Simultaneous observations of aerosol–cloud–albedo interactions with three stacked unmanned aerial vehicles

Abstract

Aerosol impacts on climate change are still poorly understood, in part, because the few observations and methods for detecting their effects are not well established. For the first time, the enhancement in cloud albedo is directly measured on a cloud-by-cloud basis and linked to increasing aerosol concentrations by using multiple autonomous unmanned aerial vehicles to simultaneously observe the cloud microphysics, vertical aerosol distribution, and associated solar radiative fluxes. In the presence of long-range transport of dust and anthropogenic pollution, the trade cumuli have higher droplet concentrations and are on average brighter. Our observations suggest a higher sensitivity of radiative forcing by trade cumuli to increases in cloud droplet concentrations than previously reported owing to a constrained droplet radius such that increases in droplet concentrations also increase cloud liquid water content. This aerosol-cloud forcing efficiency is as much as −60 W m⁻² per 100% percent cloud fraction for a doubling of droplet concentrations and associated increase of liquid water content. Finally, we develop a strategy for detecting aerosol–cloud interactions based on a nondimensional scaling analysis that relates the contribution of single clouds to albedo measurements and illustrates the significance of characterizing cloud morphology in resolving radiometric measurements. This study demonstrates that aerosol–cloud–albedo interactions can be directly observed by simultaneous observations below, in, and above the clouds.

Since the introduction of the Intergovernmental Panel on Climate Change (IPCC) nearly two decades ago, the uncertainty of the impact of anthropogenic aerosols has remained the least constrained radiative forcing component in understanding global climate change. Aerosols are generally considered to exert a cooling effect at the earth's surface by directly scattering incoming solar radiation back to space (1) or influencing cloud properties such as enhancing cloud albedo (2) and lifetime (3). For the first time, the IPCC 2007 assessment (4) provides a best estimate for the aerosol direct and indirect effects (−0.5 ± 0.4 and −0.7_−1.1^+0.4 W m⁻² globally, respectively); however, the magnitude of the uncertainty is still more than a factor of 2 greater than those estimates and is still greater than the sum of uncertainties of all other components contributing to climate forcing (i.e., greenhouse gases, ozone, surface albedo, and solar irradiance).

The effect of aerosols on cloud microphysical and radiometric properties, known as the “indirect effect,” consists of a two-part challenge—(i) response of cloud droplet number, n_D, and cloud liquid water path to a change in aerosols and (ii) response of cloud albedo, α_c, caused by changes in cloud microphysical properties. Aerosols are linked to cloud droplet formation by their physicochemical properties (5) and cloud dynamics such as updraft velocity (6), entrainment (7), and precipitation (3); whereas cloud radiative properties are determined by the number and size distribution of droplets and the cloud's geometric thickness (2). Numerous studies have demonstrated that an increase in aerosols increases n_D (8–12); however, fewer have actually measured the impact of aerosols on α_c (13, 14). Several studies have not detected the expected effect of aerosols on clouds, possibly because of variations in liquid water path (15–17), and employ aerosol and cloud properties in radiative transfer models to estimate radiative forcing.

The two analytical expressions given below illustrate the dependence of cloud optical depth, τ_c, and cloud albedo, α_c, on the cloud liquid water content, lwc; the effective radius, R_eff; the height of the cloud, h; the density of water, ρ_w; and an asymmetry parameter, g, for scattering of radiation by clouds.

Eq. 2 is a highly simplified expression that is derived by employing the two-stream approximation (18) to the radiative transfer equation and by assuming that the drops do not absorb solar radiation. Neither of the two simplified expressions nor the assumptions used to derive them are used in sophisticated modeling studies, but are written here to convey the essential physics of the aerosol–cloud interactions.

To this end, the Maldives AUAV Campaign (MAC) demonstrates the ability to observe aerosol and cloud microphysical properties and solar radiation fluxes simultaneously by using emerging technology (i.e., autonomous unmanned aerial vehicles, AUAVs) as new tools in atmospheric sciences (19). The MAC results presented here demonstrate that it is possible to directly observe the aerosol–cloud–albedo interactions and establish the link between aerosols and clouds—and ultimately, between changes in aerosols and subsequent changes in cloud radiative properties. We also compare the relative contribution of the aerosol direct and indirect effects, deduced directly from aircraft measurements to demonstrate the capabilities of a new observing system for addressing long-standing problems in atmospheric sciences.

Results

Part I: Aerosol–Cloud Relationship.

MAC is characterized by the arrival of a persistent, elevated aerosol layer to an already polluted boundary layer and is categorized into two periods based on vertical profiles of aerosols and key atmospheric state parameters (Fig. 1; Table 1). During period I of the experiment, aerosol transport is mostly confined to the boundary layer (<1,000 m above sea level, masl), whereas period II is characterized by an aerosol layer that peaks at 2,000 masl (≈2,500 cm⁻³). Aerosol concentrations, N_CN, at the surface are relatively high in both periods (1,000–1,250 cm⁻³) compared with clean marine conditions of a few hundred cm⁻³ (20). Wind fields above the boundary layer (850 mbar) change from an easterly flow south of the Maldives associated with the Intertropical Convergence Zone (ITCZ) to a northwesterly flow that carries long-range transport of pollution to the Maldives from the Arabian Peninsula and South Asia. Although N_CN in the boundary layer does not significantly change on the arrival of the elevated layer, the new aerosol (presumably aged anthropogenic emissions and dust) causes a shift in the size distribution at the surface and aloft (21), and doubles the accumulation mode diameter (Fig. 2a). Such a shift in aerosol size distributions affects the subset of aerosols that serve as seeds for cloud formation, i.e., cloud condensation nuclei (CCN). Indeed, the CCN spectra change, especially at low S_c, as concentrations at 0.1% S_c, N_{CCN_0.1}, and increase by 60% in period II (Fig. 2b; Table 1). Activation indices, the ratio of the observed critical diameter to that of a soluble salt (22), were close to unity at 0.1% S_c in both periods, indicating that activated particles contain a large soluble fraction. The CCN spectra approach the soluble limit at lower supersaturations (Fig. 2b) implying that chemical composition changes as a function of size probably because of cloud processing or contributions from sea salt.

Fig. 1.

Vertical profiles of total aerosol concentration (a), relative humidity (b), temperature (c), and equivalent potential temperature (d) are shown for each period. Thick lines show average profiles and shading designates 1-σ variability. Horizontal dashed lines show average sampling altitudes of the in-cloud AUAV. Lifting condensation levels (LCLs) are shown by plus (period I) and square (period II) markers on left axes.

Table 1.

Average aerosol and cloud properties for each sample period.

Period	I	II
Dates	March 6–19	March 20–28
Back trajectory (3,000 masl)	Marine recirculation	Arabian Peninsula and India
Flight	1, 4, 5, 7	9, 10, 11, 12, 13, 14
Distance, km	1,125	1,690
Cloud fraction, %	6.7 ± 2.3	5.0 ± 2.0
nD, cm−3	205 ± 124 [186 ± 117]	284 ± 150 [359 ± 189]
nD,max, cm−3	411 ± 265 [361 ± 249]	551 ± 251 [665 ± 356]
Reff, μm	6.74 ± 0.62 [7.0 ± 0.48]	6.45 ± 0.46 [6.56 ± 0.45]
Cloud events	144 [23]	96 [34]
Cloud width, m	423 ± 333 [295 ± 199]	401 ± 277 [299 ± 143]
lwc, g·m−3	0.26 ± 0.19 [0.25 ± 0.18]	0.31 ± 0.20 [0.41 ± 0.25]
NCN, cm−3	1269 ± 622	1255 ± 258
NCCN_0.1, cm−3	502 ± 113	812 ± 54
α (clear-sky; 3,000 masl)	0.067 ± 0.003	0.079 ± 0.006

Variations reported as 1-σ. In-cloud transit distance determines cloud width. Brackets denote stacked flight averages used in analysis of Fig. 5.

Fig. 2.

Aerosol size distributions (a) and CCN spectra (b) at MCOH during periods I and II. Calculated CCN spectra for soluble [i.e., (NH₄)₂SO₄] and insoluble limits based on aerosol size distributions in a are also shown in b. D_p is the dry diameter.

Although the indirect effect relates change in cloud albedo to the input of aerosol, the first challenge is to establish the connection between cloud microphysical properties and aerosol number concentrations, in particular, CCN. Vertical profiles of aerosols and comparisons with ground observations at MCOH (Fig. 1) indicate a well mixed boundary layer, which allows a reasonable comparison between ground-based CCN measurements and cloud observations (23, 24). The MAC observations show a ≈40% increase in n_D during period II, which is inextricably linked to the change in aerosol properties (25) (i.e., shift in size distribution) and enhancement in N_{CCN_0.1} (Fig. 3). N_{CCN_0.1} correlate to maximum cloud droplet concentrations, n_D,max (r² = 0.62; Fig. 3); however, the correlation is stronger during period I (r² = 0.89) when a N_{CCN_0.1}:n_D,max ratio of 0.86 ± 0.09 indicates peak in-cloud S_c close to 0.1%. Changes in aerosol properties increase n_D in period II, and the N_{CCN_0.1}:n_D,max ratio decreases (0.63 ± 0.09), which is indicative of lower in-cloud S_c. The marked difference in the N_{CCN_0.1}:n_D,max relationship between the periods suggests more competition for water vapor during period II either from enhancement of CCN that activate at lower S_c, changes in vertical velocity, or entrainment of drier air from the aerosol layer above the clouds.

Fig. 3.

Relationship between flight rms-averaged CCN concentrations (at 0.1% S_c; N_{CCN_0.1}) and maximum in-cloud droplet concentrations (n_D,max) for all cloud flights. Diamonds indicate simulated N_{CCN_0.1} in period II based on aerosol size distributions and assumed chemistry. The dashed line represents the 1:1 correlation.

Part II: Cloud–Albedo Relationship.

The results in part I establish the link between aerosols (particularly CCN) and n_D; yet, the second and more difficult part in assessing aerosol–cloud interactions is relating changes in microphysical properties to changes in cloud albedo, α_c—a central requirement for understanding the aerosol indirect effect. Through coordinated, stacked flights using three AUAVs—one sampling at cloud level and the other two above the clouds—radiometric and microphysical properties of individual clouds were simultaneously observed. We use a detailed analysis of a single cloud [Fig. 4 and supporting information (SI) Fig. S1] to illustrate the simultaneous α_c response at different altitudes. In addressing the effects of aerosols on cloud optical properties, this analysis introduces a scaling function (see SI Text, Eq. S7, Fig. S2), which demonstrates the importance of cloud morphology in interpreting albedo measurements.

Fig. 4.

Distance along flight track showing the simultaneous measurements of cloud microphysical and radiometric properties from three AUAVs at different altitudes. The in-cloud measurements (n_D, R_eff, and droplet distribution) are shown in Bottom. The observed broadband albedo (0.3–2.8 μm) is shown by the red line in Top and Middle. Model response in Top are calculated based on characteristic length, w_*, and cloud albedo, α_c = 0.826—black and green dashed lines from 1,200 masl cloud top and 600 and 400 m cloud width, respectively; blue dashed line from 1,300 masl cloud top and 400 m cloud width. The shaded blue region illustrates Δα_c ± 0.1. Thin horizontal dashed lines show the minimum albedo during the leg. Altitudes for the AUAVs are shown in each profile. The red marker in Bottom shows the location of the AUAV for the image in Fig. S1.

In the case study (Fig. 4), the in-cloud AUAV (at 796 masl) flew through a cloud and measured an average n_D and effective radius, R_eff, of 372 ± 236 cm⁻³ and 6.8 ± 0.6 μm, respectively. The average lwc is calculated to be 0.49 ± 0.34 g·m⁻³. At the same time, both AUAVs above the cloud (at 1,625 and 3,250 masl) captured the radiometric response of the same cloud element. The up-welling broadband pyranometers on the middle and upper AUAVs measured a net increase of 183 ± 3 W m⁻² and 16 ± 1 W m⁻², respectively, which translates to an maximum increase in measured albedo, α_m, of 0.168 ± 0.002 and 0.010 ± 0.003 compared with the cloud-free background. Most of the increase in α_m is because of the cloud directly below the AUAVs; however, the ocean, islands, and other clouds in the region also contribute. A camera, mounted on the in-cloud AUAV illustrates the complicated 3D structure (Fig. S1) and shows several smaller clouds in the immediate vicinity of the main cloud. By using the onboard image, we estimate the cloud top and width to be ≈1,250 masl and 500 m and use this spatial information to simulate the expected albedo response of both above-cloud AUAVs based on the scaling relationship (see Discussion). The cloud optical depth (τ_c = 70 ± 49) is estimated from Eq. 1 (h = 650 ± 50 m assuming vertically homogeneous lwc), which yields α_c = 0.826 ± 0.10 (Eq. 2) by using an asymmetry parameter (g = 0.865) from Indian Ocean Experiment (INDOEX) observations of trade cu (26). Although we cannot conduct an exact closure between cloud properties and radiometric response (precise measurements of the 3D structure are needed), a simple pixel plane-parallel radiation model shows relatively good agreement with the measured enhancement in albedo at both altitudes. The different simulations in Fig. 4 show the expected albedo response for best estimates of cloud height and width—1,200 masl cloud top and 600 m cloud width correlates well with both AUAV observations—and compare changes in spatial relationships (i.e., cloud morphology) to changes in cloud optical properties (i.e., an enhancement in α_c by 0.2). The main discrepancy between measured and simulated values of the radiometric response is caused by the uncertainty in the cloud morphology. These results emphasize the complexity in detecting the indirect effect and show that aerosol–cloud effects can easily be masked by uncharacterized differences in cloud morphology.

By using the measurements for all stacked flights, data from 41 individual trade cu were analyzed to determine the effect of n_D (and lwc) on α_c at 1,600 masl (Fig. 5). The analysis only includes cases when the down-welling flux was constant (i.e., no clouds overhead), clouds were separated by at least 60 s (to avoid influence from neighboring clouds), and solar zenith angles <30° (to minimize uncertainties in zenith angle corrections). Despite variability in cloud geometric thickness, we observe an enhancement in the maximum observed albedo above a cloud, α_m,pk, with increasing n_D and lwc (r² > 0.468; Fig. 5). Observations of these 41 samples show nearly a twofold increase in average n_D and lwc (Table 1; values in brackets) and resulting Δα_m,pk = 0.01 (0.0199 ± 0.003 to 0.0303 ± 0.006 rms in periods I and II). There is little correlation between α_m,pk and cloud width (r² = 0.093; derived from in-cloud transit distance) and the influence of neighboring clouds is negligible because of the low cloud fraction.

Fig. 5.

Relationship between maximum measured albedo, α_m,pk, at 1,600 masl for 41 clouds and simultaneous observations of average n_D and lwc. The large markers with a center black dot denote the period averages in brackets in Table 1. The single anomalous point in period II (α_m,pk ≈ 0.23) is not included in the average.

Based on differences of n_D and lwc for these 41 samples (Table 1; values in brackets), the expected change in cloud albedo or albedo susceptibility, Δα_c,

is ≈0.16 ± 0.03 (h ≈ 600 m and Δh h⁻¹ = 0; see SI Text, Eq. S13 for derivation).

Because the MAC observations were made several hundred meters above the cloud tops, relating the observed Δα_m,pk = 0.01 (Fig. 5) to Δα_c requires information about the size of the cloud (including 3D information) and its distance to the above-cloud AUAV. The nondimensional change in albedo (Eq. 5) for these 41 samples, α_* = Δα_m,pk Δα_c⁻¹ = 0.01/0.16 = 0.063 ± 0.044, corresponds to measurements 600 m above cloud top (between 400 and 1,000 m when accounting for the uncertainty), which is in general agreement with experimental conditions (Fig. 1).

Studies of shiptracks (27) have shown satisfactory agreement with Δα_c and report a weak correlation between lwc and n_D which removes the 2Δlwc/lwc⁻¹ and 3Δh/h term in Eq. 3. Nonetheless, Platnick and Twomey (28) point out an enhanced susceptibility when an increase in n_D is also accompanied by an increase in lwc. Our results suggest trade cu α_c is about a factor of 3 more susceptible to changes in n_D than previous estimates because of the strong correlation between n_D and lwc. In addition, simulations have suggested that h may increase or decrease with n_D (29, 30). However, the Δα_c effect of a 15% change in h (29) is still seven times less than the doubling of n_D for MAC conditions.

Cloud Properties.

Trade cu sampled during MAC are similar to nonprecipitating clouds observed during INDOEX (31) categorized as “transitional” and “polluted” with R_eff < 10 μm and cloud widths of several hundred meters (20). However, an interesting distinction of trade cu observed during MAC is the strong correlation between n_D and lwc (r² = 0.85; Fig. 6); which constrains the variability in R_eff. R_eff is independent of n_D (r² = 0.09) and cloud width (r² = 0.01). Similar results (32, 33) point to inhomogeneous mixing where evaporation timescales are much shorter than those of turbulent mixing allowing some of the droplets to be completely evaporated (34, 35). Vertical profiles of trade cu show subadiabatic lwc (Fig. 7), where a slight increase in R_eff is negated by a decrease in n_D. The linear trend between n_D and lwc is consistent throughout MAC, supporting conclusions that changes in aerosols dominate the variability in τ_c for a similar thermodynamic environment.

Fig. 6.

Relationships between n_D and lwc for all clouds sampled during MAC. Dashed lines represent robust fits for each period. Green and red markers represent periods I and II, respectivley. The large markers with a center black dot illustrate the period averages.

Fig. 7.

Vertical profiles of n_D (Left), R_eff (Center), and lwc (Right) of trade cu during flight 18. The thin dashed line illustrates the expected lwc profile for an adiabatic cloud, lwc_ad.

In period II, there is more water vapor available for condensation and a slightly less stable boundary layer, which is consistent with Table 1 and the lower lifting condensation level (LCL) in Fig. 1. Nonetheless, MAC observations show a 15% lower lwc for the same n_D in period II because of a slight decrease in R_eff. Vertical profiles of relative humidity (Fig. 1b) show drier air above the clouds in period II, consistent with simulations of trade cu that entrainment of dry air above the clouds exerts a strong influence on lwc (36).

Discussion

Scaling Analysis.

Discerning subtle changes in α_c from surface heterogeneities and surrounding clouds requires an understanding of the spatial scales of the earth's surface and clouds relative to the height of the measurement. Sensors with a cosine response (i.e., pyranometers) receive a greater contribution of their signal from objects that are closer and normal to the sensor as demonstrated by the two AUAVs flying above the cloud at different altitudes in Fig. 4. The observed response, α_m, is related to a single cloud of width, W, albedo, α_c, and its distance below the AUAV, z. The nondimensional response in albedo, α_*,

is calculated for a cloud projection that is plane-parallel and directly below the sensor. By integrating and substituting a nondimensional length, w_* = z W⁻¹, the scaling relationship becomes

and provides the sampling constraints to detect relatively small changes related to aerosol indirect effects (Fig. S3). For example, the AUAVs in the above study (Fig. 4) sampled at characteristics lengths, w_* ≈ 0.75 (1,625 masl) and 4.9 (3,250 masl), which translate to instrument responses of 17% and 0.5%, respectively (Eq. 5) of the total change in α_c of the cloud below. Although the albedo measurements for both AUAVs show variations in the baseline up to ≈0.005 peak-to-peak, the maximum difference in α_m induced by Δα_c = 0.2 is 0.03 and 0.001 (i.e., 0.2 × 17% and 0.2 × 0.5%) for the middle and top AUAVs, respectively. The blue-shaded region in Fig. 4 denotes the expected changes in α_m related to aerosol–cloud interactions. Δα_c = 0.2 corresponds to a factor of 2 increase in n_D and lwc (Table 1; Eq. 3) between periods I and II. This case study shows that albedo measurements are sensitive to cloud morphology, particularly when the cloud element occupies a small fraction of the pyranometer's field of view. In addition, the variability in cloud properties (even those not along the flight track) needs to be fully characterized to quantify the contribution of each cloud element to α_m.

As cloud fraction increases, cloud–cloud radiative effects (37) may enhance the net reflected flux at intermediate cloud fractions. Contribution from neighboring clouds certainly influences the measured pyranometer signals as shown in the elevated signal after the main cloud in Fig. 4. Therefore, extrapolating the MAC results of trade cu to cloud systems with larger spatial scales is not so straightforward. Formation flights with manned aircraft over stratocumuli, SCu, in the Pacific Ocean have been conducted (38) with unexplained discrepancies between simulated and measured cloud albedos. In that experiment, a second aircraft flew 1 km above clouds. At this height above the clouds, 50% of the signal comes from a 3,000-m diameter region making it difficult to relate changes in α_m to variations in the cloud field. Simulations of unbroken marine SCu have shown that horizontal variability of α_c can vary by more than a factor of two because of changes in τ_c (39, 40). Because the horizontal variability scales to roughly twice the thickness of the cloud deck, nonlinear variability in α_c of surrounding cloud cells may mask the aerosol–cloud–albedo interactions on the larger scale of the cloud system.

Aerosol-Cloud Effect.

The magnitude of the aerosol effect on cloud albedo, ΔF_{aerosol-cloud}, from period I to II is estimated by

where F_o is the incident downward flux (incoming TOA diurnal average: ≈425 W m⁻²), T_a is the transmittance through the atmosphere at 1,600 masl (T_a = 0.925), f_c is the cloud fraction during period II (f_c = 0.05), and Δα_c = 0.05–0.16 (Table 1, values outside and in brackets). The observations measure an aerosol-cloud effect of −1 to −2.9 ± 0.7 W m⁻². The slight decrease in cloud fraction (from 6.7% to 5%) changes the forcing by ≈1 W m⁻²; however, this decrease cannot be attributed to an aerosol effect and has been excluded to isolate the aerosol effect on cloud optical properties. By normalizing with respect to cloud fraction, the diurnal average top-of-atmosphere aerosol-cloud forcing efficiency is as much as −60 W m⁻² per 100% cloud fraction for a doubling in n_D and associated increase in lwc. In comparison, clear-sky albedos increased from period I to period II because of the elevated pollution layer (Table 1)—the direct top-of-atmosphere effect of this elevated aerosol layer is −4.3 ± 2.5 W m⁻². These findings provide direct observations of the aerosol direct and indirect effects and a basis for reducing the uncertainty of anthropogenic aerosols as the least constrained radiative forcing component in understanding global climate change.

Materials and Methods.

The MAC experiment was conducted from March 6 to 31, 2006, to observe long-range transport of dust and anthropogenic pollution from the Arabian Peninsula and South Asia (19). The observing system consists of three AUAVs in a synchronous vertical formation with miniaturized instruments for measuring aerosols, clouds, and radiometric fluxes (19). The stacked flights are programmed to autonomously sample the same vertical column below, in, and above the clouds to study how the boundary layer aerosols feeding the trade wind cumuli modify cloud microphysical and radiative properties. Vertical stacking was performed autonomously between 500 and 3,000 m and generally maintained <300 m horizontal separation, which translates to <10-s latency between the aircraft. See Movie S1 for a demonstration of the coordinated flight. The Maldives Climate Observatory on Hanimaadhoo Island (MCOH; 6.77°N; 73.18°E), a long-term monitoring station for the Atmospheric Brown Cloud (ABC) project (41, 42), serves to complement and validate the airborne measurements. In addition, a streamwise thermal-gradient CCN instrument (43) measured CCN spectra between 0.1% and 0.6% supersaturation, S_c, at MCOH.

Aerosol, cloud, radiometric instruments, and an integrated data acquisition system were deployed with a total payload weight and power <5 kg and 30 W, respectively. The AUAV payload is mission-specific and outfitted to perform a particular set of atmospheric measurements. Much of the instrument suite includes commercially available instruments that have been repackaged or redesigned to minimize weight and volume. These measurements include aerosol number and size distribution, aerosol absorption, cloud drop size distribution, solar radiation fluxes (visible and broadband), temperature, pressure, and relative humidity. A shrouded aerosol inlet was designed to minimize aerosol sample biases. The data collected during MAC has been validated by using standard calibration routines in conjunction with comparisons to ground-based instruments in both laboratory and in situ settings.

The in-cloud platform measures cloud droplet number and size distributions of trade cu—and a real-time video camera assists in targeting the clouds. The aerosol-radiometric AUAVs (i.e., above and below cloud) measure aerosol concentrations, size distributions, and visible and near-IR absorption coefficients as well as broadband (0.3–2.8 μm) and visible (0.4–0.7 μm) albedo. Occasionally, the “below-cloud” AUAV flew above the clouds such that two AUAVs measure cloud albedo at different altitudes simultaneously. Radiometric data are corrected for pitch and roll and cloud albedos are normalized based on a clear sky model to remove solar zenith angle dependence (44). Only cases when the solar zenith angle is <30° are used for this study to minimize 3D effects of trade cu on albedo measurements. Clear sky albedos were measured at 3,000 masl by averaging multiple consecutive legs where no clouds were sampled and the observed albedo did not change. Cloud events are counted when droplet concentrations, n_D, exceed a threshold (7 cm⁻³) for at least 3 s (i.e., cloud width ≥100 m).