3.1. Optical Properties
 Figure 1 shows the annual and seasonal variations of mean AOD at 500 nm and α at 440–675 nm with their corresponding standard deviations for the period of 2006–2009. The annual mean AOD for each year was 0.79 ± 0.51, 0.78 ± 0.48, 0.72 ± 0.40 and 0.63 ± 0.34, respectively. Standard deviations are greater than 50% of the mean, indicating that AODs varied greatly throughout the years. Figure 1 also reveals a gradual reduction in annual mean AOD of about −3.4%, −8.4% and −9.3% per year from 2006 to 2009 during the spring, autumn and winter seasons, respectively. The decrease in AOD in autumn is statistically significant with a confidence level of 0.1 (P = 0.1), but is not in other seasons. A similar analysis for the summer season was not done because of a lack of AOD retrievals during the summer of 2009.
Figure 1. Seasonal and annual mean (a) AOD (500 nm) and (b) α (440–675 nm) with corresponding standard deviations. Data are from 2006 to 2009 at Taihu.
Download figure to PowerPoint
 In terms of seasonal variation, maxima in overall mean AOD occurred in spring and summer (0.76 ± 0.40 and 0.81 ± 0.48, respectively). In spring, dust aerosols are transported out of northwest China and can increase aerosol loading over Taihu [Tsai et al., 2008; J. Liu et al., 2011]. This is illustrated in Figure 2, which shows the aerosol particle size distribution for each season. Relatively high concentrations of coarse-mode particles are present during spring (Figure 2a). In summer, high values of AOD occur because of higher relative humidity during this season inducing deliquescence and particle growth and an increase in fine-mode anthropogenic aerosols (seeFigure 2b) due to the build-up of local pollution resulting from the presence of a persistent and stagnant synoptic meteorological system over the Asian continent [Kim et al., 2007]. During autumn and winter, relatively strong winds to the lower Yangtze River Delta region, diluting pollutants and lowers the aerosol loading [Pan et al., 2010]. Frequency distributions of AOD shown in Figure 3a, illustrate the broad range of values observed from 2006 to 2009, and the high aerosol loading prevalent year-round over the region. Accumulated frequencies of AOD in the range less than 0.5, 0.5 to 1.0 and greater than 1.0 are about 34.9%, 45.2% and 19.9%, respectively. Only 2.5% of all values of AOD are less than 0.2.
Figure 2. Seasonal mean aerosol particle size distributions for (a) spring (March–May), (b) summer (June–August), (c) autumn (September–November) and (d) winter (December–February). Data are from 2006 to 2009 at Taihu.
Download figure to PowerPoint
Figure 3. Histograms of (a) AOD (500 nm), (b) α (440–675 nm), (c) SSA (675 nm) and (d) ASY (675 nm). Data are from 2006 to 2009 at Taihu.
Download figure to PowerPoint
 The Ångstrom exponent, which varies with particle size, is derived from AOD retrievals made at 440 nm and 675 nm. Figure 1b shows that the annual mean α was 1.14 ± 0.30, 1.23 ± 0.25, 1.24 ± 0.27 and 1.19 ± 0.27 from 2006 to 2009, respectively. Seasonal mean α during the study period was 1.09 ± 0.30, 1.26 ± 0.27, 1.31 ± 0.20, and 1.22 ± 0.23 for spring, summer, autumn and winter, respectively. The magnitude of α was smallest in spring and largest in autumn. This variation may reflect changes in the origin of aerosol particles and transport routes [Léon et al., 2009], although these values fall well within the range of the standard deviations. An influx of dust particles from the northern/northwest regions of China, carried in by winds associated with the Asian monsoon and continental anticyclones [Tsai et al., 2008; J. Liu et al., 2011], is the likely cause for the relatively low springtime minimum. It is worth noting that aerosol particles can affect the Asian monsoon circulation as well [Niu et al., 2010]. Figure 3b shows that α ranges from 0.1 to 1.9 with a peak in the distribution between 1.2 and 1.5. This illustrates that the aerosol particle type in this region is highly variable. The accumulated frequencies of α less than 1.0 and greater than 1.0 are about 20.6%, and 79.4%, respectively. Only 1.6% of all values of α was less than 0.5 and occurred in spring when dust events occasionally impact the site.
 Table 1 summarizes annual and seasonal variation in mean SSA and ASY at 675 nm, as well as annual minimum and maximum values of each quantity. From 2006 to 2009, annual mean values of SSA were 0.921 ± 0.032, 0.906 ± 0.038, 0.907 ± 0.033 and 0.922 ± 0.024, respectively. The seasonal mean SSA over the period of 2006–2009 was 0.922 ± 0.028, 0.925 ± 0.042, 0.917 ± 0.031 and 0.892 ± 0.034 in spring, summer, autumn and winter, respectively. Low values of SSA in winter are mainly due to the dominance of absorbing urban aerosol particles prevalent during the heating period. Figure 3c shows that the majority of SSA values (84%) fall in the range of 0.85 to 0.95. The wide range of SSA values (0.726–0.992 in terms of instantaneous values) illustrates there were quite different aerosol particle types and optical characteristics at this location.
Table 1. Seasonal Mean, Minimum, Maximum and Annual Mean SSA and ASY at 675 nm Retrieved From AERONET Measurements Taken at Taihu From 2006 to 2009a
|SSA (675 nm)|
|2006||0.934 ± 0.021||0.939 ± 0.034||0.936 ± 0.027||0.896 ± 0.027||0.838||0.982||0.921 ± 0.032|
|2007||0.921 ± 0.025||0.908 ± 0.047||0.911 ± 0.031||0.888 ± 0.042||0.726||0.992||0.906 ± 0.038|
|2008||0.916 ± 0.033||0.947 ± 0.019||0.900 ± 0.031||0.893 ± 0.027||0.804||0.978||0.907 ± 0.033|
|2009||0.918 ± 0.027||n/a||0.931 ± 0.017||0.915 ± 0.026||0.810||0.975||0.922 ± 0.024|
|all||0.922 ± 0.028||0.925 ± 0.042||0.917 ± 0.031||0.892 ± 0.034|| || ||0.912 ± 0.035|
|ASY (675 nm)|
|2006||0.664 ± 0.022||0.664 ± 0.033||0.645 ± 0.026||0.653 ± 0.034||0.584||0.744||0.658 ± 0.029|
|2007||0.654 ± 0.036||0.697 ± 0.041||0.650 ± 0.037||0.652 ± 0.036||0.565||0.775||0.656 ± 0.039|
|2008||0.652 ± 0.039||0.670 ± 0.046||0.634 ± 0.045||0.648 ± 0.032||0.555||0.755||0.649 ± 0.039|
|2009||0.651 ± 0.030||n/a||0.665 ± 0.023||0.679 ± 0.023||0.594||0.721||0.657 ± 0.028|
|all||0.655 ± 0.034||0.681 ± 0.042||0.649 ± 0.036||0.652 ± 0.034|| || ||0.655 ± 0.036|
 Magnitudes of ASY were relatively constant from 2006 to 2009 (see Table 1). Annual mean values were 0.658 ± 0.029, 0.656 ± 0.039, 0.649 ± 0.039 and 0.657 ± 0.028 in 2006, 2007, 2008 and 2009, respectively. Seasonal mean values of ASY for the entire period were 0.655 ± 0.034, 0.681 ± 0.042, 0.649 ± 0.036, 0.652 ± 0.034 in spring, summer, autumn and winter, respectively. Nearly 84% of all values of ASY fell in the range of 0.6–0.7 (Figure 3d). Hygroscopic growth may explain why SSA and ASY are largest during the summer season. For a fixed composition, hygroscopic growth under humid conditions can lead to an increase in particle size (therefore increasing ASY) and cause enhanced forward scattering (therefore increasing SSA) [Jeong et al., 2007; Xia et al., 2007].
 Figure 4 presents the seasonal mean values of spectral aerosol SSA and ASY at 440, 675, 870 and 1020 nm with corresponding standard deviations at Tahihu during 2006–2009. The SSA showed a relatively weak dependence on wavelength with a slight increase from 440 to 675 nm and almost invariable over 675–1020 nm. The annual mean SSAs at the four wavelengths are 0.892 ± 0.038, 0.912 ± 0.035, 0.909 ± 0.041 and 0.908 ± 0.046, respectively. The seasonal mean ASY showed a decreasing trend with wavelength in summer, autumn and winter. But in spring, ASY decreased over 440–870 nm and slightly increased over 870–1020 nm, which is related to the dust activity in spring [X. Yu et al., 2009]. The annual mean ASYs at the four wavelengths are 0.720 ± 0.026, 0.655 ± 0.036, 0.629 ± 0.040 and 0.626 ± 0.043, respectively.
Figure 4. Seasonal means of the spectral (top) SSA and (bottom) ASY at 440, 675, 870 and 1020 nm with corresponding standard deviations. Data are from 2006 to 2009 at Taihu.
Download figure to PowerPoint
3.3. Role of Dust Aerosols on Aerosol Vertical Distribution
 By virtue of the polarization sensitivity of the MPL deployed at Taihu, the role of non-spherical particles, and in particular, dust, on the seasonal vertical distribution of aerosols is investigated. Polarized lidar measurements can be used to determine the characteristic parameter “d” of the Mueller matrix describing the polarizing effects from randomly oriented particles in the atmosphere [Flynn et al., 2007]. Specifically, the parameter “d” identifies the degree to which the scattering event modifies the polarization vector. In the absence of multiple scattering, “d” is near zero for spherical particles and greater than zero for non-spherical particles. The parameter “d” is explicitly related to the commonly described lidar “linear depolarization ratio”δlinear, a potential tool for differentiating different aerosol types [Sakai et al., 2000; Rajeev et al., 2010].
 The following criteria are used to identify dust with a high probability of mixing with local emissions: (1) δlinear must be larger than 0.1 [Cavalieri et al., 2010] in at least five consecutive vertical bins for most of the day; (2) at least one of the five-day back trajectories arriving at the 0.5-km and 2.5-km levels over Taihu pointed to the dust source region or passes over it before reaching the site; (3) no precipitation is detected during the time it takes for the air mass to arrive at the site; and (4) dust activities are detected near the site on the day in question or in the dust region before the air mass arrives at the site, based upon the aerosol index from NASA's Ozone Monitoring Instrument (OMI) and surface observations from the dust monitoring network established by the China Meteorological Administration (http://www.duststorm.com.cn/).
 The linear depolarization value of 0.1 we apply indicates the presence of dust (which is generally mixed with other spherical particles, known as polluted dust, over Taihu). This is lower than dust particle linear depolarization from several studies at other locations, such as ∼0.3 over North Africa [Z. Liu et al., 2011], 0.31 for pure Saharan mineral dust [Freudenthaler et al., 2009], and ∼0.2 over Chungli, Taiwan, during a strong Asian dust storm event [Nee et al., 2007]. However, most of these studies focused on pure dust particles near source regions or strong dust activities located away from dust regions, where the non-spherical dust particle fraction is much larger than the spherical particle fraction. This value chosen forδlinear(0.1) is similar to the threshold value used for identifying polluted dust in the Cloud-Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO) automated aerosol classification algorithm [Omar et al., 2009] and for defining weak/moderate dust layers by Mishra et al. . Using these criteria, 13, 6, and 7 vertical extinction profiles in spring, autumn, and winter, respectively, were identified as under the influence of dust aerosols. None of the summertime profiles indicated any dust activity.
 Figure 11 presents rich quantitative information about the vertical profiles of aerosols obtained from polarized lidar measurements. These images illustrate the intensity of the normalized relative backscatter (NRB) (Figures 11a and 11c) and the log10 of linear depolarization ratio (Figures 11b and 11d). Figures 11a and 11b and Figures 11c and 11dshow images of lidar profiles for relatively dust-free conditions (January 2, 2009) and heavy dust conditions (March 16, 2009), respectively, with the strikingly different linear depolarization ratios indicative of the very different aerosol types on these two days. It is apparent thatFigures 11a and 11bcorrespond a day with a well-defined boundary layer dominated by spherical particles and a slightly elevated aerosol layer whileFigures 11c and 11d show an abundance of highly depolarizing aerosols at the surface for much of the day along with extensive elevated layers exhibiting varying depolarizing properties.
Figure 11. (a and c) The intensity of the normalized relative backscatter (NRB) and the (b and d) logarithm of linear depolarization ratio on a dust-free day (Figures 11a and 11b: January 2, 2009) and a dusty day (Figures 11c and 11d: March 16, 2009).
Download figure to PowerPoint
 Figure 12shows seasonal mean aerosol extinction profiles in the presence of dust (red lines) and absence of dust (black lines) in spring, autumn and winter; horizontal bars represent standard deviations. Differences in extinction between dusty and non-dusty cases in spring and fall are greater than during wintertime. In spring, from the surface to 2.5 km and between 3 km and 4.5 km, aerosol extinction coefficients in the presence of dust are larger than those in the absence of dust, illustrating that dust aerosols enhance aerosol extinction near the surface and at higher altitudes [Tsai et al., 2008; Rajeev et al., 2010; J. Liu et al., 2011]. Between 0.5 km and 1.5 km in autumn, aerosol extinction coefficients in the presence of dust are smaller than those in its absence, indicating that non-dust aerosol particles contribute to larger values of extinction. No significant difference between wintertime extinction profiles is found for dusty and non-dusty cases. No significant elevated dust layers were detected, which may be due to dust transport patterns and dust deposition in the lower atmosphere, which is typical in eastern and southern China [Zhou et al., 2002; Tsai et al., 2008].
Figure 12. Seasonal mean aerosol extinction coefficient profiles at 527 nm for cases with dust (red lines) and without (black lines) in (a) spring, (b) autumn and (c) winter from June 2008 to May 2009. Horizontal bars represent standard deviations.
Download figure to PowerPoint
3.4. Aerosol Radiative Effects
 The Santa Barbara DISORT Atmospheric Radiative Transfer (SBDART) model [Ricchiazzi et al., 1998] is used to estimate broadband (0.25 to 4.0 μm) total, diffuse, and direct shortwave irradiances and to evaluate aerosol particle direct radiative effects. It is based on the low-resolution band models developed for LOWTRAN 7 atmospheric transmission and the DISORT radiative transfer model. Inputs to the model include the vertical profiles of water vapor and ozone, obtained by partitioning total column water vapor amounts from AERONET retrievals and ozone amounts from OMI according to a standard model atmosphere [McClatchey et al., 1972]. Aerosol particle extinction coefficient profiles were derived from the MPL, normalized by total AOD from the Cimel, while SSA and ASY were assumed constant as derived following AERONET. The aerosol properties, including AOD, SSA and ASY at four AERONET wavelengths (i.e., 440 nm, 675 nm, 870 nm, and 1020 nm) were used to interpolate and extrapolate into the spectral divisions of the SBDART model [Xia et al., 2007]. Surface albedo data were applied from the MODIS Level 2 Collection 5 spectral surface reflectance product (MOD09). Shortwave irradiances under cloud-free conditions were simulated with and without aerosol particles and were then used to determine aerosol direct radiative forcing (ADRF) at the surface (SFC) and at the TOA. A previous study shows that SBDART simulations of downwelling broadband flux at the surface agrees exceptionally well with ground-based measurements, and that modeled upwelling TOA fluxes were compatible with Clouds and Earth's Radiant Energy System satellite retrievals in terms of absolute differences [Li et al., 2010]. They also showed that the combined error caused by uncertainties in main input parameters, including AOD, α, SSA, ASY, surface reflectance, and ozone amounts, is 8.76 ± 3.44 W/m2.
 ADRF is commonly used term for quantifying the direct effect of aerosols on the atmospheric energy budget. It is usually expressed as instantaneous or diurnal mean radiative forcing [Xia et al., 2007]. The definition of diurnal mean radiative forcing is often given as
where F(t) represents instantaneous radiative forcing values. The aerosol radiative forcing within the atmosphere (ATM) is defined as the difference between radiative forcings at the TOA and SFC.
 The left plot in Figure 13a depicts the seasonal and annual mean diurnally averaged ADRF during the study period. Note that ADRF in summer is for June only because there were no SSA and ASY retrievals made in July and August. Annual mean ADRF estimates at the SFC, the TOA and within the ATM were −34.8 ± 9.1, −8.2 ± 4.8 and 26.7 ± 9.4 W/m2, respectively. The magnitude of ADRF at the SFC is comparable with values reported in other studies made in eastern China for different periods (usually one year) or different locations. For example, the largest negative values for ADRF at the surface ranged from −32 to −20 W/m2 over eastern China [Li et al., 2010]. Xia et al.  reported that the annual mean SFC ADRF at Taihu was −38.4 W/m2, based on ground-based radiation measurements collected from September 2005 to August 2006. Global mean estimates of ADRF at the SFC, at the TOA, and within the ATM from observations over land are −11.9, −4.9 and 7.0 W/m2, respectively, and from modeling, are −7.6, −3.0 and 4.0 W/m2, respectively [H. Yu et al., 2006, 2009]. The magnitudes of ADRF at the SFC and within the ATM at Taihu are more than three times greater than global mean values over land [Yu et al., 2006; Xia et al., 2007], implying strong cooling at the surface and warming of the lower troposphere. Relatively high RH over eastern China can partly explain the large aerosol cooling effect at the SFC because the ADRF estimate within the surface boundary layer is strongly dependent on RH [Cheng et al., 2008]. Enhanced warming within the lower troposphere is not surprising because eastern China experiences high aerosol mass loading and an abundance of absorbing aerosol particles like smoke [Li et al., 2007].
Figure 13. (a) Seasonal and annual aerosol direct radiative forcing (ADRF) at the surface (SFC), the top-of-the-atmosphere (TOA), and within the atmosphere (ATM) and (b) seasonal changes in surface downwelling global, direct and diffuse shortwave irradiances induced by aerosols over the period of June 2008 to May 2009.
Download figure to PowerPoint
 It is important to understand how relative proportions of direct and diffuse downwelling surface fluxes are modulated by the presence of aerosols. For example, Gu et al.  found that diffuse radiation results in higher light use efficiencies by plant canopies, which has implications for the study of the global carbon cycle. Figure 13b shows seasonal and annual diurnal mean values of ADRF for global, direct and diffuse radiation. In the presence of aerosols, direct shortwave fluxes reaching the surface were reduced by −109.2 ± 49.4 W/m2 and diffuse shortwave fluxes were enhanced by 66.8 ± 33.3 W/m2 in terms of annual means.
 The aerosol radiative forcing efficiency (ARFE) is the rate at which irradiance within a certain wavelength range changes per unit of AOD, and is an indicator of the radiative forcing potential of a given type of composite aerosol [Pathak et al., 2010]. It is also useful for quantifying and comparing aerosol particle radiative effects, as well as seeing more clearly how their optical properties impact ADRF at different sites under a wide range of conditions [Yu et al., 2006; Wang et al., 2010]. The annual mean magnitudes of ARFE in our study were −54.4 ± 5.3, −13.1 ± 1.5 and 41.3 ± 4.6 Wm−2 τ−1 at the SFC, TOA and ATM, respectively, which are significantly larger than the mean values over China given by Li et al.  (−35.1, −0.5 and 34.1 Wm−2τ−1 at the SFC, TOA and ATM, respectively). However, the ARFE at the SFC at Taihu is clearly smaller than that calculated at some cities in northern China, such as Xianghe (−65.4 ± 4.7 Wm−2 τ−1) and Beijing (−61.2 ± 3.5 Wm−2 τ−1) [Li et al., 2010]. For context, studies made at other locations show that mean ARFE at the SFC at the Kaashidhoo Climate Observatory (KCO), Maldives is −72.2 ± 5.5 Wm−2 τ−1 [Bush and Valero, 2002], −75 Wm−2 τ−1 over the Indian Ocean [Ramanathan et al., 2001a], −85 Wm−2 τ−1 in the Middle East [Markowicz et al., 2002], −83 Wm−2 τ−1 over northeastern India [Pathak et al., 2010], and −89.4 Wm−2 τ−1 induced by black carbon and organic carbon aerosol over the South Asian region [Wang et al., 2007].
 No previous studies conducted in China attempt to apportion the radiative energy trapped in the atmosphere in the presence of aerosols from the surface to the upper atmosphere. Thanks to lidar measurements available for this study, this term is computed here in terms of the radiative heating rate, which is an indicator of the climatic impact of aerosols, and defined as
where is the heating rate (K/day), g is the acceleration due to gravity, Cp is the specific heat capacity of dry air at constant pressure, F is the atmospheric forcing and dP is the change in atmospheric pressure. 527 nm particle extinction profiles retrieved from the MPL provide key information describing vertical particle distributions, which is the first of many factors influencing the radiative heating rate. Unfortunately, vertical profiles of SSA require multiwavelength approaches [Cattrall et al., 2005]. As such, the column-mean SSA derived by AERONET is assumed in this investigation.
 Figure 14 shows seasonal diurnal mean vertical profiles of aerosol particle heating rate (black solid lines), with corresponding standard deviations (gray horizontal lines). The vertical distribution of heating rate is consistent with the vertical distribution of mean aerosol extinction coefficients (i.e., as the aerosol particle extinction coefficient increases, the heating rate increases). The maximum heating rate was 0.87, 1.91, 0.87 and 1.00 K/day in spring, summer (June only), autumn and winter, respectively. The heating rate in summer does not vary smoothly with height, which implies that the energy distribution at different altitudes can significantly change. This would modify atmospheric static stability and influence convection [Ramanathan et al., 2001b]. In spring, the heating rate is fairly constant within the mixed layer of the atmosphere. In autumn, the heating rate increases with height. This would have an impact on convective instability and thermal profiles, which could induce inversions at lower altitudes, which is not conducive to lofting of pollutants from near the surface. The relatively large change in heating rate with height within the lower troposphere in winter would also have a significant impact on convective instability. From the surface to 2 km, the annual mean heating rate due to aerosols is 0.74 K/day with seasonal means of 0.64, 1.26, 0.56 and 0.50 K/day in spring, summer, autumn and winter, respectively.
Figure 14. Seasonal mean aerosol heating rate profiles over Taihu in (a) spring, (b) summer, (c) autumn, and (d) winter from June 2008 to May 2009. Horizontal lines represent standard deviations.
Download figure to PowerPoint
 One uncertainty in the calculated ADRF and heating rate profiles is the lack of knowledge about the vertical variability of aerosol optical properties, especially SSA. Guan et al. found that the difference in radiative forcing for different SSA profiles is small at the surface, while at the TOA a change of ∼10% occurs if absorbing aerosol is assumed in the elevated layer in lieu of all absorption occurring in the boundary layer. These uncertainties become obvious when radiative transfer calculations assume a constant column-mean SSA for the multilayer aerosol vertical distribution, especially when aerosols originate from different source regions for each layer [Wang et al., 2010]. Based on lidar depolarization measurements, elevated dust layers were not found, although, this does not rule out changes in particle properties such as size or composition. No cases of multiple aerosol layers were found that decrease associated uncertainties in our calculations. McFarlane et al.  provide some suggestions on improving our understanding of the vertical distribution of aerosol optical properties, such as use of a multiple wavelength lidar or a combination of the MFRSR with longwave sensors.