1. Introduction
Aerosol particles are important constituents of the earth’s atmosphere. They affect climate directly by scattering and absorbing solar radiation, and they affect climate indirectly by acting as cloud condensation and freezing nuclei (Haywood and Boucher 2000). The magnitude of aerosol forcing on climate is still considered highly uncertain because of, for example, their strong temporal and spatial variation. Further, aerosol particles affect air quality as well as visibility (Cabada et al. 2004), and they are known to have an impact on human health by contributing to numerous respiratory (Brunekreef and Holgate 2002) and cardiovascular ailments (Von Klot et al. 2005) through yet unknown mechanisms.
Aerosol particles are traditionally divided into four different modes based on their size: nucleation mode (Dp < 25 nm), Aitken mode (25 nm < Dp < 90 nm), accumulation mode (90 nm < Dp < 1 μm), and coarse mode (Dp > 1 μm), where Dp is the particle diameter. Alternatively, nucleation and Aitken modes can be denoted as ultrafine particles. All size modes are important: small particles are the most dangerous regarding their health impacts and large particles are the most efficient in directly impacting climate. Accurate information about the particles’ physical properties and local concentrations are best obtained using in situ measurements, but remote sensing measurements are the best way to observe the spatial and temporal distributions of aerosols with adequate resolution and coverage. The most important aerosol remote sensing techniques are satellite-based spectrometric measurements and ground-based lidar measurements. The former provide an excellent spatial coverage, but there are also many challenges, especially when continental boundary layer aerosol particles are of interest. In addition, satellite instruments are unable to observe aerosols below clouds. Likewise, lidars have restrictions related to, for example, rainy or misty weather, but their advantage is an excellent vertical resolution and the ability to make measurements near the ground. This is of an utmost importance, for example, in air-quality applications.
Elastic backscatter lidar is a traditional form of lidar where the transmitter emits laser pulses at a single wavelength and the detector measures the backscattered signal at the same wavelength. For research purposes, elastic backscatter lidars have been used, for example, to characterize atmospheric boundary layer and its aerosol content. Lagrosas et al. (2005) studied the correlation between portable automated lidar data and suspended particulate matter concentration at Ichihara, Japan, and found good correlations when either the lidar signal showed heavy backscatter near the ground or the atmosphere was relatively clear. During the Development and Validation of Tools for the Implementation of European Air Quality Policy in Germany (VALIUM) measuring campaign (Hannover, Germany), Münkel et al. (2007) found fairly good correlations between the total mass of particles with diameters less than 10 μm (PM10 particles) and near-range backscattering from lidar in dry conditions [relative humidity (RH) < 62%] using a Vaisala CT25K ceilometer (previous generation to Vaisala CL31). The determination of boundary layer mixing height based on aerosol backscattering has also been studied with Vaisala CT25K and CL31 ceilometers (Emeis et al. 2004; Münkel et al. 2007). In this study, a Vaisala CL31 ceilometer-type lidar was used. Although ceilometers are designed for measuring cloud heights and vertical visibility, their performance is sufficient for monitoring boundary layer structures. These studies show that a simple commercial ceilometer-type lidar can provide us with useful aerosol-related information. They do not tell, however, whether ceilometers could be used to obtain quantitative information about aerosols, such as number concentrations or total volumes. One of the advantages of the CL31 ceilometer is that the backscattered signal can be detected at very close range, about 30 m above the ground (Emeis et al. 2004). Because the lowest detection range limit for more sophisticated research lidars can be as high as a few hundred meters (e.g., Gimmestad 2005), the motivation to do further investigations on whether the CL31 ceilometer can actually be used as a quantitative instrument in aerosol research is justified.
The aim of this work was to study the capability of using a commercial ceilometer-type lidar in aerosol research. Backscattering was measured with a lidar; for comparison, the aerosol optical properties were modeled theoretically based on particle size distributions measured in situ. To facilitate the comparison, the lidar measurements were conducted in a vicinity of an aerosol in situ measurement site. The theoretical computations also provided estimates for the relative importance of different physical factors (i.e., particle size, composition, and shape) that affect lidar backscattering by aerosol particles.
2. Theory
Accurate light scattering properties of aerosol particles are essential for conducting reliable remote sensing observations. The scattering properties of a single particle depend on the particle size, shape, and composition. Although it is known that the scattering properties of nonspherical particles can be significantly different than those of spherical particles, the assumption of spherical particles is still widely used in remote sensing applications. This can lead to large errors when retrieving aerosol optical properties. The assumption of particle shape can be critical when inverting lidar measurements because backscattering and extinction depend on the particle shape (Mishchenko et al. 1997; Gobbi et al. 2002).
a. Basic concepts
b. Modeling approach
The purpose of optical modeling is to establish how the optical properties of the particles (e.g., phase function, scattering, and extinction cross sections) depend on their physical properties. This allows us to use the in situ–measured physical properties to compute what the lidar returns should be. Because the particle shape and refractive index are not measured and thus need to be assumed, we limit ourselves to estimating an expected range for the lidar returns.
For spherical and homogeneous aerosol particles, one only needs to know the particle complex refractive index m and diameter to calculate their optical properties analytically and exactly using the Mie theory. For nonspherical particles, the situation is more complicated, as additional shape information is needed and there is no exact scattering theory for arbitrarily shaped particles. In this case, one can either use simple nonspherical particles such as spheroids, Chebychev shapes, or polyhedrons, or adapt numerical methods such as the discrete dipole approximation (DDA; Draine and Flatau 1994). All liquid aerosol particles can be safely considered spherical, but all solid particles are to some degree nonspherical.
Here, we assume that the impact of possible nonsphericity on the optical properties of our urban aerosol cannot be larger than that for mineral dust. Thus, we assume that, shapewise, the optical properties of our urban aerosol particles will lay between solutions for spheres and dust-like particles. Whether this range covers all the eventualities is a matter of debate, but we point out the following: both laboratory measurements (e.g., Volten et al. 2001; Muñoz et al. 2001) and computer simulations (e.g., Moreno et al. 2006; Veihelmann et al. 2006; Nousiainen et al. 2006) imply that collections of nonspherical particles tend to scatter light in a remarkably similar fashion and clearly differently from spherical particles. It thus seems plausible that an urban aerosol mixture composing of spherical liquid particles and a wide variety of differently nonspherical particles would have optical properties somewhere between the chosen extremes.
For the calculations, major-to-minor axes ratios from 1.0 to 2.6 both for oblate and prolate spheroids with an interval of 0.2 were used. The shape distribution thus consists of 17 different spheroids. The optical properties of spheroids were calculated using the so-called T-matrix method (e.g., Waterman 1971; Mishchenko and Travis 1998). It unfortunately suffers from numerical convergence problems when extreme axis ratios and particles that are large compared to the wavelength are considered. This forced us to exclude some of the most extreme axis ratios from the shape distributions for particle sizes above 8 μm in diameter, which led to a slight underestimation of the impact of shape on the results. However, because of the small number of large particles in the size distributions, the overall impact was very small.
The calculated optical properties were computed for 79 narrow size bins from 3 nm to 20 μm in diameter, assuming a uniform size distribution within each size bin. From these, the size-averaged optical properties could be calculated for any size distribution (in our case, the in situ size distributions) simply by weighing the mean optical properties for each size bin by the number of particles in that size bin and summing the results over the whole size distribution. Because of computational burden, only 5 particle sizes were used within each size bin, totaling 395 sizes for the whole range from 3 nm to 20 μm in diameter. A relatively small number of sizes was not a problem because the results were also averaged over 17 different shapes. Because the speed of computation was not an issue in the Mie calculations and because no shape averaging was involved, 250 sizes were used within each size bin for the spherical particles. Surface area equivalence was adapted to specify the spheroid size.
3. Measurements
a. Measurement site
The measurements were conducted at the Kumpula campus of the University of Helsinki, located in a heterogeneous urban area about 5 km from the center of Helsinki, Finland. The lidar and meteorological measurements were carried out on the roof of the Department of Physics about 25 m above the ground, whereas in situ aerosol measurements were conducted at the Station for Measuring Forest-Ecosystem-Atmosphere Relations (SMEAR) III, about 250 m southwest of the Department of Physics.
The most relevant meteorological information was obtained with a Vaisala MILOS 520 automatic weather station, an optical Present Weather Sensor PWD21 with enhanced prototype software, and a weighing rain gauge by OTT Messtechnik (Kempten, Germany). All these instruments were located at the roof of the Physicum building next to the lidar. The meteorological measurements from MILOS consisted, for example, of wind speed, wind direction, air temperature, relative humidity, and atmospheric pressure; however, PWD21 provided data about rainfall and visibility.
b. In situ measurements
Aerosol size distribution measurements at ground level were conducted at the SMEAR III station by combining results from two different instruments. Submicron size distribution was measured with a differential mobility particle sizer (DMPS; Aalto et al. 2001) for particles between 3 and 950 nm in diameter. The DMPS system consists of two parallel DMPS (twin DMPS) with closed-loop flow arrangements (Jokinen and Mäkelä 1997). The first set was dedicated to sizes of 3–50 nm in electrical mobility–equivalent diameter, which were classified with a short (L = 0.109 m) Vienna-type differential mobility analyzer (DMA; Winklmayr et al. 1991) operated with 20 min−1 sheath flow and 4 min−1 aerosol flow rates. Particles in this range were counted with a TSI model 3025 ultrafine condensation particle counter (CPC; Stolzenburg and McMurry 1991). The second system classifies particles between 10 and 950 nm with a medium length (L = 0.28 m) Vienna-type DMA with flow rates of 5 and l min−1 for sheath and sample flows, respectively. In the second system, particles were counted with a TSI model 3010 CPC.
The sample for the DMPS system was drawn through a circular stainless steel tube and a rain cover. The main inlet extends through the roof of the SMEAR III container to a height of approximately 4.5 m above the ground. The flow rate was approximately 50 min−1. The DMPS sample was extracted from the centerline of the main flow. Prior to size segregation and counting, the sampled particles were dried with a Topas silica gel drier to ensure that the measurements are conducted for dry particles. With the DMPS setup, aerosol particle number size distribution between 3 and 950 nm is obtained with 10-min temporal resolution.
Supermicron aerosol number distribution was monitored with a TSI 3321 aerosol particle sizer (APS), which classifies ambient aerosol particles based on their aerodynamic diameter. The lower and upper detection limits are at 550 nm and approximately 20 μm in diameter, respectively. The APS sampling setup consisted of a 16.7 min−1 Thermo Andersen PM10 inlet. Five per minute were directed to APS, four of which as sheath air and one as the aerosol sample flow. The inlet removes particles larger than about 10 μm in diameter from the sample flow, but the cutoff is not ideal. Because of this, a small fraction of over-10-μm particles were detected by the APS, but their contribution to total number concentration as well as their effects on backscattering were minuscule.
The in situ data consist of dry-aerosol size distributions with 10-min temporal resolution measured with DMPS and APS instruments. Together, the instruments covered a particle size range from 3 nm to 20 μm in diameter (particles from 10 to 20 μm being strongly undersampled). The two instruments classify the particles based on two different equivalent diameters: aerodynamic and electrical mobility–equivalent diameters for the APS and DMPS systems, respectively. The merging of the two size distributions was made by means of two parameters: shape factor κ and density ρ. The shape factor increases the DMPS sizes, whereas densities larger than unity move the APS size ranges toward smaller sizes. The overlapping size range is optically important and thus the parameters were varied between atmospherically relevant values. Occasionally, this produces unphysical sharp features in the merged size distributions. To suppress this artifact from the optical properties, the overlapping region was filtered with a five-point running mean. In a base case, κ = 1.4 and ρ = 1.8 g cm−3 (e.g., Bundke et al. 2002) were utilized to convert the measured size distributions into a combined distribution with 79 logarithmically equally spaced bins. Values larger than unity for κ indicate that the particles are assumed to be nonspherical, which is a good assumption for urban aerosol particles. Because the main identified chemical compounds in the Helsinki area for the PM2.5 are sulfate and ammonium (Pakkanen et al. 2001), the value of 1.8 g cm−3 for the base case density is justified.
c. Lidar measurements
The lidar measurements were carried out using a vertically pointing Vaisala CL31 ceilometer, which is a commercial elastic backscatter lidar originally intended for measuring cloud heights. It operates at 910-nm wavelength and detects clouds up to 7.5-km altitude, but the aerosol backscattering can be detected at most up to about 1–2-km altitude, depending on the meteorological conditions and the strength of aerosol backscatter. The temporal resolution of the lidar is 2 s, with each 2-s interval being an average of 16 384 transmitted pulses. The vertical resolution can be set to either 5 or 10 m, from which the latter was chosen. It is noted that the vertical resolution is nominal; in reality, adjacent range gates are not independent, as the lidar’s 110-ns pulse length results in a 16.5-m-long contributing volume. Because the lowest 20 m of the lidar measurements were affected by optical cross-talk and software correction, the lowest usable lidar measurement altitude was 30 m above the device.
As described in section 2, the volume backscattering coefficient β of aerosol particles can be derived from the received power [see Eq. (3)] if σ is known. Unfortunately, the lidar operates at a wavelength located within a water vapor absorption band, so the extinction is not solely due to scattering by aerosols and air molecules and could not be solved from the lidar ratio, even if σ for aerosols was known. In principle, one could also solve σ by conducting lidar measurements at different zenith angles and looking at the change in backscattered power resulting from varying pathlengths to a desired altitude. However, our measurement data showed that the assumption of horizontal homogeneity required by this method would generally not apply.
To obtain the contribution to β solely from aerosol particles, backscattering from air molecules and hydrometeors needed to be removed from β derived from the lidar measurement. Backscattering from hydrometeors can be very strong, and it is impossible to distinguish the aerosol contribution from the backscattered signal if hydrometeors are present within the measurement volume. The easiest way to exclude hydrometeor cases is simply to study dry days when there is no rain, drizzle, or mist present. Then, the lidar-measured β, from which molecular contribution is removed, can be expected to represent the aerosol contribution. This can then be compared with β calculated theoretically from the in situ–measured aerosol size distribution, even though some deviations resulting from high relative humidity (hygroscopic growth) and different sampling volumes still exist. The measured β were typically about twice the contribution from molecular scattering, so about half of the signal originated from aerosols.
Scattering by gas molecules depends mainly on the number concentration of molecules in a unit volume, which primarily depends on temperature and pressure and thus may vary. The molecular contribution to β at 910-nm wavelength was approximated by Eq. (2.135) in Measures (1984) with 10-min temporal resolution using the measured temperature and pressure.
4. Results and discussion
The purpose of the study was to establish the capabilities of a ceilometer-type lidar as a quantitative aerosol instrument. To this end, optical modeling based on light scattering theories was used to convert the in situ aerosol measurements into simulated lidar measurements that could be compared with the actual ceilometer measurements. The calculations and the lidar measurements were carried out at 910-nm wavelength, but the results can be applied to other wavelengths using the scale invariance rule of electromagnetic scattering (Mishchenko et al. 2002, p. 147).
Unfortunately, the in situ measurements do not unambiguously specify the physical aerosol properties necessary to predict lidar backscattering exactly. For one thing, the measured size distribution is for dry, not ambient, aerosol. Also the form of the size distribution depends on the values of shape factor κ and density ρ used. Further, direct information about the composition or shape of the aerosol particles is not available. Some allowance thus needs to be made when comparing the simulations and lidar measurements; more importantly, the impact of shape and composition on backscattering needs to be established.
Three days were selected for the studies (20 June, 21 June, and 20 August 2005, denoted as days 1, 2, and 3, respectively). Each day represented typical meteorological summer conditions in Helsinki, but the particle size distributions varied in a different manner. All of the days were rainless and clear, with the exception of a few upper clouds. In addition, a significant amount of large pollen particles were not present that could affect the lidar measurements while not registering in the in situ instruments. Moderate winds during the night and early morning on day 1 indicated that some vertical mixing took place during most of the day. Relative humidity stayed between 45% and 55% and did not have a clear diurnal cycle. On days 2 and 3, on the other hand, significant vertical mixing took place only during the daytime. The relative humidity had a clear diurnal cycle, varying from 35% to 75% on day 2 and from 48% to 90% on day 3.
a. Effect of particle size
According to Hussein et al. (2004), ultrafine particle number concentrations contribute more than 90% of the total number concentrations in the urban atmosphere in the Helsinki area. The dry-particle size distribution was established by combining measurements from DMPS and APS instruments, as described in section 2b. Different values of aerodynamic shape factor κ and density ρ were tested to establish their impact on the particle size distribution (Fig. 1) and the theoretically calculated values of β, σ, and R. The values of 1.0 or 1.4 for κ, 1.0 or 1.8 g cm−3 for ρ, and all their combinations were used to get an estimate for the uncertainty related to these parameters. As mentioned in section 2b, the default values in this study were κ = 1.4 and ρ = 1.8 cm−3. Results showed that the concentrations in the optically active diameter range from 0.1 to 3 μm depended considerably on the values of κ and ρ. The values of theoretically calculated β decreased about 70% if the value of κ was decreased from 1.4 to 1.0 while ρ was kept at 1.8 g cm−3. In contrast, when ρ was decreased to 1.0 g cm−3 while κ = 1.4, β increased about 250%. Interestingly, the difference in the values of β was negligible if the combination of ρ = 1.0 g cm−3 and κ = 1.0 was used.
Effective radii and variances were calculated to characterize the variation of the size distribution during the example days. The diurnal behavior of reff and υeff was similar for all days. Generally reff varied around 0.3 μm during nighttime and around 0.5 μm during daytime. However, as Fig. 2 shows, low values of reff (∼0.3 μm) were also observed during the afternoon. This was connected to a rapid increase in ultrafine particle concentrations during the rush hour, which decreased the relative concentration of accumulation- and coarse-mode particles in the size distribution and reduced reff. The diurnal variation of υeff was opposite to the variation of reff, the maximum values being observed during nighttime and minimum values during daytime. This is at least partially connected to the definition of υeff [Eq. (8)], where reff is in the denominator.
In addition to the size distribution, the backscattering efficiency depends also on the scattering efficiency Qsca and the phase function p(π, r). Both of these parameters depend on the particle size, as illustrated in Fig. 3. For particles much smaller than the wavelength, the Rayleigh theory predicts Qsca to be proportional to the fourth power of the radius. In Fig. 3, this is seen as a quasi-linear increase for submicron particles; the maximum value of about 4 is reached around particle diameter 1.4 μm, after which it asymptotically approaches the value of two, the value predicted at the large-particle limit. The behavior of p(π, r) for spherical and nonspherical particles was very similar for ultrafine and accumulation-mode particles, but clearly different in the coarse mode, where largest differences were obtained. The value of 1.5 predicted from the Rayleigh theory was obtained for ultrafine particles. It is noted that, for Qsca, the Rayleigh theory seemed to be valid for larger particles than for p(π, r). Similar behavior has been noticed, for example, between the extinction efficiency and the asymmetry parameter (Mishchenko et al. 2002, p. 312).
The particle sizes that contribute most to backscattering were centered in two modes: the first around 0.34 μm and the second around 2.0 μm in diameter. As Fig. 1 shows, the particle number concentration in the first mode was nearly four orders of magnitude larger than in the second mode. Further, the results show that half of the total backscattering originated from particles larger than about 0.5 μm in diameter and only 1% originated from particles larger than about 8 μm (Fig. 4). About 0.5% of the total backscattering originated from ultrafine particles. These results also illustrate the difficulty of a near-infrared optical instrument to detect very small particles.
b. Effect of particle composition
The effect of composition on backscattering was estimated by varying the complex refractive index m, where the imaginary part refers to the particle absorption. The values of β, σ, and R for different Re(m) and Im(m) were calculated using the observed dry-aerosol size distributions. These calculations were carried out assuming particles to be spherical, because calculations for nonspherical model particles are much slower and more complicated, and shape is expected to have only a secondary impact to the dependence of scattering on composition (e.g., Nousiainen 2007) when size-distribution-integrated values are considered.
The main components of PM2.5 particles in the Helsinki area are the particulate organic matter and secondary inorganic aerosols (Sillanpää et al. 2006). The fraction of sea salt, elemental carbon, and other minor components are found to be significantly smaller. The main components for PM10 particles are soil-derived particles and particulate organic matter. Here, the values of 1.5 and 1.3 were used for Re(m) and denoted as mhi and mlo, respectively, with Im(m) fixed to 0.0. The value of mlo is close to the refractive index of liquid water, whereas mhi is representative of many dust aerosol types [both soil and volcanic (e.g., d’Almeida et al. 1991)] as well as sea salt and ammonium sulfate. Values close to mhi have also been used, for example, for biomass burning aerosols (e.g., Ross et al. 1998) and particulate organic matter (e.g., Mallet et al. 2003). The effect of absorption, on the other hand, was studied with Im(m) values of 0.000, 0.001, and 0.005, which are denoted as abs0, abslo, and abshi, respectively, whereas Re(m) was fixed to 1.5. The value of abshi has been used, for example, for particulate organic matter, but for nitrates and sulfates Im(m) is significantly smaller; Gosse et al. (1997) measured the values of Im(m) between 3.49 × 10−7 and 4.94 × 10−7 at 910-nm wavelength for sulfates and nitrates. The refractive index for soot differs from these values and can be as high as Re(m) = 1.75 and Im(m) = 0.435 (e.g., Liu and Mishchenko 2007). It is noted that mhi and mlo, as well as abs0 and abshi, can be considered as expected extreme values for the effective refractive index for the urban aerosol mixture.
The comparison of the results for mhi and mlo, as well as for abs0 and abshi, gave an idea of how much the composition could affect lidar backscattering. The estimate for the composition-related error or uncertainty thus obtained overestimated the uncertainty faced in practice, as one can always simply choose a refractive index between the known extremes. Although backscattering is not a monotonic (let alone linear) function of the refractive index, larger values of Re(m) generally lead to stronger backscattering, especially when size-distribution-integrated quantities are considered (e.g., Mishchenko et al. 2002, p. 268). Thus, considering only the extreme cases in our study to estimate the error/uncertainty resulting from composition is appropriate.
The results showed that the refractive index affected backscattering substantially. Both β and σ increased significantly (while R decreased) when composition was changed from mlo to mhi. This is clearly seen in the right panel of Fig. 5, where the impact to these quantities resulting from composition is expressed as mhi/mlo ratios. Although the size distribution varied substantially during the day, the mhi/mlo ratios for β and σ were systematically well over unity: about 3.0 for β and 1.9 for σ. For R, the mhi/mlo ratio was typically about 0.6.
The results also showed that the effect of composition on β and σ depended to some degree on the value of reff (i.e., the size distribution). A comparison of Figs. 4 and 5 reveals that a small mhi/mlo ratio for σ seemed to be related to large reff, whereas, for β, the dependence seemed to be opposite and somewhat weaker. This behavior for σ is understood when the impact of composition on Cext was plotted as a function of size (not shown). For submicron particles, Cext was systematically and considerably smaller for mlo, whereas, for supermicron particles, it was nonsystematically either smaller or larger. In fact, at the large-particle limit, Cext was exactly 2 times the cross-sectional surface area and did not depend on the composition at all. Consequently, the change of composition from mlo to mhi had larger impact on σ when reff was small.
Finally, the overall effect of absorption was found to be minor. The ratios abshi/abs0 for β and σ were about 0.89 and 1.04, respectively; for abslo/abs0, the corresponding values were 0.97 and 1.01. The most probable reason why these values were over unity for σ is that the absolute value of m is larger for absorbing particles, providing stronger interaction between matter and radiation. For R, the abshi/abs0 and abslo/abs0 ratios were, on average, 1.16 and 1.05, respectively.
c. Effect of particle shape
The effect of particle shape was studied by comparing the results for shape distributions of spheroids (see section 2b) with those for spheres. We used three different shape models with a fixed m: spheres, spheroids with an n = 0 shape distribution (so-called equiprobable shape distribution), and spheroids with an n = 3 shape distribution (see section 2b). These cases are denoted as sph, shan=0, and shan=3, respectively.
Changing the particle shape from spherical to nonspherical decreased backscattering. Similar results have been obtained (e.g., Mishchenko et al. 1997) by using the shape distribution n = 0. The decrease was considerable but not quite as large as when changing the composition from mhi to mlo. The values of the ratio sph/shan=3 for β and σ were, on average, 1.71 and 1.24, respectively (Fig. 5); for sph/shan=0, the corresponding values were 1.35 and 1.14. For the lidar ratio R, this resulted in sph/shan=3 and sph/shan=0 ratios of 0.70 and 0.84, respectively. As is seen for the n = 3 shape distribution, the results deviate the most from the results for spherical particles, whereas the equiprobable shape distribution is situated between the two extremes.
The larger the values of the sph/shan=3 ratio for β were, the larger the reff. This is consistent with the well-known fact (e.g., Mugnai and Wiscombe 1980) that the impact of nonspherical particle shape on scattering becomes stronger with increasing particle size, which can also be seen in Fig. 2. However, this dependence was not as pronounced for σ as it was for β, as Fig. 5 shows.
d. Comparison between simulated and measured backscattering
The purpose of this work was to investigate whether a simple off-the-shelf ceilometer-type lidar can provide useful information about ambient aerosols. To assess this, the so-called observed β(βobs) derived from the lidar measurement by Eq. (12) were compared with simulated β(βsim) based on in situ size distributions and optical modeling. The comparison is meaningful only if the simulated and measured values of β can be considered representative. Thus, the values measured at the lidar’s third range gate were selected for the comparisons (section 3c).
For each day, the measured and simulated values of β were compared at 10-min temporal resolution. The comparisons showed that the largest differences between βsim and βobs were obtained when particles were assumed spherical with the refractive index mhi (Fig. 6), leading to βsim values mainly 3.6 times higher than βobs. Particle absorption did not change the differences between βsim and βobs notably, but changing the particle shape from spherical to nonspherical decreased the differences considerably. For shan=3, βsim was about 2.0 times higher than βobs. The best agreement was obtained with spherical model particles with refractive index mlo, when βsim was mainly 1.2 times higher than βobs. This, however, represents a rather unrealistic refractive index for dry aerosol, the measured size distribution represented. The differences between βsim and βobs were thus generally considered large but remained nearly constant most of the time. Most likely, by simulating β with nonspherical particles (shan=3) and a refractive index somewhat lower than mhi, better agreement with measurements could be obtained.
The possible factors affecting the values of βsim and βobs must be carefully considered when assessing whether the lidar measurements are quantitatively correct or not. As the results showed, most of the time, the differences between βsim and βobs remained nearly constant. This indicates that there was most likely some systematic error in βsim, βobs, or both. On the other hand, the occasional large deviations between βsim and βobs were most likely caused by some coincidental event. Systematically erroneous values of βsim could be explained by the uncertainty related to particle composition and shape, the parameters κ and ρ used in the merging of the in situ size distributions, an inaccurate sampling efficiency by DMPS and APS instruments, and/or hygroscopic growth of particles that may make the measured dry-aerosol size distributions inadequate for predicting the lidar signal. The values of βobs could also contain some systematic error resulting from calibration and/or the overlap function in the lidar measurement algorithm. On the other hand, occasional large differences between βsim and βobs are most likely explained by local aerosol sources or unfavorable meteorological conditions.
As shown in section 4a, the values of βsim can shift quite considerably when using different combinations of κ and ρ. For example, if the values of κ and ρ were set to 1.0 and 1.8 g cm−3, respectively, the simulated values decrease so that they would still be larger than the measured βobs for mhi but lower for mlo and shan=3. Thus, with this combination of κ and ρ, the agreement between the simulated and measured values would generally improve. On the other hand, with κ = 1.0 and ρ = 1.4 g cm−3, the simulated values of β would be typically 3–10 times larger than the measured values, depending on the simulation case considered.
If the lidar’s third range gate used here is affected by an erroneous overlap correction, it would lead to a systematic error of βobs. This would also partly explain why βobs is consistently lower than βsim. As Fig. 7 reveals, there are good reasons to expect that there are unresolved overlap issues in the measured profile. For example, the calibrated lidar-measured profile seemed to have a persistent minimum at the sixth range gate, even though the profile itself varied during the day. Compared to the values to other nearby range gates, the minimum value at the sixth range gate is unrealistic and most likely an artifact from the overlap function. In fact, the sixth range gate minimum was observed even in rain echoes. This also raises a question how realistic the values in other range gates adjacent to the sixth range gate are. Regarding Fig. 7, the third and ninth range gate values could be the nearest, more or less, reliable data that could be used, and even they could be affected. In fact, the profile would look more reasonable if values were interpolated from the ninth to the third range gate. Because the lidar’s measurement algorithm is treated as a black box, we can only make assumptions on what problems the algorithm might have.
The occasional strong disagreements between βsim and βobs were observed in all cases, regardless of the parameters used in the simulations. These exeptions are usually seen at night and during morning hours, as in Fig. 6, when βsim increased but βobs did not. During these episodes, βsim could be 10–19 times higher than βobs, depending on which simulation case is considered. This is most likely a representativeness issue: even though the measurement equipment used in this study were located near each other, the measurement volumes were not identical and thus the measured aerosol size distribution was not necessarily representative of the aerosol within the measurement volume of the lidar. The peaks that can be seen in βsim but not in βobs are most likely related to some local particle source near the in situ measurement site that, because of insufficient vertical mixing or unsuitable wind direction, was not seen by the lidar.
5. Summary and conclusions
The aim of this work was to investigate whether a commercial ceilometer-type lidar could be used as a quantitative aerosol measurement instrument. The advantage of a ceilometer-type lidar is its ability to detect backscattered signal at much closer range than most of the more sophisticated research lidars. For this purpose, ceilometer measurements were compared with theoretically modeled backscattering based on real, measured dry-particle size distributions. In addition to volume backscattering coefficient β, theoretical calculations provided information about the volume extinction coefficient σ, lidar ratio R, and their dependence on the physical properties of aerosol particles. To obtain an estimate for particles’ different physical characteristics affecting the lidar signal, particle composition [Re(m), Im(m)] and shape were varied in the calculations. Three example days from summer 2005 were selected for this study, each representing typical meteorological summer conditions in our urban measurement site.
For simulations, two different cases for Re(m), and three cases for Im(m) were considered. The effect of particle shape was studied using spheres and two different shape distributions for spheroids. For simplicity, the effect of particle composition was only studied for spheres. Likewise, when the effect of particle shape was studied, the particle composition was fixed. Because we did not have accurate information about the real particle composition or shape, the main objective was to establish reasonable a range of uncertainty for the simulated quantities resulting from the uncertainty in the actual particle composition and shape, rather than attempting to find a perfect match with modeled and measured values.
According to model calculations, the uncertainty resulting from the unknown aerosol composition at our measurement site can affect the simulated volume backscattering coefficient βsim by a factor of 3. The effect of particle composition was dominated by the variation of Re(m), whereas particle absorption [Im(m)] had only a minor contribution to the uncertainty. Likewise, the unknown particle shape can affect βsim by a factor of 2. The comparison of βsim and βobs, derived from the lidar measurement, revealed that βsim was systematically larger than βobs, but the difference remained, most of the time, nearly constant and within the uncertainty ranges associated with βsim. The differences between βsim and βobs were larger for βsim based on spherical aerosol particles than on nonspherical aerosol particles. Likewise, the differences were smaller with Re(m) = 1.3 (mlo) and Re(m) = 1.5 (mhi). However, such a low Re(m) is unrealistic with a dry-particle size distribution. On the other hand, a somewhat-larger refractive index combined with nonspherical particle shapes would be likely to give good fits with the measurements.
It should noted that the values βsim used in the comparison were based on a size distribution that has been merged from two parts using a fixed shape factor κ (1.4) and density ρ (1.8 g cm−3). Sensitivity tests showed that the values of κ and ρ can significantly affect the computed βsim values and could play a part in the observed differences with βobs. For example, decreasing κ from 1.4 to 1.0 decreased βsim about 70%, improving the agreement with βobs regardless of the assumptions concerning particle composition or shape. If uncertainties of this magnitude are involved in the values of these parameters, then the uncertainty in the size distribution is actually the largest error source in the investigation. Further, the size distribution could also be affected by an inaccurate sampling efficiency for the DMPS and APS instruments.
The values of βobs might also have some systematic errors. Because the ceilometer-type lidar used here is a so-called off-the-shelf device, it has been used “as is,” with the measurement algorithm considered as a black box. Consequently, certain factors, such as the overlap function, cannot be studied in detail. An erroneous overlap correction could partly explain why βobs were consistently smaller than βsim.
For all the simulation cases, the largest differences between βobs and βsim were observed during nighttime and early morning hours, when βsim increased rapidly and βobs did not. Obviously, the comparison between βsim and βobs is meaningful only if the in situ–measured particle size distribution can be considered representative of the size distribution in the lidar measurement volume. The unrepresentativeness of the size distribution is the most probable single reason causing the occasional large disagreement between βsim and βobs. Other possible factors affecting the difference are hygroscopic growth of particles, local aerosol sources, wind direction, and vertical mixing.
To conclude, the results obtained in this study are promising. Although perfect agreement between measured and simulated backscattering was not obtained, the differences were within the uncertainties involved. This indicates that the absolute accuracy of this instrument might well be sufficient for quantitative aerosol measurements in some applications. However, for that purpose, an accurate calibration of the overlap effect would be important. Several issues were not investigated in this study, such as performance of the lidar under different meteorological conditions, the influence of relative humidity on the lidar measurements, and whether we can obtain statistical aerosol information from ceilometer measurements. These issues will be addressed in a follow-up paper.
Acknowledgments
The research was partially funded by the Maj and Tor Nessling foundation (AMS), the Lidar in aerosol research project of the Finnish Funding Agency for Technology and Innovation TEKES (AMS, TN), and by the Academy of Finland (TN; Contract 212979). Vaisala Oyj is acknowledged for providing the lidar for the project. Janne Räsänen, Hannu Talvitie, and Reijo Roininen from Vaisala Oyj and Erkki Siivola from the University of Helsinki are acknowledged for the installation of the CL31 ceilometer and technical support. Risto Hillamo from the Finnish Meteorological Institute is acknowledged for providing the APS data used in this study. We are also grateful to Markku Kulmala for strongly advocating the initiation of the lidar research in the department.
REFERENCES
Aalto, P., and Coauthors, 2001: Physical characterization of aerosol particles during nucleation events. Tellus, 53B , 344–358.
Ansmann, A., and Müller D. , 2005: Lidar and atmospheric aerosol particles. Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, C. Weitkamp, Ed., Springer, 105–141.
Brunekreef, B., and Holgate S. , 2002: Air pollution and health. Lancet, 360 , 1233–1242.
Bundke, U., and Coauthors, 2002: Aerosol optical properties during the Lindenberg Aerosol Characterization Experiment (LACE 98). J. Geophys. Res., 107 , 8123. doi:10.1029/2000JD000188.
Cabada, J. C., Khlystov A. , Wittig A. , Pilinis C. , and Pandis S. N. , 2004: Light scattering by fine particles during the Pittsburgh Air Quality Study: Measurements and modeling. J. Geophys. Res., 109 , D16S03. doi:10.1029/2003JD004155.
d’Almeida, G. A., Koepke P. , and Shettle E. P. , 1991: Atmospheric Aerosols: Global Climatology and Radiative Characteristics. Deepak, 561 pp.
Draine, B. T., and Flatau P. J. , 1994: Discrete-dipole approximation for scattering calculations. J. Opt. Soc. Amer., 11 , 1491–1499.
Emeis, S., Münkel C. , Vogt S. , Müller W. J. , and Schäfer K. , 2004: Atmospheric boundary-layer structure from simultaneous SODAR, RASS, and ceilometer measurements. Atmos. Environ., 38 , 273–286. doi:10.1016/j.atmosenv.2003.09.054.
Gimmestad, G., 2005: Differential-absorption lidar for ozone and industrial emissions. Lidar: Range-Resolved Optical Remote Sensing of the Atmosphere, C. Weitkamp, Ed., Springer, 187–212.
Gobbi, G. P., Barnaba F. , Blumthaler M. , Labow G. , and Herman J. R. , 2002: Observed effects of particles nonsphericity on the retrieval of marine and desert aerosol optical depth by lidar. Atmos. Res., 61 , 1–14.
Gosse, S. F., Wang M. , Labrie D. , and Chylek P. , 1997: Imaginary part of the refractive index of sulfates and nitrates in the 0.7-2.6-μm spectral region. Appl. Opt., 36 , 3622–3634.
Hansen, J. E., and Travis L. D. , 1974: Light scattering in planetary atmospheres. Space Sci. Rev., 16 , 527–610.
Haywood, J., and Boucher O. , 2000: Estimates of the direct and indirect radiative forcing due to tropospheric aerosols: A review. Rev. Geophys., 38 , 513–543.
Hussein, T., Puustinen A. , Aalto P. P. , Mäkelä J. M. , Hämeri K. , and Kulmala M. , 2004: Urban aerosol number size distributions. Atmos. Chem. Phys., 4 , 391–411.
Jokinen, V., and Mäkelä J. M. , 1997: Closed loop arrangement with critical orifice for DMA sheath/excess flow system. J. Aerosol Sci., 28 , 643–648.
Lagrosas, N., Kuze H. , Takeuchi N. , Fukagawa S. , Bagtasa G. , Yoshii Y. , Naito S. , and Yabuki M. , 2005: Correlation study between suspended particulate matter and portable automated lidar data. J. Aerosol Sci., 36 , 439–454. doi:10.1016/j.jaerosci.2004.10.007.
Liu, L., and Mishchenko M. I. , 2007: Scattering and radiative properties of complex soot and soot containing aggregate particles. J. Quant. Spectrosc. Radiat. Transfer, 106 , 262–273.
Mallet, M., Roger J. C. , Despiau S. , Dubovik O. , and Putaud J. P. , 2003: Microphysical and optical properties of aerosol particles in urban zone during ESCOMPTE. Atmos. Res., 69 , 73–97.
Measures, R. M., 1984: Laser Remote Sensing: Fundamentals and Applications. Wiley, 510 pp.
Mishchenko, M. I., and Travis L. D. , 1998: Capabilities and limitations of a current Fortran implementation of the T-matrix method for randomly oriented rotationally symmetric scatterers. J. Quant. Spectrosc. Radiat. Transfer, 60 , 309–324.
Mishchenko, M. I., Travis L. D. , Kahn R. A. , and West R. A. , 1997: Modeling phase functions for dustlike tropospheric aerosols using a shape mixture of randomly oriented polydisperse spheroids. J. Geophys. Res., 102 , (D14). 16831–16847.
Mishchenko, M. I., Travis L. D. , and Lacis A. A. , 2002: Scattering, Absorption, and Emission of Light by Small Particles. Cambridge University Press, 445 pp.
Moreno, F., Vilaplana R. , Muñoz O. , Molina A. , and Guirado D. , 2006: The scattering matrix for size distributions of irregular particles: An application to an olivine sample. J. Quant. Spectrosc. Radiat. Transfer, 100 , 277–287. doi:10.1016/j.jqsrt.2005.11.069.
Mugnai, A., and Wiscombe W. J. , 1980: Scattering of radiation by moderately nonspherical particles. J. Atmos. Sci., 37 , 1291–1307.
Münkel, C., Eresmaa N. , Räsänen J. , and Karppinen A. , 2007: Retrieval of mixing height and dust concentration with lidar ceilometer. Bound.-Layer Meteor., 124 , 117–128. doi:10.1007/s10546-006-9103-3.
Muñoz, O., Volten H. J. , de Haan F. , Vassen W. , and Hovenier J. W. , 2001: Experimental determination of scattering matrices of randomly oriented fly ash and clay particles at 442 and 633 nm. J. Geophys. Res., 106 , 22833–22844.
Nousiainen, T., 2007: Impact of particle shape on refractive-index dependence of scattering in resonance domain. J. Quant. Spectrosc. Radiat. Transfer, 108 , 464–473.
Nousiainen, T., Kahnert M. , and Veihelmann B. , 2006: Light scattering modeling of small feldspar aerosol particles using polyhedral prisms and spheroids. J. Quant. Spectrosc. Radiat. Transfer, 101 , 471–487. doi:10.1016/j.jqsrt.2006.02.038.
Pakkanen, T. A., and Coauthors, 2001: Sources and chemical composition of atmospheric fine and coarse particles in the Helsinki area. Atmos. Environ., 35 , 5381–5391.
Ross, J. L., Hobbs P. V. , and Holben B. , 1998: Radiative characteristics of regional hazes dominated by smoke from biomass burning in Brazil: Closure tests and direct radiative forcing. J. Geophys. Res., 103 , (D24). 31925–31941.
Sillanpää, M., and Coauthors, 2006: Chemical composition and mass closure of particulate matter at six urban sites in Europe. Atmos. Environ., 40 , (Suppl.). 212–223.
Stolzenburg, M., and McMurry P. , 1991: An ultrafine aerosol condensation nucleus counter. Aerosol Sci. Technol., 14 , 48–65.
Veihelmann, B., Nousiainen T. , Kahnert M. , and van der Zande W. J. , 2006: Light scattering by small feldspar particles simulated using the Gaussian random sphere geometry. J. Quant. Spectrosc. Radiat. Transfer, 100 , 393–405. doi:10.1016/j.jqsrt.2005.11.053.
Volten, H., Muñoz O. , Rol E. , de Haan J. F. , Vassen W. , Hovenier J. W. , Muinonen K. , and Nousiainen T. , 2001: Scattering matrices of mineral aerosol particles at 441.6 nm and 632.8 nm. J. Geophys. Res., 106 , (D15). 17375–17401.
Von Klot, S., and Coauthors, 2005: Ambient air pollution is associated with increased risk of hospital cardiac readmission of myocardial infarction survivors in five European cities. Circulation, 112 , 3073–3079.
Waterman, P. C., 1971: Symmetry, unitarity, and geometry in electromagnetic scattering. Phys. Rev., 3D , 825–839.
Winklmayr, W., Reischl G. , Lindner A. , and Berner A. , 1991: A new electromobility spectrometer for the measurement of aerosol size distributions in the size range from 1 to 1000 nm. J. Aerosol Sci., 22 , 289–296.
Combined in situ aerosol number size distribution in the Kumpula campus on day 3, averaged over 1450–1500 UTC with different particle densities ρ and shape factors κ.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
The diurnal variation of the in situ effective radius reff (black) and effective variance υeff (gray) during day 2. For the size distribution, values of κ = 1.4 and ρ = 1.8 g cm−3 were used.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
Calculated phase functions p(π, r) and scattering efficiencies Qsca as a function of particle diameter for spherical and nonspherical particles for m = 1.5 + 0.0i. Each value represents a size-bin average. The values of p(π, r) are denoted with triangles (spherical) and asterisks (nonspherical), whereas Qsca are denoted with black dots (spherical) and white squares (nonspherical).
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
The cumulative fraction of theoretically modeled β as a function of particle size for day 3 from 1450 to 1500 UTC. The cumulative function is calculated using size distribution represented in Fig. 1 with κ = 1.4 and ρ = 1.8 g cm−3. The particles were assumed to be spherical with m = 1.5 + 0.0i.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
The effect of (right) particle composition and (left) shape on theoretically calculated β, σ, and R, during day 2, represented as mhi/mlo and sph/shan=3 ratios.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
Comparison of measured and theoretically calculated β for day 3 represented as a ratio of modeled-to-observed β. Results are shown for mhi, mlo, abshi, and shan=3 cases. The size distribution is based on κ = 1.4 and ρ = 1.8 g cm−3.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
Calibrated example lidar profiles for day 3. Each profile represents a 1-h average. Backscattering from the air molecules has not been removed from the profiles.
Citation: Journal of Atmospheric and Oceanic Technology 26, 11; 10.1175/2009JTECHA1252.1
