2.1 Selection of the aerosol microphysical parameters

As anticipated, an average continental aerosol type (i.e., describing clean to moderately polluted continental conditions; e.g., Hess et al., 1998) was targeted in this study, this being the aerosol type expected to dominate over Europe. Based on a scheme originally proposed by d'Almeida et al. (1991) and a large set of observational evidence (e.g., Van Dingenen et al., 2004), in this work the size distribution is described as an external mixture of three size modes. These are (in order of increasing size range) (1) a first ultrafine mode; (2) a second fine mode, mainly composed of water-soluble particles; and (3) a third mode of coarse particles.

A three-mode lognormal size distribution described by Eq. (1) is employed for this purpose:

In Eq. (1), r_i, σ_i, and N_i are respectively the modal radius, the width, and the particle number density of the ith aerosol mode (i= 1, 2, 3). At each computation, r_i and σ_i are randomly chosen within a relevant variability range. Values of N_i are conversely obtained by firstly randomly choosing the total number of particles, N_tot, to be included in the whole size distribution ($N_{\mathrm{tot}}=N_{\mathrm{1}}+N_{\mathrm{2}}+N_{\mathrm{3}}$) and then by applying specific rules for the number mixing ratio, $x_i=N_i/N_{\mathrm{tot}}$, of each component to this total. To reproduce clean to moderately polluted continental conditions, the value of N_tot is made variable between 500 and $\mathrm{1}\times {\mathrm{10}}^{-\mathrm{4}}$ cm⁻³ (e.g., Hess et al., 1998; Van Dingenen et al., 2004). As the result of different sources and processes, the three modes are also assumed to have a different composition, which impacts the optical computations through the relevant particle refractive index (m_i), with both its real and imaginary components ($m_i=m_{\mathrm{r}\mathit{\_}i}-i\times m_{\mathrm{im}\mathit{\_}i})$. The Mie theory for spherical particles of radius r_i and refractive index m_i is then used to compute the extinction and backscatter coefficients (see below).

Table 1Aerosol parameter values as reported in literature for continental-type aerosols.

^aThe refractive index is at λ=532 nm. ^bThe refractive index is at λ=550 nm.

Download Print Version | Download XLSX

Table 2Variability ranges used in this study. Values refer to ground and dry conditions (see text for details).

Download Print Version | Download XLSX

A description of the assumptions made for each mode and relevant parameter, mostly based on literature data (Table 1), is given hereafter; the summary of the relevant variability chosen for each parameter is provided in Table 2.

1. First mode. This ultrafine mode is the one more directly simulating fresh anthropogenic emissions. The number mixing ratio x_i=1 ($N_{i=\mathrm{1}}/N_{\mathrm{tot}}$) of this mode is let variable between 10 % (rural conditions; Van Dingenen et al., 2004) and 60 % (more polluted conditions; Hess et al., 1998). The variability in its modal radius (r₁=0.005–0.03 µm) is chosen to include nucleation-mode particles to Aitken-mode particles. To take into account the wide variability in species within this ultrafine mode, from non-absorbing (e.g., inorganic particles) to highly absorbing materials (e.g., black carbon), wide ranges of variability have been set for its refractive indexes (at λ=355 nm: m_{r_1} in the range 1.40—1.8, and m_{im_1} in the range 0.01–0.47; see Table 2 for the corresponding values at λ=532 and 1064 nm).

2. Second mode. The second aerosol mode accounts for 40 %–90 % of N_tot, with (dry) r₂ between 0.03 and 0.1 µm. Its composition (m_{r_2}, and m_{im_2}) is also made highly variable so as to include water-soluble inorganic and organic particles (Hess et al., 1998; BG04a; Dinar et al., 2008). In this case, at λ=355 nm, m_{r_2} is in the range 1.40–1.7 and m_{im_2} is in the range 0.0001–0.01 (Table 2).

3. Third mode. This coarser aerosol mode (modal radius r₃ in the range 0.3–0.5 µm) is mainly intended to account for soil-derived (dust-like) particles that are a primary continental emission. A quite narrow variability is thus fixed for its m_{r_3} and m_{im_3} values (1.5–1.6 and 0.0001–0.02, respectively, at 355 nm). The relevant number mixing ratio x₃ (N₃∕N_tot) is set as variable between 0.01 % and 0.5 %, with this mode contributing mostly to the total aerosol volume (thus mass) rather than to the total number of particles.

As mentioned, refractive indexes were also made wavelength dependent, as this feature is also typically observed as linked to the different particle composition. In particular, for the second mode (water-soluble particles) we include an increase with the wavelength of the upper boundary values of m_{im_2} and a decrease in m_{r_2} at λ=1064 nm (d'Almeida et al., 1991). For the (dust-like) third-mode particles, the upper boundary values of m_{im_3} are set to decrease with increasing wavelengths (Gasteiger et al., 2011; Wagner et al., 2012).

For convenience, the aerosol parameter boundaries summarized in Table 2 refer to dry particles and to ground level. However, the effect of a variable relative humidity (RH), its variability with altitude, and the generally observed decrease in particle number with altitude is also considered in the model. More specifically, the number of particles in each mode, N_i, and RH are both made altitude dependent through the following equations (Patterson et al., 1980; BG01):

The altitude z is variable here between 0 and 5 km. N_i(0) and H_i in Eq. (2) are the number of particles at the ground and the scale height for each mode, respectively.

To describe the altitude effect, in Eq. (2) an exponential decrease with height of the particle number density is assumed. To rescale the particle number density of the different modes, $H_{i=\mathrm{1},\mathrm{2}}$ is set equal to 5.5 km (Barnaba et al., 2007) while H_i=3 (coarse particles) is set to 0.8 km (Barnaba et al., 2007). In Eq. (3), the additional term (1 + dRH) allows a further variability with respect to the mean RH(z) profile assumed; here dRH is randomly chosen between −60 and +60. Values of RH greater than 95 % are discarded to avoid divergence.

Additionally, while the first and third modes are assumed to be water insoluble, the second mode (i=2) is fully hygroscopic. Aerosol humidification is thus considered to act on both particle size and refractive indices of the second aerosol mode (e.g., BG01) as

In Eqs. (4) and (5), r_{2_RH} and m_{2_RH} are the RH-corrected modal radius and refractive index for the second mode, respectively; r_{2_0} and m_{2_0} are the particle dry modal radius and refractive index, respectively; m_w is the water refractive index (assumed to be equal to $\mathrm{1.34}-i\mathrm{7}\times {\mathrm{10}}^{-\mathrm{9}}$, and $\mathrm{1.33}-i\mathrm{1.3}\times {\mathrm{10}}^{-\mathrm{9}}$, $\mathrm{1.33}-i\mathrm{2.9}\times {\mathrm{10}}^{-\mathrm{6}}$ at 355, 532, and 1064 nm, respectively).

Finally, following Barnaba et al. (2007), an increase in the width of the size distribution with altitude (Eq. 6) has been introduced for the first and second aerosol modes:

In fact, Barnaba et al. (2007) showed that this was necessary to better reproduce the observed decrease in the lidar ratio (LR) with altitude, and is likely related to a broadening of the particle size distribution with aging.

Once the value of each microphysical parameter is randomly selected within its relevant variability range, and once corrections are applied following Eqs. (2)–(6), each resulting aerosol size- and composition-resolved distribution is used to compute the aerosol S_a and V_a, as well as to feed a Mie code (assumption of spherical particles; Bohren and Huffman, 1983) to compute β_a and α_a (BG01; see also Fig. 1). Overall, the equations used are as follows.

Here Q_bsc (r_i, λ, m_i) and Q_ext (r_i, λ, m_i) are, respectively, the backscatter and the extinction efficiencies. As mentioned, the optical computations are made at the three different wavelengths: 355, 532, and 1064 nm (i.e., those of Nd:YAG laser harmonics, the most common wavelengths used by ground-based and spaceborne aerosol lidars).

Since in our simulations the third aerosol mode is intended to represent dust-like particles, an empirical correction for non-sphericity is also applied to the Mie-derived optical properties of this mode. This procedure is based on BG01, which uses the results of Mishchenko et al. (1997) obtained for surface-equivalent mixtures of prolate and oblate spheroids.

2.2 Model simulation results

In Fig. 2 we show the results of 20 000 simulations of continental aerosol optical and physical properties derived randomly, varying the relevant aerosol size distributions and compositions as described in the previous section. In particular, the results for α_a, S_a, and V_a are shown as a function of β_a in Fig. 2a, b, c (blue crosses) referring to λ=1064 nm. For each variable (A), the average value per bin of β_a and relevant standard deviations (〈A〉±dA) are shown as red dots and vertical bars, respectively. Note that 10 equally spaced bins per decade of β have been considered and that 〈A〉±dA values are only shown for bins containing at least 1 % of the total number of pairs. Corresponding relative errors (dA∕〈A〉) are depicted in Fig. 2d, e, f. Some sensitivity tests of these model outputs to the variability in the input microphysical parameters employed are provided in Appendix A.

Figure 2Scatter plots of (a) α_a (km⁻¹), (b) S_a (cm² cm⁻³), and (c) V_a (cm³ cm⁻³) vs. backscatter β_a (km⁻¹ sr⁻¹) and relevant relative errors (d, e, f) as derived from 20 000 model computations (blue points) at λ=1064 nm. Red dots and error bars are the average values per decade of β and their standard deviations; green lines are the seventh-order polynomial fit curve of the 20 000 points.

Download

Based on these results, at step two of the procedure (see scheme in Fig. 1), we derive aerosol-specific mean relationships linking aerosol extinction, surface area, and volume (α_a, S_a and V_a) to its backscatter (β_a). For this purpose, we used a seventh-order polynomial fit in log–log coordinates. The choice of a seventh-order polynomial fit was made for homogeneity with BG01 and BG04a. These relationships are shown as green lines in Fig. 2a, b, c while the relevant fit parameters are reported in Table 3 referring to λ=1064 nm (fit parameters related to computations at λ=355 and 532 nm are given in Table B1 and Table B2, Appendix B).

Table 3Parameters of the seventh-order polynomial fits ($y=a_{\mathrm{0}}+a_{\mathrm{1}}x+a_{\mathrm{2}}{x}^{\mathrm{2}}+a_{\mathrm{3}}{x}^{\mathrm{3}}+a_{\mathrm{4}}{x}^{\mathrm{4}}+a_{\mathrm{5}}{x}^{\mathrm{5}}+a_{\mathrm{6}}{x}^{\mathrm{6}}+a_{\mathrm{7}}{x}^{\mathrm{7}}$) for λ=1064 nm, with x=log(β_a) (in km⁻¹ sr⁻¹) and $y=\log ({\mathit{\alpha }}_{\mathrm{a}},S_{\mathrm{a}},or{V}_{\mathrm{a}})$ in km⁻¹, cm² cm⁻³, and cm³ cm⁻³, respectively.

Download Print Version | Download XLSX

The red vertical bars of Fig. 2 also highlight the ranges of α_a, S_a, and V_a, which are statistically significant, i.e., those in which, at λ=1064 nm, the model provides at least 1 % of the total points per corresponding bin of β_a. These are 10⁻⁴–10⁻¹ km⁻¹, 10⁻⁷–10⁻⁵ cm² cm⁻³, and 10⁻¹³–10⁻¹⁰ cm³ cm⁻³, for α_a, S_a, and V_a, respectively, corresponding to the backscatter range $\mathrm{9}\times {\mathrm{10}}^{-\mathrm{5}}\le {\mathit{\beta }}_{\mathrm{a}}\le \mathrm{4}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹. In terms of aerosol property variability, the relative errors associated with α_a and V_a show almost no dependence on β_a, with values between 30 % and 40 %. Conversely, the modeled aerosol surface area exhibits a larger dispersion, with relative error values spanning the range 40 %–70 %, and decreasing as β_a increases.

Figure 3(a, b, c) Scatter plots of LR (sr) versus β_a (km⁻¹ sr⁻¹) at (a) 355 nm, (b) 532 nm, and (c) 1064 nm (blue points). The seventh-order polynomial fit curve (green lines) and the average values per decade of β together with their standard deviations (red points and red vertical bars, respectively) are also reported. Horizontal black lines are mean values of the weighted-LR and ±1 SD (solid and dashed lines, respectively). (d, e, f) Relative errors associated with the model-derived LR at (d) 355 nm, (e) 532 nm, and (f) 1064 nm.

Download

A key parameter for the inversion of lidar signals is the LR, i.e., the ratio between α_a and β_a (Ansmann et al., 1992). In Fig. 3 we thus show the results of our simulations in terms of LR vs. β_a at the three λ (355, 532, and 1064 nm, Fig. 3a, b, c, respectively) and relevant dLR ∕ LR values (Fig. 3d, e, f, respectively). The color code is the same as in Fig. 2. Additional horizontal black lines have been inserted representing values (solid central lines) of the weighted-LR ± 1 standard deviation (dotted side lines), i.e., the LR weighted by the number of simulated points in each considered backscatter bin. The weighted-LR values derived at 355, 532, and 1064 nm are 50.1±17.9, 49.6±16.0, and 37.7±12.6 sr, respectively. Figure 3 also allows showing that the statistically significant regions of simulated backscatter values shift towards smaller values with increasing λ (e.g., at λ=355, the β_a extending region is $\mathrm{4}\times {\mathrm{10}}^{-\mathrm{5}}$–$\mathrm{2}\times {\mathrm{10}}^{-\mathrm{2}}$ km⁻¹ sr⁻¹, whereas, at 532 nm, it ranges between $\mathrm{2}\times {\mathrm{10}}^{-\mathrm{5}}$ and $\mathrm{1}\times {\mathrm{10}}^{-\mathrm{2}}$ km⁻¹ sr⁻¹). Furthermore, Fig. 3 reveals a quite different shape of the LR vs. β_a functional relationships (green curves) at different wavelengths. At 355 and 532 nm the curve is concave, with quite similar LR maxima of the fitting curve (54.3 and 53.8 sr at approximately ${\mathit{\beta }}_{\mathrm{a}}=\mathrm{4}\times {\mathrm{10}}^{-\mathrm{4}}$ and $\mathrm{2}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹, respectively). At 1064 nm the curve is conversely monotonic, with a flex point at β_a=3–$\mathrm{4}\times {\mathrm{10}}^{-\mathrm{4}}$ km⁻¹ sr⁻¹. A larger data dispersion also characterizes the results at λ=355 and 532 nm (LR values from 10 to 90 sr) in comparison to λ=1064 nm (LR in the range 18–80 sr, except for a minor number of outliers). This translates into different LR relative errors at UV, VIS, and infrared (IR) wavelengths. At 1064, dLR∕LR slightly decreases for increasing backscatter, with values of around 35 %. At the shorter wavelengths, dLR∕LR increases as a function of β_a, with a large (>40 %) relative error for values of ${\mathit{\beta }}_{\mathrm{a}}>\mathrm{2}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹.

Table 4Mean weighted LR at 355, 532, and 532 nm derived in this work and comparison to the corresponding aerosol subtypes (clean continental, CC, and polluted continental, PC) from relevant literature.

^a Derived using the extinction-related and backscatter-related Ångström exponents given by Amiridis et al. (2015). ^b See the explanation in the text for the two different values.

Download Print Version | Download XLSX

To insert our results into a more general context, we compared the derived, model-based weighted-LR values to some LR data reported in the literature (Table 4). In particular, we selected some of the works using the aerosol model developed to invert the CALIPSO lidar data. For example, Omar et al. (2009) consider six different aerosol subtypes: clean continental (CC), clean marine (CM), dust (D), polluted continental (PC), polluted dust (PD), and smoke (S). Our model-derived LR at 532 nm falls in the middle of the range (35–70 sr) fixed by the CALIPSO CC and PC aerosol classes. The work by Papaggianopoulos et al. (2016), in which the LR values are adjusted according to EARLINET observations, reports a LR range at 532 nm of 47–62 sr. At the same wavelength, the aerosol range defined by the LIVAS climatology (LIdar climatology of Vertical Aerosol Structure for space-based lidar simulation studies; Amiridis et al., 2015) is 54–64 sr. In both cases, our model seems to be closer to the LR values of the CC aerosol type, which is compatible with our intention to simulate the clean to moderately polluted continental aerosol type. At 532 nm, our LR value is also reasonably in between the CC and PC LR values derived by Omar et al. (2009), but again closer to the CC LR value. The very small decrease in LR values between 532 and 355 nm estimated by LIVAS for the CC aerosol is also consistent with our results. Similarly, our model predicts a lower mean LR in the near IR with respect to the green, in agreement with results of Amiridis et al. (2015) in CC conditions and not with those in polluted conditions. Table 4 also includes the continental aerosol LR values estimated in the work of Düsing et al. (2018) through comparison between airborne in situ and ground-based lidar measurements. Our model is in good agreement with their LR values at 355 and 532 nm. At 1064 nm, the algorithm developed by Düsing et al. (2018) provided a value of LR of around 15 sr. Conversely, in the same study the authors found that, rather, a value of LR = 30 sr gives the better agreement between their Mie and lidar-based α_a, this value being closer to our model-derived one at 1064 nm (LR = 37.7). The difference between these two values is explained by the authors to be probably due to the estimation of the aerosol particle number size distribution, a critical parameter for a reliable modeling of aerosol particle backscattering.

Table 5Extinction-to-volume conversion factors, $c_{\mathrm{v}}=V_{\mathrm{a}}/{\mathit{\alpha }}_{\mathrm{a}}$ (and corresponding mass-to-extinction efficiency values, MEE $={\mathit{\alpha }}_{\mathrm{a}}/(V_{\mathrm{a}}\cdot {\mathit{\rho }}_{\mathrm{a}}$), computed employing ρ_a=2 g cm⁻³) of continental particles as derived from our model at different wavelengths.

Download Print Version | Download XLSX

As a last added value of the outcome from our model-based results, we derive here and provide in Table 5 extinction-to-volume conversion factors, $c_{\mathrm{v}}=V_{\mathrm{a}}/{\mathit{\alpha }}_{\mathrm{a}}$ (e.g., Ansmann et al., 2011) at three different wavelengths (355, 532, 1064 nm) and compare these to similar outcomes from other studies. To our knowledge, values of continental particles c_v at three wavelengths are only available in Mamouri and Ansmann (2016). Note that c_v is also proportional, through the particle density ρ_a, to the inverse of the so-called mass-to-extinction efficiency (MEE, i.e., ${\mathit{\alpha }}_{\mathrm{a}}/(V_a\cdot {\mathit{\rho }}_{\mathrm{a}}))$, a parameter important in several aerosol-related applications (e.g., the estimation of PM from satellite AOT or in modules of global circulation and chemical transport models to compute aerosol radiative forcing effects; Hand and Malm, 2007). For convenience, model-derived MEE values are also included in Table 5.

3.1 Comparison of the modeled aerosol optical properties to EARLINET measurements

As mentioned, EARLINET Raman stations perform coordinated measurements 2 days per week following a schedule established in 2000 (Bösenberg et al., 2003). Overall, the EARLINET database includes the following categories: climatology, CALIPSO, Saharan dust, volcanic eruptions, diurnal cycles, cirrus, and others (forest fires, photo smog, rural or urban, and stratosphere). To be comparable to our results, we used EARLINET β_a and α_a coefficients at 355 and 532 nm within the quality-assured (QA) climatology category (Pappalardo et al., 2014). However, note that additional data filtering was necessary to screen out residual, likely unreliable values within this QA climatology category. In particular, we only selected those EARLINET QA data further satisfying the following criteria:

• β_a and α_a coefficients evaluated independently, i.e., only obtained using the Raman method (Ansmann et al., 1992);

• β_a and α_a>0;

• LR < 100;

• relative errors on β_a and α_a<30 %.

Then, we selected those sites in Europe expected to be mostly impacted by continental aerosols and having the largest datasets (e.g., at least 100 points) at 355 and 532 nm. Overall, five sites satisfied these conditions (Table 6), namely Madrid (Spain), Potenza and Lecce (Italy), and Leipzig and Hamburg (Germany). Finally, being interested in continental conditions here, we filtered out those measurement dates affected by desert dust at the measuring sites, i.e., we removed from our “model–measurement comparison dataset” all the dates within the EARLINET climatology category also belonging to the EARLINET Saharan dust category.

Table 6Main characteristics of the dataset of the EARLINET continental sites considered in this study. The listed dataset refers to the data downloaded from the EARLINET site (last access on the 11 January 2018). NA – not available.

Download Print Version | Download XLSX

Figure 4 depicts the results of the model–measurement comparison at the sites fulfilling our requirements in terms of LR vs. β_a at λ=355 nm (the corresponding results at λ=532 nm, including Madrid in place of Hamburg, are given in Appendix C, Fig. C1). The colored area represents the model-simulated data range, while the color code indicates the absolute number of simulated values (i.e., counts) in each β_a–LR pair. The EARLINET-measured values are reported as open black circles. Note that, since the model simulations are performed over an altitude range of 0–5 km (see Sect. 2.1), only those simulations corresponding to the altitude range (Δz) covered by the measurements at each EARLINET station were taken into account here. Figure 4 shows that the model results encompass the measured LR vs. β_a data well, with a few measurements outside the modeled range (most of the exceptions are found for Potenza). Statistically, the highest number density of simulated data fits the observations well, with the exception of Hamburg (Fig. 4a), which has the lowest number of measured data (it is not an EARLINET station any longer; see Table 6).

Figure 4Scatter plots of LR (sr) versus β_a (km⁻¹ sr⁻¹) at 355 nm as simulated by our model (colored region) and measured by EARLINET lidars (open black circles) in Hamburg (Germany) (a), Lecce (Italy) (b), Leipzig (Germany) (c), and Potenza (Italy) (d). The colored area is the region of simulated values, the color code indicating the number of simulated values in each β_a–LR pair (see legend). In particular, the color 2-D histogram is computed using a semilogarithmic box consisting of 10 equally spaced bins per decade of β_a on the x axis and five spaced LR values on the y axis.

Download

Table 7Mean LR discrepancies between our model results and EARLINET measurements and weighted LR at 355 and 532 nm for the considered EARLINET stations. LC: Lecce; LE: Leipzig; PO: Potenza; HH: Hamburg; MA: Madrid.

Download Print Version | Download XLSX

In Fig. 5 the previous results at λ=355 nm are converted in terms of mean LR per bin of β_a for both model (blue) and observations (red, again, only β_a bins containing at least 1 % of the total modeled data were considered). This view shows that there is a general good agreement between the modeled and the measured LR values, and in their variation with β_a. Some major deviations are found for Potenza and are further discussed in the following. The model–measurement agreement shown in Fig. 5 was evaluated in quantitative terms by computing mean LR relative differences at both λ=355 and 532 nm; i.e., we derived ([(LR${}_{\mathrm{mod}}-{\mathrm{LR}}_{\mathrm{meas}})/{\mathrm{LR}}_{\mathrm{meas}}]\cdot \mathrm{100}$) values, for which LR_mod and LR_meas are the LR values computed by the model and derived using lidar measurements, respectively. These values are reported in Table 7 for each considered EARLINET station, together with the measurement-based mean LR in each observational site (computed weighting the number of observations per β_a-spaced bins).

Figure 5Model-simulated (blue) and lidar-measured (red) LR vs. β_a mean curves at 355 nm calculated per 10 equally spaced bins per decade of β_a at the (a) Hamburg (HH), (b) Lecce (LC), (c) Leipzig (LE), and (d) Potenza (PO) EARLINET lidar stations. Vertical bars are the associated standard deviations.

Download

Results in Fig. 5 and Table 7 also give some hints on the capability of the aerosol type assumed (and its admitted ranges of variability) to reproduce real continental aerosol conditions at different sites across Europe. In fact, the four continental sites selected with our criteria are still expected to be partially impacted by different aerosol types.

• A good agreement between the model and the observations in terms of LR mean values is found for Hamburg (Fig. 5a), with mean LR differences of the order of 5 % (Table 7). Still, the measured LR values have a high variability and their distribution is positioned towards high values of β_a ($\mathrm{1}\times {\mathrm{10}}^{-\mathrm{3}}$ to $\mathrm{4}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹). This could be due to the presence of different aerosol types as slightly polluted marine and polluted aerosol (Matthias and Bösenberg, 2002).

• A good accord for Leipzig (Fig. 5c) also indicates that this site is mostly dominated by pure continental particles. In fact, the distribution of observed LR points in Fig. 4, which covers β_a values ranging from $\mathrm{2}\times {\mathrm{10}}^{-\mathrm{4}}$ to $\mathrm{3}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹, is well centered to the modeled simulations' highest density (counts >40). Table 7 shows that at both wavelengths mean discrepancies with LR measurements stay well below 10 %.

  The highest differences in Fig. 5 are found at some southern Europe EARLINET sites.

• In Lecce (Fig. 5b), the best agreement between model and observations is found for the lowest values of β_a (between $\mathrm{9}\times {\mathrm{10}}^{-\mathrm{4}}$ and $\mathrm{1}\times {\mathrm{10}}^{-\mathrm{3}}$ km⁻¹ sr⁻¹; see Table 7). Also, the increase from 10 % to 18 % in the discrepancies at 355 and 532 nm indicates some model problems in correctly reproducing the spectral variability in the optical properties, suggesting some mismatch between modeled and real aerosol sizes at this site (see discussion below).

• In Potenza (Fig. 5d), a significant difference between the mean LR curves emerges for β_a values $>\mathrm{6}\times {\mathrm{10}}^{-\mathrm{4}}$ km⁻¹ sr⁻¹ , with observed LR values lower than those simulated here.

These discrepancies could be due to the influence of marine aerosols at both stations (De Tomasi et al., 2006; Mona et al., 2006; Madonna et al., 2011), which is expected to produce lower LR values for high values of β_a (e.g., BG01). In fact, Madrid shows better performances, with dLR∕LR values comparable to those in Leipzig.

To provide some insight into the reasons of the model–measurement differences at the Lecce and Potenza sites, some specific model sensitivity tests have been performed and are reported in Appendix D. In particular, for Lecce, we found that better agreement between the observed and simulated LR vs. β_a behavior at 355 nm is obtained by reducing the variability range of N_tot (from 500–10 000 to 500–5000 cm⁻³ at ground). This indicates that Lecce is likely affected by cleaner continental aerosol-type conditions. The sensitivity simulations performed for understanding the mismatches with Potenza measurements show that an extension of the variability range of the coarse-mode radius is needed to reproduce the observed decrease in LR for increasing backscatter (Fig. 5d). This suggests a contribution of coarse particles larger than that assumed (Appendix D). This is compatible with the suspect of marine air contamination, although at this stage we are not able to exclude additional contamination of coarser particles of soil origin.

Overall, mean LR differences between our average continental model and data at selected continental sites in Europe remain lower than 20 % (Table 7), indicating the model reasonably reproduces the clean-to-moderately polluted continental aerosol conditions we intended to simulate well.

3.2 Application of model results to Nimbus CHM 15k ALC measurements

To test and validate the model-based inversion methodology, we used the derived functional relationships (Sect. 2.2) to invert and analyze the measurements of some ALICEnet ALCs (Lufft Nimbus CHM 15k systems). These instruments are biaxial ceilometers that emit laser pulses at 1064 nm (Nd:YAG laser, class M1) with a typical pulse energy of 8 µJ and a pulse repetition rate of about 6500 Hz. The instruments have a specified range of 15 km and fully overlap at around 1500 m (Heese et al., 2010). The manufacturer provides the overlap correction functions (O(z)) for each system. As shown recently by Wiegner and Geiß (2012) and Wiegner et al. (2014), a promising strategy to retrieve the aerosol backscatter coefficient from ALC measurement is adopting the forward solution of the Klett inversion algorithm (Klett, 1985). This solution requires a known calibration constant of the system (i.e., absolute calibration, c_L) and an assumption of the LR. The advantage with respect to the backward solution is that calibration is not affected by the low SNR in the upper troposphere and it is needed occasionally. Furthermore, starting close to the surface, the data retrieval allows us to resolve aerosol layers in the boundary layer even if their optical depth is high. The forward solution of the Klett inversion algorithm is thus adopted here. For convenience, we report here the equations used within our procedure to obtain β_a from ALC measurements, which are also described in Wiegner and Geiß (2012, Eqs. 1–3):

with

and

Here, β_m and α_m are the molecular backscatter and extinction coefficients calculated from climatological monthly air density profiles and z²P(z) is the ALC range-corrected (z) signal (P) (also referred to as RCS), which are the raw data obtained by the considered ALCs. As anticipated, knowledge of the calibration constant c_L is needed to solve Eq. (13) (and thus 11, forward solution). In our analysis of ALC daily records, the constant c_L has been obtained using the “backward approach” (Rayleigh calibration) applied to nighttime cloud-free ALC signals averaged over 1 or 2 h at 75 m height resolution. This allows for using the best c_L retrieval (that is the nighttime lowest noise one) in the forward solution of the lidar equation, which guarantees operating over the best signal-to-noise range of the ALC signal.

3.2.1 Model-based retrieval of aerosol optical properties

Operatively, inversion of the aerosol properties, α_a(z) and β_a(z), is performed using an iterative technique since we need to correct the backscatter signal at each altitude z for extinction losses. The iterative procedure is stopped when convergence in the integrated aerosol backscatter (IAB $={\mathrm{\Sigma }}_{\mathrm{0}}^{\mathrm{zcal}}{\mathit{\beta }}_{\mathrm{a}}\left(z\right)$) is reached (e.g., BG01). At each step, aerosol extinction is derived using the functional relationship α_a=α_a (β_a) of Table 3.

Figure 6Time–height cross section of the aerosol extinction coefficients α_a (km⁻¹), as derived at 1064 nm on 26 June 2016 by the ALICEnet ALC of Saint-Christophe in the Aosta Valley (northern Italy). The orange circle points and the pink line are the AOT values (right y axis) measured by a colocated POM-02L radiometer and estimated from the ALC following our approach.

Download

Figure 7(a) Geographical map of the ALC network ALICEnet. The red circles highlight the selected sites for this study: Saint-Christophe in the Aosta Valley (ASC), San Pietro Capofiume (SPC), and Rome Tor Vergata (ASC). (b)–(d) Histograms of the differences between the hourly-mean coincident AOTs at 1064 nm as derived by ALCs and measured by photometers at ASC, SPC, and RTV, respectively. The different colors (red, blue, and black) depict the different inversion schemes: model-based inversion scheme, LR = 38 sr and LR = 52 sr. In each panel the values of the average measured AOT (and its associated standard deviation) and of the number of considered pairs are also reported.

An example of the outcome of this retrieval methodology is depicted in Fig. 6. It shows the time–height (24 h, 0–6 km) contour plot of α_a retrieved at 1064 nm during a whole day of measurements (26 June 2016) performed by the ALICEnet system of Saint-Christophe in the Aosta Valley (ASC, 45.7^∘ N, 7.4^∘ E 560 m a.s.l., northern Italy; Fig. 7a). Time and altitude resolutions are 1 min and 15 m, respectively. Note that ALC data are cloud-screened using the cloud mask of the Lufft firmware.

The AOT is obtained by vertically integrating the ALC-derived α_a(z) from the surface up to a fixed height z_AOT, above which the aerosol contribution is assumed to be negligible. In Fig. 6, the ALC-derived AOT values at 1064 nm (pink curve, with a temporal resolution of 5 min) are superimposed on the extinction contour. Reference AOT values from a colocated sun–sky radiometer (a Prede POM-02 system) are shown by orange circles. These were extrapolated at 1064 nm from the instrument 1020 nm channel using the Ångström exponent derived fitting AOT values at all the radiometer wavelengths. This example illustrates the very good performances of our model-assisted inversion scheme and the capability of this approach to extend the (daylight-only) radiometer observations to nighttime.

Table 8Main characteristics of the ALC and colocated sun–sky radiometer equipment located at the considered ALICEnet sites.

Download Print Version | Download XLSX

To evaluate the performances of our model-assisted retrieval of α_a(z) over a more statistically significant dataset, the same approach illustrated in Fig. 6 was applied to a longer record at the ASC site, plus Nimbus CHM-15k ALC datasets from two additional ALICEnet sites: San Pietro Capofiume (SPC, 44^∘39 N, 11^∘37 E, 10 m a.s.l.) and Rome Tor Vergata (RTV, 41.88^∘ N, 12.68^∘ E, 100 m a.s.l.). The location of the instruments is shown in Fig. 7a (red circles), while some information on system types and site characteristics is given in Table 8. The data analyzed here were collected during the following periods: April 2015–June 2017, June 2012–June 2013, and February 2014–September 2015 for ASC, SPC, and RTV, respectively.

At those sites, reference AOTs were collected by three colocated sun–sky radiometers, namely using two SKYNET Prede sun–sky radiometers at ASC and SPC (POM-02L and POM-02, respectively, http://www.euroskyrad.net/, last access: 29 October 2018) and an AERONET (AErosol Robotic Networ; Holben et al., 1998) Cimel CE 318-2 instrument operational at RTV (https://aeronet.gsfc.nasa.gov, last access: 29 October 2018, Rome Tor Vergata station, data level 2.0). Only AOT values between 0.01 and 0.2 at 1064 nm were considered. This range allows for excluding the data points with a 1064 nm AOT lower than the sun photometer expected accuracy (dAOT = 0.01) and those for which we found aerosol extinction to cause significant deterioration of our ALC signal. Overall a total of 1237, 268, and 850 AOT pairs were analyzed at ASC, SPC, and RTV, respectively.

Also note that, although CHM-15k data are already corrected for the O(z) function provided by the manufacturer, the variation in the ALC internal temperature was shown to lead to O(z) differences of up to 45 % in the first 300 m above ground (Hervo et al., 2016). For this reason, in our analyses the lowest valid altitude of the CHM-15k for both the SPC and RTV systems was fixed to be about 400 m. A linear fit of the first two valid ALC points is then used to extrapolate α_a(z) down to the ground (z₀). Conversely, due to the optimal characterization down to the ground of O(z) provided by Lufft for the CHM-15k system installed at ASC, values at z₀ at this site are not those extrapolated but actually those measured. The maximum altitude of aerosol extinction vertical integration to derive the AOT, z_AOT, was selected as the first height above 4000 m at which the range-corrected signal (RCS) has a SNR < 1.

Table 9Results of the comparison between the AOT measured by sun photometers and the one derived by ALCs (model-based and fixed-LR inversion schemes) at three ALICEnet stations. Mean differences (expressed in terms of $<\mathrm{dAOT}>=<({\mathrm{AOT}}_{\mathrm{ceil}}-{\mathrm{AOT}}_{\mathrm{phot}})>$, $<\left|\mathrm{dAOT}\right|>$ (module), $<\mathrm{dAOT}/\mathrm{AOT}>$, and $<|\mathrm{dAOTI}/\mathrm{AOT}|>$ ) are reported with their standard deviations.

Download Print Version | Download XLSX

Results of the long-term AOT comparison are summarized in Fig. 7 and Table 9. For each site under investigation, Fig. 7 shows the histograms of the AOT differences between the hourly-mean coincident AOTs as derived by the ALCs and measured by the sun photometers (red curve; corresponding AOT vs. AOT scatter plots at the three considered sites are given in Appendix E). To evaluate the advantage of our approach with respect to more standard lidar inversions, we also computed AOT differences using two fixed-LR values. In particular, we used LR = 52 sr (i.e., the value suggested by the E-PROFILE network, black lines) and LR = 38 sr (i.e., the weighted mean LR value derived from our model; see Sect. 3, blue lines). Figure 7 shows that the best agreement is found at ASC. The distribution of AOT difference has a maximum of around 0 for each of the three inversion schemes, with very low dispersion. The full width at half maximum, FWHM, is in fact around 0.015, and approximately 55 % of the data are included in the interval −0.01–0.01, which is even within the expected error of photometric measurement. For SPC and RTV, the red and blue histograms are peaked around 0, whereas the black ones are shifted, with maxima around 0.01–0.02 and 0.02–0.03 for SPC and RTV, respectively. These two sites have higher dispersion (FWHM = 0.03), and approximately 30 % of the data are included in the interval −0.01–0.01 for the red and blue histograms at both sites, which is probably due to the different aerosol load affecting the different ALICEnet stations. As pointed out by the low value of the average AOT computed at ASC for the analyzed dataset (〈AOT〉=0.027), low pollution levels generally characterize this site, with some exceptions due to wind-driven aerosol transport from the nearby Po Valley (Diémoz et al., 2014, 2018a, b). Conversely, RTV (〈AOT〉=0.044) and especially SPC in the Po Valley (〈AOT〉=0.076) are characterized by higher aerosol content and pollution levels, which explain the larger histogram dispersions. Note that the high frequency of fog events in winter markedly reduces the number of analyzed AOT pairs at the SPC site, while some desert-dust-affected days at both SPC (e.g., Bucci et al., 2018) and RTV (e.g., Barnaba et al., 2017) were removed from our datasets (no desert-dust-affected dates in ASC).

Table 9 summarizes the long-term performances of the model-based procedure in deriving quantitative AOT from the ALC systems at the three investigated sites. It includes values of the average differences between the ALC-derived and sun-photometer-measured AOT (both bias 〈dAOT〉, and absolute difference $\langle \left|\mathrm{dAOT}\right|\rangle$, with associated standard deviations) obtained using both the proposed model-based approach and the fixed-LR inversions. For the SPC and RTV sites, these numbers show that the best ALC–photometer accordance is reached when employing either the model-based or the fixed-LR = 38 sr inversion scheme. In fact, these two approaches have similar performances in terms of mean dAOT values ($\langle \left|\mathrm{dAOT}\right|\rangle =\mathrm{0.011}$, 0.013 and 0.013, 0.014 for SPC and RTV, respectively), mean percent error ($\langle \left|\mathrm{dAOT}\right|\rangle /\langle \mathrm{AOT}\rangle =\mathrm{0.16}$, 0.19 and 0.31, 0.33), and a very low mean relative bias ($\langle \mathrm{dAOT}\rangle /\langle \mathrm{AOT}\rangle =-\mathrm{0.043}$, 0.005 and 0.088, 0.11). Conversely, the fixed-LR = 52 sr retrieval produces an overestimation of AOT in both SPC and RTV ($\langle \mathrm{dAOT}\rangle /\langle \mathrm{AOT}\rangle =\mathrm{0.33}$ and 0.44), with larger discrepancies between retrieved and observed AOTs ($\langle \left|\mathrm{dAOT}\right|\rangle =\mathrm{0.021}$ and 0.026, $\langle \left|\mathrm{dAOT}\right|\rangle /\langle \mathrm{AOT}\rangle =\mathrm{0.38}$ and 0.49). For the ASC site, due to the low aerosol content, the differences among the inversion schemes are almost negligible.

Overall, for the three sites, the statistics over the long-term datasets employed showed good results of the model-based approach with similar behavior of the retrievals with a fixed LR of 38 sr, while a fixed LR value of 52 sr produces an overestimation of the AOT at SPC and RTV. As different sites have different (and not known a priori) characteristic LR values, these results highlight the potential of the model-based approach to derive quite accurate β_a and α_a coefficients without the need to choose and fix an arbitrary LR value.

3.2.2 Model-based retrieval of aerosol volume (and mass)

In this section we provide examples of the applicability of the proposed approach to derive air-quality-relevant parameters. In particular, we use the ALC, β_a-retrieved data, and seventh-order polynomial fit linking β_a (at λ=1064 nm) to V_a (see also Table 3 and Fig. 2c) to derive the aerosol volume (and mass).

The ALC V_a estimates were first compared to aerosol volume derived in situ at the ASC site by two different optical particle counters (OPCs) on 29 December 2016 and 5 September 2017. For the case on 29 December 2016, a TSI optical particle sizer (OPS) 3330 was employed. This instrument has 16 channels that can be programmed to provide the number concentration at different (and logarithmically spaced) diameter size ranges within the interval 0.3–10 µm. Further details can be found in the TSI manual (2011). For the case on 5 September 2017, the Fidas^® 200s OPC was used. This spectrometer is able to retrieve high-resolution particle spectra (size measurements between 0.18 and 18 µm, with 32 channels per decade; Pletscher et al., 2016). For both dates, Fig. 8 shows the time (x axis, 24 h) vs. height (left y axis) contour plots of the ALC-based retrieval of the aerosol volume concentration (cm³ cm⁻³). The OPC-derived aerosol volume concentration measured at ground level is reported as a function of time (x axis) on the right y axis (grey curve). The corresponding ALC-derived volume concentration (integrating the ALC data between 0 and 75 m) is shown by a pink curve (same right y axis). Daily mean volume concentration values derived by OPCs and by ALC are also plotted (grey cross and pink triangle, respectively). The horizontal bar in the upper part of the figure indicates the ranges of RH measured at ground level during the analyzed cases.

Figure 8Time–height cross section of the aerosol volume concentration at Saint-Christophe in the Aosta Valley for 29 December 2016 (a) and 5 September 2017 (b). The right y axis reports the volume concentration measured at the surface through TSI and Fidas^® 200s OPCs (a, b, grey curves) and the ALC-derived volume concentration at 75 m (pink curves). The grey crosses and the pink triangles refer to the daily mean aerosol volume value derived by OPC and ALC measurements, respectively. The horizontal bars in the upper part of the panels indicate the ranges (RH<60 %, $\mathrm{60}\mathit{\%}<\mathrm{RH}<\mathrm{90}\mathit{\%}$, and RH>90 %) of the measured in situ RH during the analyzed days.

Download

The OPC-to-ALC comparison is certainly affected by intrinsic factors, such as differences in the atmospheric layer sampled (at ground and integrated between 0 and 75 m, for OPC and ALC, respectively) and in the probing methods (in situ and remote sensing, dried air sampled by OPC and ambient conditions sampled by the ALC). Furthermore, as mentioned in Sect. 4.2.1, a major critical issue of ALC retrievals at low levels is the correction for the overlap function, which needs to be experimentally characterized and verified for each instrument.

These issues are visible in the given examples of Fig. 8. In fact, in Fig. 8a, the agreement between the ALC-derived and the TSI OPC aerosol V_a values is good between 00:00 and 07:00 UTC. In the following hours both instruments register an increase in the aerosol volume, although with some discrepancies in absolute values. Starting from 18:00 UTC, the ALC derives an aerosol volume concentration higher than the OPC concentration by a factor of 3–3.5. This disagreement could be related to both the presence/arrival of fine particles (<0.3 m) not measured by the optical counter (see for example Diémoz et al., 2018a), or to aerosol hygroscopic effects (increase in volume associated with hygroscopic growth seen by the ALC but not by the OPC that dries the air samples). This latter effect is confirmed by the large RH values (RH > 90 %) measured after 18:00 UTC. Figure 8b shows a good agreement between the ALC-derived and Fidas OPC V_a values, in particular until 04:00 UTC and after 16:00 UTC. Some differences emerge around 07:00 UTC and between 11:00 and 15:00 UTC, when the ALC volume is lower by a factor of 2 compared to the in situ Fidas V_a values. The smaller minimum detectable size of the Fidas OPC instrument with respect to the OPS is likely the reason for the better agreement between ALC and OPC V_a values on this test date. In this case, the effect of RH seems to be less important, and indeed RH values remain lower than 90 %.

In general, high RH values (RH ≥90 %) are known to markedly affect the aerosol mass estimation from remote-sensing techniques and its relationship with reference PM_2.5 or PM₁₀ measurement methods, usually performed in dried conditions (e. g. Barnaba et al., 2010; Adam et al., 2012; Li et al., 2016, 2017). This theme is also discussed in Diémoz et al. (2018a) for the ALC measurement site of Fig. 8. Nevertheless, even with the mentioned limitations, results in Fig. 8 show the potential of the developed method in providing sound values of aerosol volume, and hence mass, in average-RH regimes well, giving support to more standard PM₁₀ air quality monitoring.

Figure 9Daily-resolved aerosol mass concentration at SPC, for the period June–July 2012, estimated from ALC-derived aerosol volume data at 225 m a.s.l. converted into mass using a fixed particle density ρ_a=2 g cm⁻³ (blue dotted line) and a variable ρ_a between 1.5 and 2.5 g cm⁻³ (shaded blue area). The red solid line is the daily PM₁₀ concentration as measured by the local air quality agency (ARPA). Vertical yellow shaded stripes indicate the presence of dust events.

Download

To give a further example in this direction, the model-assisted retrievals of aerosol mass over a longer time period were used to derive daily-mean aerosol mass concentrations (PM₁₀, a measurement typical of air quality stations). For this purpose, for the 2-month period June–July 2012, we derived daily mean values of aerosol volume at the SPC site using the functional relationships V_a=V_a(β_a) and then converted these into mass (PM₁₀) using typical values of aerosol densities (ρ_a). Results are shown in Fig. 9. It compares the daily average PM₁₀ concentration measured in situ at SPC by the Italian Regional Environmental Protection Agency (ARPA, red solid curve) and the model-assisted, ALC-derived daily mass concentration obtained assuming both a fixed particle density ρ_a=2 g cm⁻³ (blue dotted curve) and a range of particle density (1.5–2.5 g cm⁻³, shaded area), this range covering approximately the typical ρ_a values at the SPC site. Yellow shaded areas indicate the presence of dust events (e.g., Bucci et al., 2018) that are excluded from the results reported in the next paragraph.

More in detail, the daily-mean, ALC-derived mass concentrations were estimated in two steps: (1) estimation of hourly mass values for the selected height and (2) computation of the daily values through the median of the hourly values. To guarantee a good daily representativeness, the second step is applied only to those days on which at least 50 % of the hourly values are available in all the following temporal ranges: 00:00–05:00, 06:00–11:00, 12:00–17:00, and 18:00–23:00 UTC. Note that, due to the uncertainties associated with the O(z) in the first hundreds of meters (as previously mentioned, the ALC system at SPC has an old firmware, and its overlap function is not optimally characterized), we used the 225 m level as more trustworthy to estimate ALC mass concentration. Conversely, during the considered period of the year (i.e., June and July), the comparison to ground-level PM₁₀ at SPC is expected to be only slightly affected by this height difference, particularly in daytime, due to the strong convection within the mixing layer. A possible exception could be in nocturnal conditions when vertical gradients in the lowermost hundreds of meters can occur. However, our statistical (3 year) ALC records show the mixing layer height at SPC to descend below 250 m only 4–5 h per day in July (usually between 22:00 and 03:00 UTC, i.e., when emissions are at a minimum). Overall, Fig. 9 confirms a good agreement between the ALC-derived and the ARPA reference PM₁₀ values, with a correlation coefficient (R) of 0.64. In fact, mean, absolute mean, and relative differences between the two series are $\langle \mathrm{d}{\mathrm{PM}}_{\mathrm{10}}\rangle =\mathrm{2.3}\pm \mathrm{6.0}$ g cm⁻³, $\langle \left|\mathrm{d}{\mathrm{PM}}_{\mathrm{10}}\right|\rangle =\mathrm{4.8}\pm \mathrm{4.3}$ g cm⁻³, and $\langle (\mathrm{d}{\mathrm{PM}}_{\mathrm{10}}/{\mathrm{PM}}_{\mathrm{10}})\rangle =\mathrm{0.14}\pm \mathrm{0.27}$. This agreement attests that the SPC site can indeed be considered an average continental site and suggests the potential of this approach to derive information on aerosol volume and mass. Still, due to the specificity of each site and to the limited period considered here, these results cannot be taken as representative of all continental sites at all times. Further studies at different places and over longer time periods would be necessary to better assess the uncertainty of the proposed retrieval, including uncertainties due to the variability in continental conditions (in terms of particle size distribution, compositions, hygroscopic effects, etc.) but also in the instrument-dependent performances (e.g., overlap corrections).