atmospheric mechanisms relevant to climate change. They have a large impact on the radiative balance of the Earth through the scattering and absorption of incoming sunlight and cloud formation [Ramanathan et al., 2001; Pöschl, 2005]. According to Forster et al. [2007], the aerosols' actual contribution to the radiative forcing is still poorly understood. One reason for this is the high variability of aerosols in space and time, which is due to changing source strength, varying synoptic or local advection, and their short lifetime of up to several weeks [Haywood and Boucher, 2000; Ramanathan et al., 2001]. Their optical properties vary due to diffusion and aging processes such as humidification, coagulation, gas to particle phase conversions, or washout due to precipitation [Schuster et al., 2006].

[3] Global coverage of aerosol optical depth (AOD) as a key parameter in determining the radiative forcing can be derived from satellite generated data. Large discrepancies still exist in the estimates of aerosol direct forcing among some modeling studies using satellite-derived AOD from different sensors by different retrieval algorithms [Li et al., 2009]. Comparisons of the global mean AOD value derived from various satellite products reveal significant deviations. The deficiencies in regional or short-term (monthly) means of AOD are even larger [Li et al., 2009]. Especially in a mountainous region such as the Alps, the aerosol content in the atmosphere is difficult to determine by means of satellite-derived estimates [Emili et al., 2010]. This is due to the complex topography and local effects such as distinct wind patterns or small-scale and short-term changes in albedo conditions especially during the winter months. A new promising approach byEmili et al. [2011]uses the Multiangle Implementation of Atmospheric Correction (MAIAC) approach of the Moderate Resolution Imaging Spectroradiometer (MODIS) to generate maps of AOD of the Alpine region with a spatial resolution of 1 km. For validating these types of products the necessity of ground-based AOD measurements is underlined. Knowledge of AOD on a local scale is further important because aerosols have a large impact on human health. They cause enhanced mortality due to cardiopulmonary diseases (heart and lung diseases), and contribute to respiratory illnesses such as asthma and bronchitis. Also allergies are found to be enhanced by the emission of traffic-related aerosols [Pöschl, 2005].

[4] Sun photometers are deployed in networks, such as the Aerosol Robotic Network (AERONET) [Holben et al., 1998, 2001] or the Global Atmosphere Watch (GAW) aerosol program of the World Meteorological Organization (WMO) [Wehrli, 2002; Baltensperger et al., 2005]. They provide AOD for a large variety of sites distributed over the globe. In Innsbruck, AOD is measured with a Precision Filter Radiometer (PFR). This type of Sun photometer was originally developed at the Physikalisch-Meteorologisches Observatory Davos, Switzerland (PMOD) for the WMO GAW aerosol network [Wehrli, 2000; Baltensperger et al., 2005]. PFRs are developed to measure background aerosol conditions and have taken part in Sun photometer intercomparisons for data quality assurance [McArthur et al., 2003; Kim et al., 2008]. The deviations between the PFR and other network Sun photometers, such as the CIMEL employed in the global AERONET network [Holben et al., 1998, 2001], are within the range of +/−0.01 AOD. This agreement is better than the desired uncertainty level of 0.015 AOD set up by the WMO [Baltensperger et al., 2005].

[5] Reports on Sun photometric AOD measurements in the Alpine region are sparse. Ingold et al. [2001] present data from the Swiss CHARM network also employing PFR instruments. AOD for the GAW background sites Jungfraujoch and Hohenpeissenberg is described by Wehrli [2008]. AOD measurements in Innsbruck with a handheld Sun photometer for selected days are given by Blumthaler et al. [1997] and Mascher [2000]. The first measurements in Innsbruck with the PFR have been performed on 19 selected days between January 2001 and September 2002 by Grabner [2003]. AOD derived from measurements of spectral UV irradiance for the Bavarian Alps is reported by Lenoble et al. [2002] and for French Southern Alps by Lenoble et al. [2008].

[6] The aim of this study is to provide an aerosol climatology for a site in an extremely complex terrain. Especially in a mountainous region influenced by local as well as regional weather patterns, the atmospheric aerosol concentration shows a high variability over short distances. With our highly accurate aerosol time series starting in 2007 we provide a basis for the validation and improvement of satellite-derived aerosol products. On the basis of a scheme byGobbi et al. [2007] relating AOD, the Ångström parameter αand the second-order coefficientγ, we interpret the data with respect to particle size. This scheme has previously proven to be beneficial in various geographic regions with different aerosol characteristics [Basart et al., 2009; Kaskaoutis et al., 2010, 2011; Gerasopoulos et al., 2011].

2. Materials and Methods

2.1. Measuring Site

[7] Aerosol optical depth is being measured by Sun photometry in Innsbruck, Austria (47.26°N, 11.39°E, 620 m above sea level (asl)) since January 2007. The measuring site is located on top of the 10-story building of the Leopold-Franzens University in the city center. It is within the atmospheric boundary layer representing urban conditions with a low to moderate pollution level. Innsbruck itself is situated in the Inn valley, which is together with the Brenner Pass one of the major transit routes across the Alps. Heavy Goods vehicle traffic along such major Alpine transport routes strongly contributes to the air pollution in these areas [Beauchamp et al., 2004]. Besides the local sources, the aerosol conditions are also influenced by the local weather situation as shown by Ingold et al. [2001] and Grabner [2003]. Especially in winter, Foehn conditions represent an example of changing the aerosol conditions. Owing to very stable boundary layer stratification during an anticyclonic situation a cold pool without vertical mixing builds up leading to comparably high boundary layer aerosol concentrations [Harnisch et al., 2008]. Owing to the Foehn these cold pools are virtually wiped out of the valley and transport air with lower aerosol loading into an Alpine valley [Frioud et al., 2004]. Besides these short-term mountain specific aerosol situations, a yearly aerosol pattern typical for an urban midlatitude site should be expected with low AODs in winter and high AODs in summer.

2.2. Instrumentation

[8] Direct solar irradiance measurements with a four-channel PFR are exploited to derive aerosol optical depth at 368, 412, 500, and 862 nm. The single channels have a nominal bandwidth of 5 nm. The field of view has a full opening angle of 2.5° and a slope of 0.7°. The PFR is operated inside a weatherproof housing. A synthetic quartz window protects the entrance. An active Peltier type thermostatic system stabilizes the temperature of the detector head at 20°C +/− 1°C over a range from −30°C to +35°C of housing temperature. This greatly reduces the sensitivity to temperature changes and should also reduce aging of the interference filters. An automated shutter exposes the detectors only during actual measurements (1.5 s every minute) so that the dose dependent degradation is minimized [Wehrli, 2008].

[9] The PFR is mounted on a 2-Axis Tracker/Positioner by Kipp&Zonen configured to operate as a solar tracker. The angular resolution of its belt drive is 0.036°, the accuracy is better than 0.1°, and the repeatability for movement in the same direction is better than 0.05°. Since the diameter of the Sun is about 0.5° the features of the Sun tracker allow for optimal capturing of the Sun's disk in the field of view of the PFR.

2.3. Calculation of Aerosol Optical Depth

[10] Aerosol optical depth τa(λ) represents the extinction of radiation passing through the atmosphere due to aerosol particles at the wavelength λ. The Sun photometer measures the direct solar irradiance at four wavelengths. The measured signal in Volts follows the Lambert-Beer-Law:

display math

[11] V0(λ) is the extraterrestrial signal at the top of the atmosphere at wavelength λand standard Sun-Earth distance R of 1 astronomical unit, m is the optical air mass factor describing the length of the irradiance path through the atmosphere expressed as a multiple of the path length to the zenith, andτt is the total columnar optical depth of the atmosphere. Molecular scattering and gas absorption as well as the aerosol extinction contribute to the total optical depth τt. The exponent from equation (1) equals:

display math

[12] The aerosol air mass ma is approximated with the air mass for water vapor which is determined according to Kasten [1965]. The Rayleigh air mass mR is calculated after Kasten and Young [1989], and τR(λ), the Rayleigh optical depth, is determined according to Bodhaine et al. [1999] with the air pressure being measured on site with a CS100 Barometric Pressure Sensor by Campbell Scientific, Inc. Here mO3τO3(λ) represents ozone absorption. The ozone optical depth τO3(λ) = aO3*Ω is computed with band-weighted ozone absorption coefficients aO3 and the ozone column Ω, which is obtained from the Ozone Monitoring Instrument (OMI) on a daily basis. The ozone air mass mO3 is calculated according to Komhyr et al. [1989]. Absorption due to NO2 is described by mNO2τNO2(λ) with mNO2 being approximated by mO3, the ozone air mass, and τNO2(λ) is calculated from a climatological mean for the NO2 column amount of 2*1015 molec/cm2 [e.g., Valks et al., 2011].

[13] To obtain the aerosol optical depth τa(λ), equation (2) is substituted into equation (1) leading to

display math

[14] Uncertainties in the aerosol optical depth δτa are calculated according to Russell et al. [1993]. The main source of uncertainty is the uncertainty in the extraterrestrial voltage (V0). V0 for the PFR N08 were determined by comparison with standard instruments N01 and N26 in Davos under good conditions on 5 days in October 2006 shortly before the instrument was set up in Innsbruck. The U95 V0 uncertainties cover the 1 σ scatter in signal ratios and the difference of means μ for both reference instruments. The uncertainty of the V0 signal at 368, 412, 500, and 862 nm are 0.47%, 0.42%, 0.8%, and 0.56%, respectively. Depending on wavelength and air mass, δτa ranges between 0.001 and 0.008. The PFR δτa are in the order of the AERONET reference instruments (0.002 to 0.009) [Eck et al., 1999] and slightly lower than the ones reported by Holben et al. [1998] or Eck et al. [1999] for the AERONET CIMEL field Sun photometers, which have an uncertainty of less than 0.01 for λ > 440 nm and less than 0.02 for λ < 440 nm.

[15] The AOD used in this study are cloud screened according to Smirnov et al. [2000]with a few adaptations to the measuring site of Innsbruck and the measurement schedule of our PFR. AOD measurements are only considered if the air mass is lower than 6. For higher air masses the probability of cloud contamination is considered too high due to the low Sun elevation angle. Smirnov's triplet criterion has been adjusted to our measurement routine. If the AOD at a specific measurement is less than 0.2, the variation in the two preceding and two following AOD values is only allowed to be 0.02. If the AOD is larger than 0.2, it is allowed to fluctuate by not more than 0.03. Thus we applied a multiplet criterion considering the two preceding and the two following measurements spanning over a 5-min period. The diurnal stability check with the smoothness and standard deviation criteria are adopted without any change fromSmirnov et al. [2000].

[16] For the analysis of the aerosol optical depth the data are also checked for tracking error and housing temperature. Both parameters are stored with each measurement. If the tracking error is larger than 0.25° the data are excluded from the analysis. This criterion is only applied rarely. Data are also excluded from the analysis if the housing temperature exceeds 35°C, which happens frequently during summertime.

2.4. From Spectral Variation of AOD to Aerosol Size Distribution

[17] The spectral behavior of the aerosol optical depth can be used to infer information on the size distribution of the aerosol, which has been discussed by several authors [e.g., Kaufman, 1993; Eck et al., 1999; Schuster et al., 2006]. Ångström [1929] quantified the spectral behavior of the AOD in an empirical formula

display math

[18] Here λ is given in μm and β represents the aerosol optical depth at 1 μm. In the solar spectrum, the Ångström parameter α is an indicator of the particle size in the atmosphere: large values of α (α > 1) represent prevailing fine mode aerosols with radii < 1 μm, α < 1 points to a high fraction of coarse mode aerosols (r > 1 μm). By taking the logarithm of the Ångström law (equation (4)) α is determined as the negative slope of the linear equation:

display math

[19] Already Ångström [1964] manifested that α is not a constant and depends on the wavelengths chosen for its calculation. By adding a quadratic term to equation (5), the more precise empirical relationship between aerosol extinction and wavelength results in

display math

[20] It was first suggested by King and Byrne [1976] and taken on by a number of authors afterward [e.g., Eck et al., 1999; Schuster et al., 2006; Atkinson et al., 2010; Kaskaoutis et al., 2010, 2011]. The quadratic coefficient γ represents the curvature of the ln(τ)-ln(λ) relationship. A negative γ indicates aerosol size distributions dominated by the fine mode and a positive γ indicates size distributions with significant coarse mode contribution [Schuster et al., 2006]. The deviation of the ln(τ) versus ln(λ) fit from linearity becomes significant for dominant fine mode aerosols, which occur during conditions with τ(500 nm) < 0.1. Such an aerosol environment is expected in Innsbruck and therefore the coefficient γ is an essential parameter to analyze aerosol conditions in Innsbruck. In our study α and β are calculated according to equation (5) as a linear regression of the four wavelengths measured with the PFR (368, 412, 500, and 862 nm). Here γ is computed by a nonlinear regression with equation (6) at the four PFR wavelengths.

[21] Gobbi et al. [2007] suggest a graphical framework to gain information on the fraction of the fine mode aerosol contribution η to the total aerosol optical depth as well as the modal radius Rf of the fine mode aerosol. Furthermore, AOD increases due to aerosol humidification can be separated from AOD increases due to the addition of coarse particles as in the case of cloud contamination [see Gobbi et al., 2007, Figure 1]. The robustness of the graphical framework of Gobbi et al. [2007] is shown by Basart et al. [2009] who characterize aerosol in northern Africa, the northeastern Atlantic, the Mediterranean Basin, and the Middle East. They compared the method of Gobbi et al. [2007] to the O'Neill fine mode algorithm products [O'Neill et al., 2003] and to Dubovik's sky radiance fine mode inversion products [Dubovik and King, 2000]. Results show good coincidence among the three methods in the coarse particle detection (η < 40%) as well as at high η values. The differences in ηare always <20%, but discrepancies were partly due to differences between the measurement frequencies of each data set (i.e., less or no sky radiance inversions with respect to direct-Sun measurements). As expected, the comparison shows a better agreement of Gobbi's graphical method with O'Neill'sη values since the outputs of Gobbi's graphical method are a subset of the products retrieved from the O'Neill algorithm [O'Neill, 2010].

[22] Gobbi's straightforward graphical framework was previously employed to study aerosol parameters in the Arabian Sea, the Bay of Bengalen [Kaskaoutis et al., 2010, 2011], and in Athens, Greece [Gerasopoulos et al., 2011]. In our study Gobbi's grid is adopted, but it was recalculated to the wavelength range of the Innsbruck PFR. Furthermore, γ (instead of the discrete difference Δα) is plotted in dependence of α. Using the quadratic coefficient γ supplies a general picture and allows for comparisons with other studies. The link between Gobbi's Δα and γ is given by the negative of the second derivative of equation (6) with respect to ln(λ): d2ln[τa(λ)]/dln(λ)2 = 2γ, and the discrete form of the curvature Δα/Δln(λ) = 2γ. Thus Δα = 2γ*Δln(λ) with Δln(λ) = 0.5[ln(λ3) + ln(λ2)] − 0.5[ln(λ2) + ln(λ1)] = 0.5[ln(λ3) − ln(λ1)]. In our case λ3 = 0.862 μm and λ1 = 0.368 μm; thus we obtain Δα = 2γ*0.5[ln(λ3) − ln(λ1)] = 0.85γ. Note, that our γ is the same as the coefficient a2 by Atkinson et al. [2010], but they present the spectral curvature α′ = −2γ.

[23] The Gobbi grid is employed on a routine basis. For the fine mode aerosols which prevail in Innsbruck due to urban and industrial pollution also quantitative information on the fine mode effective radius can be obtained with the method by Gobbi et al. [2007]. For special cases, such as Saharan dust events, when in addition to the fine mode aerosols also coarse mode aerosols are expected the King inversion algorithm [King et al., 1978] has been applied. With this inversion algorithm also quantitative information on the coarse mode radii are acquired. The King inversion algorithm is based only on direct solar irradiance measurements. More sophisticated inversion algorithms such as the one suggested by Dubovik and King [2000] need input derived from sky radiance measurements, which are not available in Innsbruck on a routine basis. The King inversion algorithm assumes only spherical aerosol particles with a constant refractive index over the range of observation. Also, the refractive index has to be supplied a priori, which is set to 1.4–0.001i according to the refractive index used by Gobbi et al. [2007] to determine the fine mode fraction/radius grid. The inverted size distribution maintains its shape under various values of the real part of the refractive index and is quite insensitive to the imaginary part [King et al., 1978]. It is discussed by King et al. [1978] and King [1982] that size distributions from inverted AOD spectra are nonunique. They depend on the range of radii used, the initial guess distribution, and on iteration number and bin size. To reduce the nonuniqueness, the range of radii has been altered during the inversion procedure with lower limits between 0.02 and 0.08 μm and upper limits between 2 and 10 μm. The iteration number is set to eight iteration steps and the bin size ranges between 15 and 50. Three initial guess Junge type size distributions of the form n(r) = crν* [Junge, 1955] are used in the King inversion despite the fact that the radius extends between 0 and ∞. C and ν* are coefficients that characterize the size distribution. Junge [1955] showed that ν* ≈ α + 2 for nonabsorbing aerosols with α > 1. The initial guess size distributions vary in ν* with ν* = α + 1.5, ν* = α + 2, and ν* = α + 2.5. Only cases where the resulting size distribution is not or marginally sensitive to varying initial radii, ν* values and bin size are used as additional information on the aerosol size distribution and modal coarse mode radii.

3. Results

3.1. Database

[24] Table 1 presents the number of available AOD measurements in various time intervals in Innsbruck during the 5 years of AOD sampling. The AOD climatology is based on quality controlled daily means. Quality control consists of a cloud screening described in the previous section. The data are also checked for tracking error and housing temperature. During the 5 years of measurement the PFR operated continuously with a few exceptions (see second row in Table 1). In 2011, the instrument operated every single day. It was the year with most measurement days due to an exceptional dry and clear period in October and November. In 2009, the Sun photometer operated without any interruptions, only on 2 days (26 and 27 September) the PFR was not operating due to maintenance. From the end of April throughout May 2008 the PFR took part in a campaign in Kolm-Saigurn, Austria. Therefore 32 days of AOD data from Innsbruck are missing. In 2007 and 2010 the data gaps are rather long. From 25 April until 31 July 2007 the PFR was deployed at the University of Natural and Applied Sciences, Vienna, Austria. The data gap from October to December 2007 is due to a hardware failure. From September to December 2010 a hidden software malfunction prevented data recording. All together data were collected on 1583 days; 790 days passed quality control and are used for further data analysis.

Table 1. Database for AOD Measurements in Innsbruck From 2007 to 2011
Number of single measurements passed QC35,74643,74747,65332,69769,499229,342
Number of measurement days2613353632593651583
Number of days passed QC112162182109225790

3.2. AOD Characteristics

[25] Figure 1 shows the time series of daily mean aerosol optical depth at 500 nm. The wavelength of 500 nm is chosen to be displayed in all following figures and tables to maintain comparability with other studies, for example those performed with data from AERONET, the largest global network for AOD measurements [Holben et al., 1998]. A comparison of our results with other GAW aerosol studies remains possible. The daily mean AOD is calculated as arithmetic mean only if at least 50 minutely AOD values passed the quality control check in accordance with Wehrli [2008]. AOD values lower than 0.05 occur mainly in the winter months. The minimum daily AOD with 0.03 occurred on 22 December 2010. The rest of the year the variability of the AOD is rather high with daily mean values up to 0.48 measured on 11 September 2008.

Figure 1.

Time series of daily mean aerosol optical depth (AOD) at 500 nm measured in Innsbruck.

[26] The frequency distribution of daily AOD at 500 nm is shown in Figure 2. The frequency distribution can be well represented by a lognormal probability distribution. This finding is in accordance with the study of O'Neill et al. [2000] who investigated multiyear AOD time series from various locations. They concluded that AOD in general is better represented by a lognormal distribution than a normal distribution with according standard deviation. The overall arithmetic mean of the AOD at 500 nm 〈τ〉 with associated standard deviation σ, the geometric mean /τ/ with associated standard deviation μ as well as the median {τ} of the 5-year data set is also displayed inFigure 2.

Figure 2.

Frequency distribution of daily means of AOD at 500 nm measured in Innsbruck from 2007 to 2011. The lognormal fit to the data is shown in red. The arithmetic mean 〈τ〉 with standard deviation σ, the geometric mean /τ/ with associated standard deviation μ and the median {τ} of the hourly means are indicated as vertical lines.

[27] Even though O'Neill et al. [2000] suggests to present aerosol data with respect to a lognormal distribution with geometric mean and its associated standard deviation, a number of authors only present the arithmetic mean [e.g., Che et al., 2011], some present various percentiles [e.g., Estellés et al., 2007] and some the median, skewness and kurtosis [e.g., Bouya et al., 2010], or at least a modal value [Bi et al., 2011]. To retain comparability with other studies, Table 2presents the overall monthly means of AOD at 500 nm in Innsbruck for the 5-year period from 2007 to 2011 with the geometric and arithmetic mean as well as the median.

Table 2. Overall Monthly Means of AOD at 500 nm in Innsbruck for the 5-Year Period From 2007 to 2011 With the Geometric and Arithmetic Mean as Well as the Mediana
  • a

    Overall monthly mean arithmetic 〈τ〉 and geometric /τ/ with the associated standard error of the mean (SEM) is shown in the first two columns. The third column displays the median {τ} at 500 nm. The overall monthly mean α and γ with their SEMs are given in columns four and five. The last column shows the number of daily means N used for calculating the SEM.

Jan0.101 ± 0.0020.087/* 1.0170.0751.42 ± 0.01−0.52 ± 0.0262
Feb0.120 ± 0.0030.106/*1.0200.0921.44 ± 0.01−0.50 ± 0.0273
Mar0.151 ± 0.0030.135/*1.0190.1391.48 ± 0.01−0.53 ± 0.0170
Apr0.181 ± 0.0040.164/*1.0240.1601.48 ± 0.01−0.48 ± 0.0186
May0.173 ± 0.0030.164/*1.0140.1581.48 ± 0.02−0.47 ± 0.0047
Jun0.167 ± 0.0070.151/*1.0370.1531.62 ± 0.01−0.57 ± 0.0152
Jul0.191 ± 0.0050.178/*1.0210.1781.67 ± 0.02−0.53 ± 0.0168
Aug0.164 ± 0.0040.151/*1.0250.1591.67 ± 0.01−0.54 ± 0.0195
Sep0.166 ± 0.0040.153/*1.0250.1491.59 ± 0.02−0.60 ± 0.0171
Oct0.143 ± 0.0030.131/*1.0200.1251.50 ± 0.02−0.62 ± 0.0166
Nov0.092 ± 0.0010.083/*1.0200.0831.38 ± 0.02−0.58 ± 0.0158
Dec0.091 ± 0.0040.082/*1.0330.0691.54 ± 0.01−0.61 ± 0.0242

[28] Figure 3shows the multiannual monthly geometric mean AOD at 500 nm for the 5-year period from 2007 to 2011 and its associated standard error of the mean (SEM). By showing the SEM it is possible to evaluate the variability of the monthly mean values. The geometric standard deviation is also depicted inFigure 3to show the variability of the daily data within a month. The climatological monthly means are calculated from daily means. In the long-term observation the month of July shows the highest AOD with 0.18. Long-term mean AOD below 0.1 occur in November, December, and January. The low SEMs of the multimonthly means (i.e., variability of the monthly means) indicate a significant yearly cycle.

Figure 3.

Multiannual monthly mean AOD at 500 nm, standard error of the mean (SEM; broad bars; Table 2, column two) and standard deviation (thin bars) for the observation period (2007–2011).

[29] A yearly statistics of the aerosol optical depth at 500 nm including the yearly mean AOD, its SEM, as well as the median, all based on daily AOD is shown in Table 3. For the yearly means the geometric mean AOD is closer to the median AOD than to the arithmetic mean AOD and ranges between 0.12 and 0.14 for the 5-year period. The year-to-year variability is not very high. The year 2007 has the lowest aerosol loading. The highest yearly AODs occur in 2009.

Table 3. Yearly Mean Arithmetic 〈τ〉 and Geometric /τ/ at 500 nm and Associated Standard Error of the Mean (SEM) for a 5-Year Period (2007 to 2011) in Innsbrucka
  • a

    The median {τ}, α, and γ with SEM are shown as well. The number of daily means used for calculating the SEM is given in the last row of Table 1. The 5-year period is 2007–2011.

20070.129 ± 0.0040.117/*1.0320.1151.48 ± 0.01−0.52 ± 0.01
20080.135 ± 0.0040.122/*1.0290.1161.53 ± 0.01−0.51 ± 0.01
20090.158 ± 0.0030.144/*1.0850.1351.49 ± 0.01−0.56 ± 0.01
20100.157 ± 0.0050.138/*1.0350.1351.52 ± 0.01−0.59 ± 0.01
20110.137 ± 0.0020.128/*1.0180.1291.58 ± 0.01−0.55 ± 0.01
Overall0.150 ± 0.0030.131/*1.0190.1311.53 ± 0.01−0.54 ± 0.01

3.3. The α and γ Characteristics

[30] On most days α ranges between 1.2 and 2, indicating a prevalence of fine mode aerosol particles in Innsbruck (see Figure 4). The overall mean α is 1.53 +/− 0.28 (+/−1σ). The α time series shows a slight tendency of higher values in summer, which can be seen in the multiannual α values shown in Figure 5. The SEM as well as the standard deviation is also shown in Figure 5 to show the variability of the monthly means as well as the variability of the daily mean α within a certain month, respectively. Table 2 shows the overall monthly mean α in June, July, and August being higher than 1.6. The months with the lowest α remaining below 1.45 are January, February, and November. There are days with α being as low as 0.3 indicating coarse mode aerosols to be present on single occasions. These days are not restricted to the months with low α.

Figure 4.

Time series of daily mean α and γ for Innsbruck from 2007 and 2011.

Figure 5.

Multiannual monthly means of α (black) and γ(blue) for the 5-year period (2007–2011) of AOD measurements in Innsbruck. Broad error bars indicate the standard error of the mean (SEM;α: magenta, γ: green); thin error bars indicate the standard deviation (α: red, γ: cyan).

[31] Figure 6 displays the dependency of α on the AOD at 500 nm also used by Holben et al. [2001]. The majority of points clusters in the 0.1 AOD/1.5 α – region. A low α below 0.5 occurs for AOD at 500 nm between 0.2 and 0.5; the lowest α of approximately 0.3 occur for high AOD at 500 nm between 0.37 and 0.5.

Figure 6.

Dependence of daily mean α on the daily mean AOD at 500 nm.

[32] The time series of daily means of the quadratic coefficient γ is also presented in Figure 4. In most cases γ remains negative with values between −1 and 0 indicating the prevalence of fine mode aerosols. A few days with γ being positive (significant coarse mode contribution) are also observed mostly coinciding with low α. The overall mean γ is −0.54 +/− 0.25 (+/−1σ). Table 2 and Figure 5 indicate the lowest multiannual γ occurring in October (−0.62) and the maximum multiannual monthly mean γ being observed in May with −0.47. The time series of γ as well as the time series of α indicate a prevalence of small aerosol particles in Innsbruck. On single occasions large particles can be observed associated with low α and high γ.

[33] Gobbi et al. [2007] present a straightforward graphical framework to obtain information on the aerosol size distribution, which is applied in Figure 7 to the aerosol distribution measured in Innsbruck from 2007 to 2011. In this graphical method the quadratic coefficient γ is plotted as a function of α in a special grid. The blue gridlines in the plot identify the fraction η of the fine mode particles from the total aerosol particles in percent. The black gridlines each portray a fixed radius size Rf of the fine mode particles in micrometers. The reference points for this classification grid is calculated for a bimodal, lognormal size distribution assuming spherical particles with a refractive index n = 1.4 − 0.001i which is typical for urban/industrial aerosol. The level of indetermination of this classification scheme is of the order of 25% for Rf and about 10% for η for a refractive index varying between n = 1.33 − 0.0i and n = 1.53 − 0.003i [Gobbi et al., 2007; Basart et al., 2009]. Possible nonsphericity of particles is not expected to impact significantly on the results [Gobbi et al., 2007]. The aerosols are classified according to their AOD with different colors. Since both α and γ are derived from spectral τ values, propagation of errors indicates the relevant indetermination to be Δα/α ∼ 2Δτλ/τλ and Δγ/γ ∼ 5Δτλ/τλ [Gobbi et al., 2007]. With our upper uncertainty in AOD of 0.008 this translates to uncertainties of > 16% in α and >40% in γ for τ < 0.1, and uncertainties of >8% in α and >20% in γ for τ < 0.2, and less for larger AOD. Gobbi et al. [2007] only use AOD > 0.15 to avoid uncertainties >30% with an error in AOD of 0.01. In Figure 7 we use all AOD values. The color code refers to different AODs, associated with different uncertainties. Figure 7 shows in the majority of cases a fine mode fraction of more than 70% with a fine mode radius between 0.1 μm and 0.15 μm. For the rare cases with a fine mode fraction of 30% or less the AOD is always larger than 0.2.

Figure 7.

Nonlinear Ångström coefficient γ as a function of α and AOD at 500 nm (color code) for Innsbruck in the years 2007 to 2011. Daily means are presented. The fine mode fraction η (blue grid lines) and the fine mode radius Rf (black gridlines) serve as a classification framework to infer information on the aerosol size distribution [see Gobbi et al., 2007]. Saharan dust events and the volcanic ash event are marked with special symbols.

3.4. Saharan Dust

[34] On a few days per year it is possible to identify Saharan dust in Innsbruck. Saharan dust events are characterized by rather high AOD compared to the previous and following days. Wiegner et al. [2011]suggests an AOD of more than approximately 0.3 in the visible spectral range for Munich, southern Germany. This roughly applies to our results, too. The surface-area-weighted mean radii (effective radii) of desert dust size distributions for dust events in continental Europe is reported to be about 0.8–2μm [Ansmann et al., 2003]. A modal coarse mode radius around 2 μm is used for Saharan dust particles in aerosol models [Koepke et al., 1997], which is also in accordance to results by Dubovik et al. [2002]. Cachorro et al. [2006] state Saharan dust particles being distributed over a wide range of sizes with effective radii ranging from 0.6 to 15 μm. Owing to the larger particle size of the Saharan dust compared to the urban/pollution aerosol a bimodal size distribution is expected during Saharan dust events, which results in a smaller fine mode fraction (η < 50%) in the Gobbi plot. In addition, a small α and a positive γ are expected during a Saharan dust event. Dubovik et al. [2002] report 1.2 > α > −0.1 during dust events and Wiegner et al. [2011] state α around 0.5 for a specific dust event over Munich. The decision on the occurrence of Saharan dust over Innsbruck is by visual inspection of the diurnal course of the AOD as well as the Gobbi plot keeping the AOD, α and γ values in mind that are mentioned above. In case Saharan dust is suspected, trajectories are looked at to prove the origin of the air mass. Example days with Saharan dust over Innsbruck are 25 June, 11 September, 13 October 2008, and 25 May 2009. On these four example days, α was 0.58, 0.28, 0.26, 0.54, and γ 0.06, 0.08, −0.01, 0.02, respectively. The AOD on these four days is also rather high with 0.47, 0.48, 0.41, and 0.24. The day 13 October 2008 is the day with the lowest α, and 11 September 2008 is the day with the highest daily mean AOD. In the αγ plots only a small fraction of the fine mode particles should be expected. The Saharan dust events are shown in Figure 7 as special symbols with AOD color code.

[35] The fine mode radii are between 0.1 μm and 0.15 μm, where the majority of fine mode radii in the long-term observations of Innsbruck aerosols are detected (seeFigure 7). Only on 13 October 2008 the fine mode radius is larger than 0.15 μm. On 11 September and 13 October 2008 the fine mode fraction is less than 30% and it is less than 50% on 25 June 2008 and 25 May 2009. This indicates the Saharan dust aerosols being superimposed on the usual Innsbruck aerosol conditions leading to a bimodal size distribution. To also gain information on the coarse mode radius, the size distribution calculated with the King algorithm (see section 2.4) is exemplified in Figure 8 for 25 May 2009. The fine mode radius derived from the King inversion is around 0.28 μm and the coarse mode radius is at 1.44 μm. The fine mode radius is around 0.15 μm larger than shown in Figure 7, which is due to uncertainties in the calculation of the fine mode fraction/effective radius grid [Basart et al., 2009] and the uncertainties involved in the King inversion algorithm. Considering the rough estimate of the refractive index, the differences in fine mode radius derived from both methods are in reasonable agreement. The coarse mode particles dominate the size distribution with a radius of 1.44 μm. This is within the range of effective radii reported by Dubovik et al., [2002], Ansmann et al. [2003], or Wiegner et al. [2011] for Saharan dust particles and also in the range employed in aerosol models [Koepke et al., 1997].

Figure 8.

Particle size distribution of aerosols for Innsbruck on 25 May 2009 between 07:00 and 09:00 UTC during a Saharan dust event. In the King inversion, the radius limits are set to 0.08 μm and 9 μm and 40 coarse bins are used. The bimodal distribution shows a radius mode around 0.28 μm and one at about 1.44 μm.

[36] The Saharan dust events are further indicated by back trajectories. The trajectories have been calculated online on the Web site of the flextra trajectory model ( [Stohl et al., 1995] hosted by the Norwegian Institute for Air Research (NILU). Trajectories from the HYSPLIT model ( (R. R. Draxler and G. D. Rolph, HYSPLIT (HYbrid Single-Particle Lagrangian Integrated Trajectory) Model, 2012, show similar results. Because of easy access to both websites the trajectories are not shown here. In the HYSPLIT model the trajectories are started in Innsbruck. The flextra back trajectories are started at the Zugspitze, which is the closest location to Innsbruck for the back trajectories available on the flextra Web site. The direct distance between the Zugspitze and Innsbruck is about 30 km. Back trajectories derived from flextra and HYSPLIT for the four selected Saharan dust events (25 June 2008, 11 September 2008, 13 October 2008, 25 May 2009) indicate the same region of end points in northern Africa.

3.5. Volcanic Ash

[37] On 14 April 2010 the main eruption of the Icelandic volcano Eyjafjallajökull injected an ash plume into the middle and upper troposphere. Owing to the atmospheric flow pattern the ash plume was transported over central Europe and arrived in the Alpine region on 17 April 2010 [Emeis et al., 2011]. On 17 April 2010 this ash plume was detected by the Sun photometer measurements in Innsbruck. The daily mean AOD at 500 nm was 0.48, which is about 0.3 larger than the average AOD for the month of April (see Table 2). Toledano et al. [2012] have observed an increase in AOD of 0.2–0.4 due to the Eyjafjallajökull ash plume at Spanish measurement sites being similar to the Innsbruck enhancement. On this day α was equal to 0.87 and γ was −0.09. The daily mean γ as a function of α for 17 April 2010 is displayed in Figure 7 (blue diamond). The fine mode fraction is between 50% and 70%. The volcanic ash aerosols are superimposed on the regular aerosol conditions. Independent ceilometer measurements in Innsbruck identified the volcanic ash from the Eyjafjallajökull eruption [Schreiter, 2010], and enhanced PM10 values due to the volcanic ash were detected in Innsbruck [Schäfer et al., 2011]. Innsbruck is at the periphery of the ash plume with a concentration around 60 to 70 μgm−3 at an altitude of 3.5 km. Ash plume concentrations as high as 1.1 mgm−3have been reported for Maisach near Munich at the same altitude for 17 April 2010. These mass concentrations have been modeled with a coupled meteorology-atmospheric chemistry model (MCCM) and confirmed with lidar measurements [Wiegner et al., 2012].

[38] For 17 April 2010 the inverted size distribution is shown in Figure 9. For a variety of ν* values the size distributions deliver nearly the same results indicating a trustful solution. A bimodal size distribution with radii modes around 0.25 μm and 0.9 μm can be identified. The fine mode around 0.25 μm represents the usual conditions in Innsbruck which are dominated by urban aerosols. The King inversion algorithm overestimates the fine mode radius by 0.1 to 0.15 μm in comparison to Gobbi's graphical method. Considering the uncertainties involved in the King inversion and in the indetermination of the Gobbi grid, the agreement in the fine mode radius derived from both methods is in the expected range of uncertainty. The second mode with the higher effective radius around 0.9 μm can be attributed to the volcanic ash. This particle radius is within the range of effective particle radii between 0.75 μm and 1.7 μm as observed in the volcanic ash plume from the Eyjafjallajökull by Gasteiger et al. [2011].

Figure 9.

Particle size distribution for 17 April 2010 between 11:00 and 11:30 UTC calculated during the Eyjafjallajökull volcanic ash event. In the King inversion, the radius limits are set to 0.08 μm and 5.9 μm and 40 coarse bins are used. The bimodal distribution shows a radius mode around 0.25 μm and one at about 0.9 μm.

4. Discussion

[39] The Innsbruck (620 m asl) aerosol pattern shows similarities with aerosol patterns of other Alpine locations such as the Swiss low altitude sites Locarno (366 m asl), Payerne (490 m asl) or Changins (430 m asl) operating in the CHARM network. Monthly mean AOD at 500 nm at the Swiss stations range between 0.05 and 0.1 in winter and up to 0.25 in summer [Ingold et al., 2001]. This seasonality for midlatitude sites was already pointed out by Ångström [1964].

[40] The closest GAW station performing AOD measurements with a PFR is Hohenpeissenberg (989 m asl) about 50 km southwest of Munich and 75 km north of Innsbruck. Its AOD climatology is similar to the Innsbruck climatology shown here. Monthly means of AOD at Hohenpeissenberg are slightly lower than in Innsbruck, which may be due to the higher elevation. Especially in winter, the Hohenpeissenberg minimum daily AOD around 0.01 [Wehrli, 2008] are lower by a factor of three to five compared to Innsbruck with wintertime minimum daily means between 0.03 and 0.05. This finding is supported by the multiannual monthly mean AOD. At Hohenpeissenberg the multiannual monthly mean in December and January is 0.03 [Wehrli, 2008]. In Innsbruck it is 0.08 for December and 0.09 for January thus by a factor of three higher. In summer, the difference between the multiannual monthly mean is lower. At Hohenpeissenberg the AOD stays rather constant from April to November with 0.12 to 0.13 [Wehrli, 2008]. In Innsbruck, the summertime AOD from April to September is between 0.15 and 0.18, which is about 25% higher than the Hohenpeissenberg AOD. The large wintertime difference between the two locations is due to their geographic locations. The Hohenpeissenberg measurement site is situated on an isolated mountain rising about 500 m above the surrounding terrain. It is often above wintertime inversion layers which favorably form during anticyclonic weather situations. These inversion layers are characterized by hardly any vertical mixing of air. Aerosols having their main sources in the boundary layer are trapped in the inversion layer not reaching the Hohenpeissenberg measurement site. Innsbruck is situated in the Inn valley within the boundary layer. Thus the boundary layer aerosols are part of the measured columnar AOD in Innsbruck.

[41] Despite the seasonal differences in AOD between the valley site Innsbruck and the elevated GAW site Hohenpeissenberg, the dependence of α on AOD at 500 nm (Figure 6) shows a similar behavior for both sites (compare Figure 6 for Innsbruck and Figure 27 from Wehrli [2008]). This indicates the industrial/urban aerosol type with a moderate level in AOD (<0.3) being observed in the majority of cases at both sites. These fine mode aerosols originate in local sources as well as in midrange and long-range transport.

[42] The local aerosol sources are traffic and industrial pollution as well as wood burning in the wintertime [Harnisch et al., 2008], which are affected by local thermally driven wind patterns such as up-slope winds during daytime [Gohm et al., 2009; Schnitzhofer et al., 2009]. Furthermore, regional wind systems can lead to air pollutant transport from regions nearby the Alps into the valleys (up-valley wind) and long-range transport may cause strong particle concentration peaks [Seibert et al., 2000; Kaiser, 2009]. Coincidence of higher AOD (0.4 < AOD < 0.5) with lower α (α ∼ 0.5; see Figure 6) indicate a considerable number of coarse mode particles in the atmospheric column, which can mainly be attributed to long-range transport of mineral dust particles originating in the Sahara.

[43] The prevalence of fine mode particles from urban/industrial pollution in Innsbruck is indicated of the αγ dependence shown in the Gobbi grid (Figure 7). The qualitative size distribution of the aerosol particles in Innsbruck is comparable to the one of Ispra, located south of the Alps in northern Italy [Gobbi et al., 2007]. At both sites α is around 1.5. Gobbi et al. [2007] do not use our γ but a difference between the Ångström exponent α derived from two wavelength intervals δα = α(440,675) − α(675,870). The connection between δα and γ is presented at the end of section 2.4. Applying this translation to the Ispra δα, Ispra and Innsbruck show γ being in the −0.5 region. From the graphical framework it can be seen that in Ispra as well as in Innsbruck fine mode aerosols with radii between 0.1 and 0.15 μm dominate. In Ispra, AOD growth is linked to coagulation-aging and hydration-type increase in Rf [Gobbi et al., 2007]. In Innsbruck, high AOD occurs in any range of fine mode fraction (see Figure 7). Growth in AOD can be attributed to the addition of coarse mode particles, which would result in a movement along constant Rf lines perpendicular to constant η lines. This feature is observed during the advection of Saharan dust or the formation of cloud droplets not being detected by the cloud screening algorithm. Figure 10 exemplifies days with increasing AOD due to humidification and in turn cloud droplet formation.

Figure 10.

(a) Daily time series AOD for 3 selected days with (b) corresponding αγ plot showing AOD growth due to the addition of coarse mode particles in the form of hydrated aerosols or just formed cloud droplets.

[44] Figure 10 (top) shows three diurnal cycles of AOD at 500 nm increasing in the course of the day. This increase in AOD is due to subvisible and forming clouds in the vicinity of the sun. The presence of clouds in the sky on these 3 days is verified by cloud camera pictures (not shown here) [Kreuter et al., 2009]. Because of low time resolution compared to the AOD measurements the cloud camera pictures cannot be unambiguously taken for cloud detection. They can be used to judge the overall cloud condition. The cloud screening algorithm uses predetermined criteria to distinguish between clear and obscured optical path between the Sun and the Sun photometer. Cloud screening criteria are always subjective due to the transition between aerosols and clouds. In the vicinity of a cloud the humidity is increased [Radke and Hobbs, 1991]. Aerosols humidify and increase in size until they become cloud droplets. This is a gray zone in any cloud detection algorithm. Koren et al. [2007] show a systematic increase in AOD in this “twilight zone” as the time interval between the measurement and the nearest cloud decreases. The Ångström coefficient α decreases (see Figure 10, bottom) the nearer to the cloud, suggesting greater contribution of large particles or undetected cloud droplets to the observed AOD closer to the cloud. The change in aerosol conditions near a cloud can be seen in Figure 10 (bottom) where a decrease in α together with an increase in γ occurs along constant Rflines implying the addition of coarse mode particles as reason for AOD growth. The cloud screening was established to find a balance between eliminating cloud contaminated and keeping cloud-free measurements. The separation between aerosols and clouds is a smooth transition and the enhancement of AOD in the vicinity of clouds is based on both thin clouds and humidified aerosols [Koren et al., 2007].

5. Conclusion

[45] This study presents and analyzes aerosol data gathered during a 5-year period at an Alpine station. This database is ideal for supporting the validation and comparison of aerosol retrieval products from satellites with high spatial resolution in a complex mountainous terrain.

[46] Especially in a valley, such as the Inn valley where Innsbruck is situated, local atmospheric patterns introduce specific situations which are superimposed on advected aerosols by long-range transport. Aerosol can be trapped under stable conditions in the boundary layer but also in higher residual layers, which can exist due to the valley topography. Underlining geographic differences on a regional scale, significant differences in the AOD seasonality at an Alpine valley site (Innsbruck) and a mountain site just north of the Alps (Hohenpeissenberg) are identified.

[47] The fine mode fraction and effective radius grid introduced by Gobbi et al. [2007]is successfully applied to classify aerosols regarding their size and to identify changes in aerosol properties over the daily course. Humidification and the addition of coarse mode particles by long-range transport are detected in daily time series. In a future application results from the Gobbi grid analysis can be incorporated in a refined iterative cloud screening process. Such a new approach can complement existing cloud screening algorithms.

[48] Our Sun photometric aerosol data complement the in situ aerosol measurements performed at several stations in the Inn valley by governmental authorities. This study will further improve the understanding of the relationship between remotely sensed AOD and in situ PM10 data in a highly complex terrain.


[49] We thank G. P. Gobbi for the recalculation of the fine mode/effective radius grid. The PFR is generously provided by the Institute for Meteorology and Geophysics of the University of Innsbruck. CMS - Ing. Dr. Schreder GmbH allocated the Sun tracker for our use. NILU is acknowledged for providing the FLEXTRA trajectories ( used in this study. The authors gratefully acknowledge the NOAA Air Resources Laboratory (ARL) for the provision of the HYSPLIT transport and dispersion model and READY Web site ( used in this publication.