Atmospheric aerosols can have important direct radiative impacts, and indirect impacts through their effect on cloud microphysical properties. The total effect of aerosol particles on the global radiation budget is negative (IPCC, 2007), and thus they are partly counteracting the warming effect caused by the increased concentrations of greenhouse gases. However, due to highly varying concentrations and composition of aerosols, and feedbacks from the climate system onto aerosols and clouds, the total effect on the global radiation budget is not well known. Aerosols can affect the number concentration of cloud droplets (Nd) and thereby the impact that clouds have on radiative transfer in the atmosphere. However, this interaction is complex and it produces the largest single uncertainty in the estimates of Earth's radiation budget (IPCC, 2007). The interaction between aerosol and cloud, and cloud formation is not a trivial process; it depends on many factors including local weather parameters along with the chemical, optical and dynamical properties of aerosol and cloud (McFiggans et al., 2006; Romakkaniemi et al., 2006; Reutter et al., 2009).
Cloud droplet number concentration during the formation of cloud droplets is determined by the aerosol size distribution and chemical composition, the amount of condensable gases, the temperature and the updraft velocity of the air parcel entering the cloud (McFiggans et al., 2006). The first two define how many particles can maximally act as cloud condensation nuclei (CCN) at some supersaturation, and the last two depend on the local meteorology. Theoretically, an increase in the aerosol concentration should lead to an increase in Nd, which has been observed in several direct measurements of aerosol–cloud interactions (ACI) (e.g. Snider et al., 2003; Komppula et al., 2005; Meskhidze et al., 2005; Portin et al., 2009). If Nd increases throughout a cloud whilst holding the liquid water path (LWP) constant, the effective radius Reff of cloud droplets decreases, and this increases cloud optical thickness (COT) and cloud albedo (Twomey, 1977). This effect has been termed the first aerosol indirect effect. Furthermore, changes in Nd affect also drizzle formation so that increased Nd suppresses rain formation efficiency and ultimately enhances cloud fraction (Albrecht, 1989). This process has been termed the second indirect effect. However, due to the complexity of the processes of cloud and drizzle formation, it is not always trivial to observe these different effects as there are several feedbacks involved in the cloud systems (Stevens and Feingold, 2009).
Beyond direct in situ measurements, it is also possible to use remote sensing tools, such as satellite instruments (e.g. Nakajima et al., 2001) or ground-based instruments as deployed in the Atmospheric Radiation Measurement (ARM) programme (e.g. Dong et al., 2008), to study the aerosol effect on clouds. The difficulty with estimating ACI remotely by satellite is a lack of coincident information of both the cloud and the aerosol residing below the cloud. Thus, some studies have used ground-based data for aerosol properties and satellite data for cloud properties (e.g. Boers et al., 2006; Lihavainen et al., 2010; Janssen et al., 2011), and in some cases satellite retrievals have been validated with airborne measurements (e.g. Schüller et al., 2003; Painemal and Zuidema, 2011). Quite often different measurement methods give different cloud responses to aerosol perturbation and several reasons have been proposed for this difference (Lihavainen et al., 2010; McComiskey and Feingold, 2012).
In this article, we focus on ACIs observed both in situ and remotely. We use the measurement data from Puijo station in Finland together with cloud retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) instruments onboard the Aqua and Terra satellites to assess the effect of aerosol concentration on Nd, COT and Reff. The relation between accumulation mode particle concentration (Nacc) from ground-based measurements and COT and Reff, from MODIS is studied to see how aerosol affects the optical and microphysical properties of low clouds. We also compare Nd and the aerosol–cloud interaction (ACI) (Feingold et al., 2001; McComiskey et al., 2009; Lihavainen et al., 2010) parameter values derived from satellite data to those from direct measurements. ACI parameters are a commonly used set of metrics that associate changes in Nd, Reff and COT to changes in aerosol burden, for example, Nacc or aerosol optical thickness. Here, we focus on the ACI that pertains to Nd. A significant amount of work has already been done regarding ACIs over ocean – for example, the VOCALS-Rex campaign (Wood et al., 2011) – but to our knowledge this is the first time that long-term ground-based aerosol and cloud droplet measurements have been compared to satellite retrievals for continental clouds.
The measurement station at Puijo (62°54′34′′N, 27°39′19′′E) is on top of an observation tower located near the city of Kuopio. The measurements are carried out at a height of 306 m above sea level and 224 m above the surrounding lake level. The regional terrain consists mainly of lakes, small hills, forests, urban areas and rural areas. Overall, the surroundings of the measurement station are quite homogeneous with regard to aerosol sources, with only a few point sources of aerosol emissions. Since 2005/06, the station has been instrumented for continuous measurements of aerosols, cloud droplets, weather parameters and trace gases (Leskinen et al., 2009; Portin et al., 2009). Below, we give a brief description of the instruments used in our study, also listed in Table 1.
The aerosol size distribution is measured with a twin Differential Mobility Particle Sizer (DMPS) system which has a measurement range from 7 to 800 nm in diameter. The device is operated with two different inlets, which allows interstitial particles and those inside cloud droplets to be distinguished. For the direct measurements of cloud droplet number concentration and their size distribution a Cloud Droplet Probe (CDP, Droplet Measurement Technologies) is used. The device has a measurement range of 3–50 µm in diameter. In the present work, we have used one-hour-averaged data for Nacc and Nd for a large area (1°×1°) analysis, and 5 minutes Nd averages for a smaller area (5×5 km) analysis. In the latter, the shorter averaging time is used to match the satellite observation time as well as possible. In the comparison of cloud properties to aerosol data, a one-hour averaging time is used to get reliable statistics from the aerosol measurements as the DMPS system used takes 12 minutes for a single distribution scan. These data are compared with 1°×1° satellite data since data at this resolution are readily available as MODIS-level three data. The elevation of cloud layers is detected with a Ceilometer (Vaisala CT25 K) located at the Savilahti research station less than 2 km from Puijo and 219 m below the Puijo station measurement level.
Cloud events, defined as when the station is surrounded by clouds with high enough liquid water content (LWC>0.02 gm−3) for at least an hour, occur most frequently at Puijo in October when the top of the Puijo tower is inside cloud on more than 40% of days. Since June 2006, there have been 450 cloud events with a total 3000 hours of data. This data set was screened to remove rain events, discontinuous clouds, icing conditions and warm days (which cause an overheating of the CDP). After careful screening we are finally left with 414 cloud hours with reliable CDP data.
In an earlier study, Portin et al. (2009) analysed the dependence of Nd on aerosol properties for the same region. They showed that, in the cloud typical to Puijo station, Nd is dependent on Nacc, which they defined to be particles larger than 100 nm in diameter. It was found that a proxy cloud droplet number concentration (Nd,p) that matched the observed Nd well can be calculated with the following relation:
Here the constants a and b were obtained by data fitting, with values of 30.13 and 0.36 respectively [these are updated values based on a larger data set than the one used by Portin et al. (2009)]. Using this equation we can estimate the droplet concentration of the clouds above the tower by assuming that the relation is representative for all boundary layer clouds observed close to the station.
We use collection 5.1 data from MODIS instruments onboard both the Aqua and Terra polar synchronous orbiting satellites. The primary quantities of interest were Reff and COT (MODIS ATBDT – cloud products; King et al., 1997; Platnick et al., 2003). We also used MODIS cloud-top temperature and cloud-top pressure data to ensure that the analysed clouds were low-level clouds. Further details on the data selection and filtering are provided in Section 3.
In this study all the quantities are averages over cloudy pixels only and therefore cloud fraction (CF) is 1. In eq. (2) k is the cube of the ratio between the volume mean radius (Rv) and the effective radius of droplets () with a given constant value of 0.67 (Martin et al., 1994) in our study, Q is the scattering efficiency whose value is approximately 2, and ρw is the density of water. For this calculation, it is assumed that LWC increases linearly with height from cloud base to cloud top (i.e. that the clouds are adiabatic). Cw is the ‘condensation rate’ for such a moist adiabatic ascent given by
Here qv and qL are the water vapour and liquid mixing ratios; Lv is latent heat of condensation of water; Cp is the specific heat capacity of dry air at constant pressure; and Γd (=g/Cp, where g is the gravitational acceleration) and Γ m are the dry and moist adiabatic lapse rates, respectively. Cw depends mainly on temperature (T), although there is also some weak pressure (P) dependence. This equation can be derived through consideration of the conservation of moist static energy during moist adiabatic ascent, that is,
Here Ra is the specific gas constant for dry air (=287.04 J kg−1 K−1). Then eq. (4) above can be obtained by dividing by dz, inserting the hydrostatic equation and recognising that the temperature change dT/dz will be the moist adiabatic lapse rate (Γm).
For the comparison of ground-based measurements and MODIS data, it is crucial that only a single layer of cloud exists. For some of our comparison single-layered cloud events were chosen using data from the Vaisala CT25 K Ceilometer and/or by limiting the MODIS-retrieved cloud-top pressure and temperature values. Using meteorological data from Puijo station and cloud-phase information from MODIS, we limited our study to non-precipitating low-level liquid clouds. A limit of satellite-derived Reff (3 µm ≤Reff≤30 µm) is also applied to make the selection of liquid-only cloud more reliable (Nakajima and King, 1990).
After screening out the possible ice phase and rain clouds, we further divided the satellite data used in this study into three categories:
The satellite-derived Reff and COT from the category III data have average values of 11.6±2.9 µm and 13.2±8 µm, respectively, for the observation period 2006–2011. These values are comparable to those presented by Janssen et al. (2011) derived over a 2°×2° area above the Hyytiälä measurement station 200 km southwest of Puijo station. Similar results can be expected because both areas are quite comparable in terms of aerosol sources. The average Nd calculated from eq. (2) is 171 cm−3. This value is higher than the range 40–100 cm−3 obtained by Janssen et al. (2011) and at the upper end of the range 45.2 cm−3 (clean pixels) – 132 cm−3 (polluted pixels) obtained by Sporre et al. (2011) for a cleaner area. One possible reason for the difference is the different method used to derive Nd, as Janssen et al. (2011) used models from Boers et al. (2006). We also tested the method used in Sporre et al. (2011) obtaining an average Nd of 156 cm−3. For comparison, the average Nd from the CDP data is 217 cm−3, which is quite close to the MODIS-measured Nd.
To study the dependence of Nd on aerosol particles we have analysed the number concentration (Nacc) of particles larger than 100 nm in diameter at the Puijo measurement station. We also tested different minimum sizes between 80 and 150 nm for the definition of Nacc particles, and found that the following results are insensitive to these choices. In Fig. 1 we compare Nacc to MODIS-retrieved Reff, COT and Nd for the category III data. The line indicates a running mean of 30 data points to make trends more readable for eye. We can see that at low values of Nacc both Reff and Nd are clearly dependent on Nacc. At high values this is not the case. The calculated Spearman correlation between MODIS-retrieved Nd and Nacc for values less than 400 cm−3 is 0.38, and the ACI parameter is 0.14 when calculated from the linear fit between Nacc values of 70 cm−3 and 400 cm−3. For larger Nacc, there is no correlation and the average Nd remains almost constant at around 215 cm−3. For COT, we found no dependence on Nacc, which is interesting given the observed dependence of Nd on Nacc since it would be expected that for a fixed adiabatic cloud thickness (H) COT would increase proportionally with Nd1/3. However, COT is also proportional to H5/3 in such an adiabatic cloud – that is, a cloud with the same assumptions as those used in deriving eqs. (2) and (3) – and thus is much more sensitive to changes in H than to changes in Nd. Such changes in H are likely to occur due to the influence of local meteorological changes, thus making it more difficult to see the effect of Nd and aerosol on COT in the rather noisy data.
Similarly to Fig. 1, in Fig. 2 we show how the in situ measured Nd from the CDP depends on the Nacc. In this case, the Spearman correlation is 0.63 for Nacc less than 400 cm−3. As expected, the correlation is higher in direct measurements as the cloud droplets and aerosols are measured in the same place. The ACI value calculated from these data was 0.16 (and 0.12 using eq. (1) with fitted parameters) on average for Nacc lower than 400 cm−3. Both ACI values are very similar to those reported by Garrett et al. (2004) who documented ACI values in the range of 0.13–0.19 using ground-based measurements of low-level liquid clouds observed near Barrow Alaska. Lihavainen et al. (2010) reported measurements conducted in a cleaner area in Northern Finland, where ACI values of 0.2 to 0.3 were obtained from ground-based measurements and approximately 0.1 from satellite measurements.
Both data sets show that Nd is only weakly, or not at all dependent on Nacc at high aerosol loadings. The reason for this finding is the complex dynamics of cloud formation. In the type of clouds observed at Puijo, the updraft velocities are quite low, and in such conditions Nd is typically only weakly dependent on aerosol particle concentration above some threshold concentration (Reutter et al., 2009; Romakkaniemi et al., 2012). When comparing the in situ and MODIS data sets, we can see some differences. For example, for the in situ measurements the running mean of Nd is still increasing at high aerosol loading, but this is not the case in the MODIS-derived Nd. This might be because for the in situ measurements there are no small Nd values with high aerosol loading, but in MODIS-derived data such data points do exist. As we are correlating local aerosol concentrations with cloud properties determined from a larger area, it is always possible that some local source might have increased the measured aerosol concentration above that of the observational area, or that some local aerosol are not effective as CCN.
In Fig. 3 we show Nd retrieved from MODIS as a function of that directly measured in the cloud (category I, red circles) or estimated from aerosol measurements at Puijo station (category II, blue circles). In the case of direct measurements we have also included the standard deviation of the observed Nd over the 5×5 km area. It can be seen that the in situ CDP and MODIS Nd values are slightly anti-correlated, with a correlation coefficient of −0.07 (but with 95% confidence limits ranging from −0.6 to +0.5) and an RMSE value of 118.9 cm−3. There is a tendency for the MODIS Nd values to be clustered around the same value, whereas corresponding CDP Nd values can reach somewhat higher values. The mean value from the in situ measurements (271 cm−3) is higher than the MODIS-retrieved one (209 cm−3). Here, we also tested different spatial averaging from MODIS and temporal averaging from CDP measurements, but the agreement between retrievals was not improved.
However, given the low number of data points the correlation could easily have occurred by chance. Indeed the 95% confidence limits of the correlation coefficient cover a very wide range suggesting that this is a strong possibility. When the data were further filtered to remove multiple-layer cloud using ceilometer, much better agreement was found (r=0.44 with upper and lower 95% confidence limits of −0.47 and 0.9, respectively). However, only seven data points remained making firm conclusions difficult.
When category II data, where the in situ Nd is estimated (Nd,p) from the measured Nacc using eq. (1) and thus the restriction that the Puijo station is in cloud is lifted, are compared to MODIS-derived Nd, the correlation is also good. But again only when the data are filtered using the ceilometer data to make sure single boundary layer cloud (cloud base height less than 800 m) existed, although there are still some points that are far from the one-to-one line. The RMSE and correlation coefficient values in this case are 99.6 cm−3 and 0.65 (the 95% confidence limits are 0.47 and 0.78), respectively. The results from the category I and category II data suggest that the height of the cloud under consideration and whether it is within the boundary layer, which is likely well mixed, is very important for matching ground measurements to satellite measurements. Additionally, the good match between MODIS and the Nd,p suggests that the MODIS retrievals themselves are not be the sole cause of the CDP discrepancy.
To further analyse the differences in cloud droplet number concentrations derived from different measurements the frequency distributions for Nd are shown in Fig. 4. Here we use data from satellite and in situ measurements from the months when the CDP was operating. Thus the distributions presented include all the data from the previous figures, and also non-coincident data from the same time period (2006–2011). It can be seen that the MODIS-derived Nd is peaked at smaller values with a median of 148 cm−3. The in situ measurement gives a slightly broader distribution of Nd values with a median of 216 cm−3.
Our in situ measurements of cloud droplet number concentration are in good agreement with several other measurements conducted in continental stratus clouds (e.g. Leaitch et al., 1992; Dong et al., 2005; Komppula et al., 2005). The mean MODIS-retrieved Nd in this study is in reasonable agreement with direct measurements and most likely the Nd retrieval method used in Janssen et al. (2011) gave too low values. However, as can be seen from eqs. (2) and (3), there are several assumptions that need to be made to calculate Nd using MODIS-retrieved COT and Reff. The calculated Nd is more sensitive to Reff than to COT and Painemal and Zuidema (2011) concluded that the MODIS-retrieved Reff systematically exceeded the in situ measured value by 15–20%. As the power of Reff in eq. (2) is −2.5, the decrease of Reff by 15–20% (when taken alone) would increase retrieved Nd by a factor of 1.5–1.75 and bring the median MODIS value from Fig. 4 into good agreement with that from the in situ data. However, Painemal and Zuidema (2011) also found that, for their assumptions of k=0.8 and of adiabatic clouds in the calculation of Nd, the overestimate of Reff by MODIS was compensated by an underestimate of the k factor (the actual value observed was ~0.88) and the observation that the clouds were fairly sub-abiabatic. They found a measured effective Cw of 0.7 times the adiabatic Cw value. However, whether the parameters observed by Painemal and Zuidema (2011) in the South-East Pacific are relevant to the Puijo region is an open question.
Thus, there are potentially large uncertainties in the estimate of Nd, which may explain some of the differences between the in situ and MODIS Nd seen here. However, it is likely that constant systematic biases in the parameters just mentioned would lead to a constant relative difference between the Nd values. This suggests that non-systematic biases were occurring in either of these parameters, the MODIS measurements or the in situ CDP measurements (or a combination). Another possible reason for discrepancy is the assumption that Nd is constant throughout the cloud, which is implicit in the Nd calculation and is also necessary to make sure that the tower observed value is not different from that higher up in the cloud. Aircraft measurements have suggested that this is generally the case in clouds with bases above ground level (Martin et al., 1994; Miles et al., 2000; Wood, 2005), although this has not been well tested for clouds near the ground. An alternative explanation is that MODIS is more likely to sample higher clouds than the surface-based CDP and thus it is possible that Nd at such elevated levels is lower than Nd near the surface due to a lower aerosol concentration.
We have compared in situ measured cloud microphysical properties to those retrieved from the MODIS instruments onboard the Aqua and Terra satellites. We have used ground-based measurements from the Puijo measurement station to study the dependence of cloud droplet number concentration (Nd) and cloud droplet effective radius (Reff) on the number concentration of aerosol particles in the accumulation mode (Nacc). As far as we know, this is the first time that Nd analysed from long-term ground-based in situ measurements have been compared to satellite observations. We have found that in situ measurements showed a clear correlation between Nd and Nacc for Nacc concentrations less than 400 cm−3. A similar but weaker correlation was seen from the remote sensing data. The ACI value (0.16) from in situ measurements was almost similar the ACI value (0.14) calculated from the MODIS-retrieved Nd. This is valid for Nd values lower than 250 cm−3 and Nacc<400 cm−3. Above this value Nd increased only slightly (in situ measurements), or levelled off at around 215 cm−3 (MODIS). Mean (median) Nd from in situ measurements was a factor of 1.2 (1.4) higher than the MODIS-derived value. However, due to several uncertainties in the method used to calculate Nd it is difficult to say if this difference was caused by, for example, a systematic error in the retrieval of Reff.
As shown, our Nd values are clearly higher than, for example, those measured in the study of Janssen et al. (2011), even though the study areas partly overlap. This suggests that the satellite-retrieved Nd is strongly dependent on the retrieval method. This must be taken into account for example when MODIS-retrieved Nd is used for model validation purposes.
The results presented above initially showed fairly poor agreement between the in situ CDP Nd and that measured from satellite. However, when the data were filtered using ceilometer data to make sure that the clouds were within the boundary layer and that only a single layer of cloud existed, the agreement was much improved. Unfortunately, the amount of data for the direct comparison was very low, and so it is difficult to draw some conclusion. Good agreement was obtained when the satellite data was compared to an Nd,p value calculated using the aerosol measurements, but again only if the data were first filtered using ceilometer measurements. This suggests that such height determination is very important when trying to match ground-based measurements to satellite measurements.
This work was supported by Maj and Tor Nessling Foundation, Academy of Finland (through the Centre of Excellence programme, project no 1118615), and by the strategic funding of the University of Eastern Finland. Analysis and visualisations used in this study were produced with MODIS level 2 and level 3 atmosphere data taken from LAADS WEB (Goddard Space Flight Centre) and Giovanni online data system, developed and maintained by the NASA GES DISC.
Dong X , Minnis P , Xi B . A climatology of midlatitude continental clouds from the ARM SGP Central Facility: part I: low-level cloud macrophysical, microphysical, and radiative properties . J. Clim . 2005 ; 18 : 1391 .
Dong X , Minnis P , Xi B , Sun-Mack S , Chen Y . Comparison of CERESMODIS stratus cloud properties with ground-based measurements at the DOE ARM Southern Great Plains site . J. Geophys. Res . 2008 ; 113 : 03204 .
Feingold G , Remer L. A , Ramaprasad J , Kaufman Y. J . Analysis of smoke impact on clouds in Brazilian biomass burning regions: an extension of Twomey's approach . J. Geophys. Res . 2001 ; 106 : 22907 .
International Panel on Climate Change (IPCC Report) . Solomon S , Qin D , Manning M , Chen Z , Marquis M , co-authors . Climate change (2007), the physical science basis . Contribution of working Group I to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change .
King M. D , Tsay S , Platnick S. E , Wang M , Liou K . Cloud retrieval algorithms for MODIS: optical thickness, effective particle radius and thermodynamic phase, MODIS Algorithm Theoretical Basis Document (ATBD-MOD-05) . 1997 ; NASA . Online at: http://modis-atmos.gsfc.nasa.gov/_docs/atbd_mod05.pdf (Accessed Aug 2012) .
Komppula M , Lihavainen H , Kerminen V.-M , Kulmala M , Viisanen Y . Measurements of cloud droplet activation of aerosol particles at a clean subarctic background site . J. Geophys. Res . 2005 ; 110 : 06204 .
Leaitch W. R , Isaac G. A , Strapp J. W , Banic C. M , Wiebe H. A . The relationship between cloud droplet number concentrations and anthropogenic pollution: observations and climatic implications . J. Geophys. Res . 1992 ; 97 : 2463 – 2474 .
Leskinen A , Portin H , Komppula M , Miettinen P , Arola A , co-authors . Overview of the research activities and results at the Puijo semi-urban measurement station . Boreal Environ. Res . 2009 ; 14 : 576 – 590 .
McComiskey A , Feingold G , Frisch A. S , Turner D. D , Miller M. A , co-authors . An assessment of aerosol–cloud interaction in marine stratus clouds based on surface remote sensing . J. Geophys. Res . 2009 ; 114 : D09203 .
McFiggans G , Artaxo P , Baltensperger U , Coe H , Facchini M. C , co-authors . The effect of physical and chemical aerosol properties on warm cloud droplet activation . Atmos. Chem. Phys . 2006 ; 6 : 2593 – 2649 .
Meskhidze N , Nenes A , Conant W. C , Seinfeld J. H . Evaluation of a new cloud droplet activation parameterization with in situ data from crystal-face and cstripe . J. Geophys. Res . 2005 ; 110 : 16202 .
Nakajima T. Y , Higurashi A , Kawamoto K , Penner J. E . A possible correlation between satellite-derived cloud and aerosol microphysical parameters . Geophys. Res. Lett . 2001 ; 28 ( 7 ): 1171 – 1174 .
Nakajima T. Y , King M. D . Determination of the optical thickness and effective particle radius of clouds from reflected solar radiation measurements, Part I: theory . J. Atmos. Sci . 1990 ; 47 : 1878 – 1893 .
Painemal D , Zuidema P . Assessment of MODIS cloud effective radius and optical thickness retrievals over the Southeast Pacific with VOCALS-REx in situ measurements . J. Geophys. Res . 2011 ; 116 : 24206 .
Portin J. H , Komppula M , Leskinen A. P , Romakkaniemi S , Laaksonen A , co-authors . Overview of the aerosol-cloud interaction at the Puijo semi-urban measurement station . Boreal Environ. Res . 2009 ; 14 : 641 – 653 .
Reutter P , Su H , Trentmann J , Simmel M , Rose D , co-authors . Aerosol- and updraft-limited regimes of cloud droplet formation: influence of particle number, size and hygroscopicity on the activation of cloud condensation nuclei (CCN) . Atmos. Chem. Phys . 2009 ; 9 : 7067 – 7080 .
Romakkaniemi S , Arola A , Kokkola H , Birmili W , Tuch T. M , co-authors . Effect of aerosol size distribution changes on AOD, CCN and cloud droplet number concentration: case studies from Erfurt and Melpitz, Germany . J. Geophys. Res . 2012 ; 117 : D07202 .
Sporre M , Glantz P , Tunved P , Swietlicki E , Kulmala M , co-authors . A study of the indirect aerosol effect on subarctic marine liquid low-level clouds using MODIS cloud data and ground-based aerosol measurements . Atmos. Res . 2011 ; 116 : 56 – 66 .
Wood R , Mechoso C. R , Bretherton C. S , Weller R. A , Huebert B , co-authors . The VAMOS Ocean-Cloud Atmosphere-Land Study Regional Experiment (VOCALS-REx): goals, platforms, and field operations . Atmos. Chem. Phys . 2011 ; 11 : 627 .