Abstract

This study focuses on the radiative properties of five subtropical marine stratocumulus cloud regions, on monthly mean scale. Through examination of the relation between total albedo and cloud fraction, and its variability and relation to other parameters, some of the factors controlling the reflectivity, or albedo, of the clouds in these regions are investigated. It is found that the main part of the variability in albedo at a given cloud fraction can be related to temporal rather than spatial variability, indicating spatial homogeneity in cloud radiative properties in the studied regions. This is seen most clearly in satellite observations but also appears in an ensemble of climate models. Further comparison between satellite data and output from climate models shows that there is good agreement with respect to the role of liquid water path, the parameter that can be assumed to be the primary source of variability in cloud reflectivity for a given cloud fraction. On the other hand, the influence of aerosol loading on cloud albedo differs between models and observations. The cloud-albedo effect, or cloud brightening caused by aerosol through its coupling to cloud droplet number concentration and droplet size, is found not to dominate in the satellite observations on monthly mean scale, as it appears to do on this scale in the climate models. The disagreement between models and observations is particularly strong in regions with frequent occurrence of absorbing aerosols above clouds, where satellite data, in contrast to the climate models, indicate a scene darkening with increasing aerosol loading.

1. Introduction

The subtropical marine stratocumulus clouds covering a considerable fraction of Earth’s surface, in regions of high insolation, are known to be of relevance for the radiation budget and climate (Ramanathan et al. 1989; Ockert-Bell and Hartmann 1992; Stevens and Brenguier 2009; Wood 2012). Their properties and potential changes in properties have therefore been the focus of many studies (e.g., Bony and Dufresne 2005; Latham et al. 2008; Clement et al. 2009; George and Wood 2010; Myers and Norris 2013). Further, the marine stratocumulus clouds have been found to be susceptible to aerosol influence via the cloud-albedo or Twomey effect (Twomey 1974, 1977), and the regions studied here have also been shown to coincide with areas of maximum negative indirect forcing in climate models (Kirkevåg et al. 2013; Carslaw et al. 2013). The same regions are also known to be difficult to correctly capture in climate models (Bender et al. 2006; Karlsson et al. 2008; Caldwell et al. 2013; Noda and Satoh 2014). Recently, however, it has been shown that the radiative properties of these clouds have improved their agreement with observations, over a generation of model development, from phases 3 and 5 of the Coupled Model Intercomparison Project (CMIP3 and CMIP5, respectively) (Engström et al. 2014). Engström et al. (2014) showed that the relation between albedo and cloud fraction in the CMIP5 models is close to linear, indicating an approximately constant cloud albedo on the regional and monthly mean time scale studied, in close agreement with satellite observations on the same spatial and temporal scale. The present analysis is similarly based on the linear relationship that emerges between albedo and cloud fraction when studied on time scales long enough to filter out the variability that stratocumulus cloud regions are known to display on short time scales (George and Wood 2010).

From Engström et al. (2014) it is clear that climate models display a greater variability in total albedo, for a given cloud fraction, than satellite data do, indicating a larger spread in cloud albedo. Engström et al. (2014) also argued that a shift in cloud albedo seen in the models between simulations of preindustrial and present-day conditions is likely caused by increases in aerosol loading. Here, we build on the results of Engström et al. (2014) by investigating the origin of the greater variability in albedo in terms of contributions from spatial and temporal variability, and by further studying the residual variability in albedo, that is, the variability not determined by cloud fraction, focusing on the contributions from variations in liquid water path (LWP) and aerosol optical depth (AOD).

When cloud fraction is fixed, the spread in estimated albedo will be governed by factors controlling the cloud optical thickness, the primary determinants being the amount of condensate in the cloud and the size of the cloud droplets () (George and Wood 2010). The size of the cloud droplets is in turn dependent on the number of available cloud condensation nuclei, determined by the amount and properties of the atmospheric aerosol (Twomey 1974, 1977). We examine the relation between the key variables albedo, cloud fraction, liquid water content (quantified by LWP), and aerosol amount (approximated by AOD), in climate models and in satellite observations.

Previous studies have compared these quantities in climate models and satellite data in a more general sense. Engström et al. (2014), for example, show that there is a spread among the CMIP5 models in their quantification of albedo and cloud fraction in the given regions, but a general agreement with Clouds and the Earth’s Radiant Energy System (CERES) and Moderate Resolution Imaging Spectroradiometer (MODIS) observations for the multimodel mean. Shindell et al. (2013) show a large range in AOD, even on a global mean scale, among the CMIP5 models. LWP too is expected to differ widely between models. Bender (2008) showed a large range in global mean LWP among the CMIP3 models, and Jiang et al. (2012) show an improvement in the CMIP5 models, but there are large remaining discrepancies between models and satellite observations. Distributions of albedo, cloud fraction, LWP, and AOD in the models and satellite data used in the present study are displayed in  appendix A and confirm the large spread among models. In this study, however, the focus is not on the quantification of albedo, cloud fraction, AOD, or LWP but rather on the relationships between these variables, and their implications.

In the case of AOD the presented method of analysis provides a means for investigating to what extent a shift in cloud albedo with aerosol loading, manifest as a cloud brightening from preindustrial to present-day conditions, can be seen on shorter time scales as variations between clean (low AOD) and polluted (high AOD) cases, under present-day conditions. Several observational studies (e.g., Conover 1966; Coakley et al. 1987; Han et al. 1994; Nakajima et al. 2001; Krüger and Graßl 2002; Roberts et al. 2008; Chen et al. 2014) have established relationships between aerosol abundance and cloud brightness in concord with the cloud-albedo effect. Other studies have questioned the climatic relevance of regional cloud brightening (Peters et al. 2011, 2014). A lack of significant hemispheric asymmetries in observed albedo and cloud optical depth, despite large differences in anthropogenic aerosol loading between the Northern and Southern Hemispheres, has also been pointed out as inconsistent with a dominating cloud-albedo effect (Schwartz 1988; Feng and Ramanathan 2010).

The aerosol–cloud interaction of the cloud-albedo effect leads to a radiative forcing to which other aerosol effects on clouds cause rapid adjustments, and it has been shown that isolating individual pathways of aerosol–cloud interaction is of limited value, due to compensational effects among them. The cloud-albedo effect is typically no longer studied in isolation, but in association with the rapid adjustments (Boucher et al. 2013; Rosenfeld et al. 2014). While global models represent an increasing number of aerosol–cloud interaction pathways, the cloud-albedo or Twomey effect can be singled out as the one included in most global models, and several models include that effect only (Ekman 2014), which justifies the focus on that aerosol–cloud interaction pathway in the present study.

2. Methods and data

a. General

We base the analysis on the method for determining cloud albedo from observations of total albedo α and cloud fraction f, described in Bender et al. (2011). The approach is based on the linear relationship between the two latter variables in a homogeneous cloud scene over a constant-albedo surface, given by

 
formula

where is the cloud albedo and is the clear-sky albedo. The mean cloud albedo can be deduced from the properties of a linear regression fit to the relation between albedo and cloud fraction. This method requires a dark surface with little variation in albedo, and not least a sufficiently homogeneous cloud scene in the sense that it is dominated by a single cloud type. Accordingly, we focus on five specific regions, all with prevailing low-level clouds; Californian (20°–30°N, 120°–130°W), Peruvian (10°–20°S, 80°–90°W), Australian (25°–35°S, 95°–105°E), Namibian (10°–20°S, 0°–10°E), and Canarian (15°–25°N, 25°–35°W) following Klein and Hartmann (1993) and also consistent with Bender et al. (2011) and Engström et al. (2014). For these regions, we consider the albedo and cloud fraction, for each point in time (on monthly mean scale) and space (within the given region and on the resolution of the model–satellite data). Particularly we study the spread around the linear relation that emerges as the variability in albedo for a given cloud fraction, as well as the organization of values of other quantities, that can be seen in albedo–cloud fraction space. By segregating the points in albedo–cloud fraction space by their LWP or AOD values, the influence of those quantities on cloud albedo can be examined.

b. Satellite observations

Monthly mean values of all-sky albedo from CERES (Wielicki et al. 1996) and daytime cloud fraction, AOD, and cloud water path from MODIS (King et al. 2003) are analyzed. Both of these instruments (MODIS and CERES) are on board Aqua and Terra, sun-synchronous polar-orbiting satellites that cross the equator at 1330 and 1030 local time, respectively. We focus here on the data collected by the instruments on Aqua, which is part of the A-Train satellite constellation, but for comparison in section 3a we show results from Terra as well. Data are obtained for the period between July 2002 and June 2011 using the MODIS 5.1 (MYD08_M3) and CERES edition 2.6 (SSF1deg-lite Ed2.6) data collections.

The albedo is based on the CERES broadband fluxes, and the cloud and aerosol retrievals are based on the MODIS visible and IR channels. The satellite footprint measurements are spatially averaged to a 1° × 1° grid. The instantaneous measurements at satellite overpass time are temporally interpolated based on constant meteorology at overpass time to create a diurnal cycle before the daily average is created. Cloud fraction and liquid water path are both derived from the cloud optical properties retrieval, calculated for daytime scenes only, and for all cloud phases (liquid, ice, and undetermined) (Platnick et al. 2003). The cloud fraction is the number of MODIS pixels identified as cloudy divided by the total number of pixels. Liquid water path is retrieved from MODIS pixels identified as cloudy, and is based on optical depth and particle size. Aerosol optical depth is based on reflectances retrieved for the MODIS pixels that are identified as clear (Remer et al. 2005; see also section 2f herein). Data are processed as described in section 2d.

c. CMIP5 models

Model output is taken from CMIP5 (Taylor et al. 2012). Monthly mean top-of-atmosphere (TOA) albedo, cloud fraction, and AOD from 20 climate models (see Table 1 for a list of included models and their spatial resolutions) are processed in the same way as the satellite data (see section 2d). The simulations used are referred to as “historical” and include observed forcings of both natural and anthropogenic origin. Twenty-five years, from January 1980 to December 2004, are utilized, and the CMIP5 models analyzed are that subset that contributed the necessary output parameters. The multimodel ensemble mean data presented are normalized to the ensemble mean values of cloud fraction and albedo. This means that for each region, model-specific offsets are calculated as the deviations of the model regional means from the multimodel ensemble regional mean, for cloud fraction and albedo respectively. These offsets are then subtracted from all data values to yield distributions that are centered around the ensemble mean albedo and cloud fraction, while maintaining the intermodel variability. Some modeling centers supply several model versions (which may have the same atmospheric component but differ in resolution or ocean model, for instance), and to avoid giving too much weight to these models in the means we include only one model from each model family, as indicated in Table 1. The results are found to be insensitive to the selection of models included in the multimodel means, and hence the conclusions are not affected by this choice. The results are also found to be insensitive to the choice of time period of the model output.

Table 1.

CMIP5 models considered in the study and the horizontal resolutions of their atmospheric components. Models included in the multimodel mean data presented are marked with an asterisk. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

CMIP5 models considered in the study and the horizontal resolutions of their atmospheric components. Models included in the multimodel mean data presented are marked with an asterisk. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)
CMIP5 models considered in the study and the horizontal resolutions of their atmospheric components. Models included in the multimodel mean data presented are marked with an asterisk. (Expansions of acronyms are available online at http://www.ametsoc.org/PubsAcronymList.)

d. Model and satellite data processing

Satellite-derived albedo, cloud fraction, and AOD are compared on the 1° × 1° resolution on which they are reported and model output on the resolution of each model respectively (see Table 1). To avoid influence of a seasonal cycle, all data are deseasonalized; that is, for all data points the mean seasonal cycle is subtracted, yielding interannual anomalies to which the mean value for all months is added. To further avoid influence of geographical dependence, the data are also, in a similar way, deregionalized. This is done by calculating a mean perturbation field for each region, as the deviation of the temporal mean in each grid cell from the total spatiotemporal mean. This perturbation is then subtracted for each month and each point in the region, and in this way persistent geographical signatures (e.g., gradients in cloudiness) are filtered out from the data. The resulting time series are hereby made independent of time and relative location within each region. These steps are precautions taken to avoid misinterpretation of large-scale seasonal and regional covariations as cloud physical relations.

e. Idealized model

In the investigation of aerosol influence on cloud albedo (see section 3b), we are aided by an idealized model, with which a first-order approximation of how the cloud-albedo effect would be expected to appear is obtained. This simple model is used to calculate an estimate of the total albedo from a random selection of data chosen from the probability density function of cloud cover over the Californian region. The LWP is allowed to vary, by randomly choosing a value between 50 and 200 g m−2 (cf. Wood 2012). To include the Twomey effect, the number of cloud droplets is assumed to increase with the number of cloud condensation nuclei (CCN) following (Jones et al. 2001). Values of CCN concentration are obtained by randomly sampling a value between 10 and 500 cm−3 (cf. Hudson and Noble 2014). To facilitate the comparison with climate models and satellite observations, the corresponding AOD distribution is estimated from the CCN concentration using the empirically based parameterization of (Andreae 2009). The resulting cloud optical depth is calculated from and LWP using

 
formula

where = 2.585 (Wood 2012) relates droplet number concentration to cloud liquid water content and droplet size, k = 0.8 (Martin et al. 1994) relates volume radius to effective radius, is the density of water, and is the moist adiabatic lapse rate, here assumed to be 5 K km−1. The choice of LWP is in each case independent of . Thereby, each LWP value will be associated with a range of values, meaning that the average LWP is the same at each cloud fraction bin, and LWP is effectively constant with respect to changes in aerosol loading, as theoretically prescribed (Twomey 1974, 1977).

Last, the variation in cloud optical depth is translated into cloud albedo using a simple radiative transfer model (Corti and Peter 2009) and taking the ratio between incident and reflected sunlight. This procedure is undertaken to obtain an estimate of the expected gradient of AOD with respect to albedo, with which the observational data can be compared. A flowchart describing the idealized model is presented in Fig. 1.

Fig. 1.

Idealized model flowchart.

Fig. 1.

Idealized model flowchart.

The simple model is based on realistically varying conditions but it does not include any dynamical or microphysical feedbacks from aerosol–cloud interactions, and should only be regarded as a tool to illustrate the cloud-albedo effect under the idealizing assumption that the LWP is invariant under variations in aerosol loading. We emphasize that the parameterization in the simple model captures the essence of the parameterizations of the cloud-albedo effect used in climate models, although most models use far more complex schemes for radiative transfer, aerosol activation and cloud microphysics.

f. Limitations in estimates of AOD and cloud fraction

AOD is used here as an indicator of the aerosol concentration, and subsequently the number concentration of CCN in the atmosphere, an approximation often made in lack of more correct data (e.g., Menon et al. 2008; Kaufman et al. 2005; Quaas et al. 2009). The relation between AOD and CCN has been shown to be valid with a high degree of correlation when considering long-term averaged data, comparing different aerosol regimes, although large local variability is expected (Andreae 2009). Further caveats apply to the use of AOD to investigate aerosol effects on clouds. AOD is a column-integrated value with no information about aerosol altitude or chemical composition, of relevance to the aerosol–cloud interaction (see section 4a). As discussed by, for instance, Nakajima et al. (2001), Andreae (2009), and Costantino and Bréon (2010), the size spectrum at a given AOD is important for the determination of the numbers of cloud condensation nuclei and droplet concentrations, and the aerosol size distribution is further related to relative humidity through hygroscopic growth. It has also been found that the present-day relationship between AOD and cloud droplet number concentration cannot be used to correctly represent the preindustrial cloud droplet number concentration (Penner et al. 2011). However, for the approach taken in this study, where the cloud-albedo effects of aerosol perturbations in present-day conditions are compared between models and observations, AOD can be used, although it is an imperfect proxy for cloud condensation nuclei.

Current generation models have been found to capture many features of the AOD distribution relatively well, although regional differences are significant, as are discrepancies in contributions from individual aerosol types. See, for example, Shindell et al. (2013), particularly showing a model underestimation of absorbing aerosol optical depth.

AOD can only be retrieved from satellite observations for cloud-free pixels. If the MODIS cloud screening algorithm, as described in Remer et al. (2005), finds an insufficient amount of cloud-free pixels in a grid cell, it does not return an AOD value and that cell is thus excluded from the analysis. The monthly gridded MODIS data are weighted by the number of ungridded 10-km retrievals in the 1° × 1° grid cell considered, as described by Remer et al. (2008). This procedure will exclude grid cells at cloud cores but is also likely to preferentially screen out grid cells at cloud edges, where effects of hygroscopic growth, subsiding shell turbulence, entrainment, and influence of scattering from neighboring clouds are important (Haywood et al. 1997; Koren et al. 2007; Charlson et al. 2007; Heus and Jonker 2008; Knight et al. 2007; Wen et al. 2007). The standard assumptions that the grid cell AOD values are representative for an entire 1° × 1° grid cell, and that these values, although not retrieved from cloud cores, are representative for the all-sky scene, are employed (cf., e.g., Koren et al. 2005; Matsui et al. 2006; Myhre et al. 2007; Quaas et al. 2009; Chen et al. 2014).

Cloud fraction is also an ambiguous quantity, and its definition is not the same between models and observations, or between different sets of observations (see, e.g., Stubenrauch et al. 2013). Differences in instrument sensitivity as well as retrieval methods and assumptions for vertical cloud structure will cause discrepancies between the estimates. Bender et al. (2011), however, found that the linear relation between albedo from CERES and cloud fraction from MODIS upon which the current analysis builds is valid also for additional independent satellite datasets, adding confidence to the relationship. In section 4, we investigate further the effects of some of the differences in cloud fraction estimates between models and observations.

The presented analysis of aerosol influence on cloud albedo relies on the use of the all-sky albedo from CERES, which makes the observed variations in albedo independent of potential biases in satellite-retrieved cloud properties like effective radius, cloud water path, or cloud masking algorithms. Further, problems related to retrievals from satellite, which have been found to bias estimates of the aerosol indirect effect (Rosenfeld and Feingold 2003), are avoided here, as the analysis of the cloud albedo effect is based on covariations in albedo and AOD, rather than on dependence on AOD.

3. Results

a. Variability

1) LWP and AOD

First we investigate the spread in LWP and AOD and their relations to total albedo variations. Figure 2 shows the variability in LWP, and the correlation between LWP and total albedo, for all models and two sets of satellite observations. The MODIS estimates from Aqua and Terra agree very well, indicating a minor influence of the difference in overpass time. For the multimodel mean, there is a general agreement with the observations in both correlation and in standard deviation, and the observed values are close to the center of the model distribution. The correlations in the models range from ~0.2 to 0.8, but are typically higher than 0.5. However, the variability range is large in the model ensemble, with the smallest and largest standard deviation values differing by a factor of ~5.

Fig. 2.

Variability in LWP expressed as the standard deviation of LWP at all points and all times, and the correlation between LWP and albedo (also spatiotemporal) in models and observations (colored symbols) for the five: (a) Californian, (b) Peruvian, (c) Canarian, (d) Namibian, and (e) Australian regions. Filled and boldfaced symbols denote correlations that are statistically significant at the 95% level, and low correlations (below 0.2) are gray shaded.

Fig. 2.

Variability in LWP expressed as the standard deviation of LWP at all points and all times, and the correlation between LWP and albedo (also spatiotemporal) in models and observations (colored symbols) for the five: (a) Californian, (b) Peruvian, (c) Canarian, (d) Namibian, and (e) Australian regions. Filled and boldfaced symbols denote correlations that are statistically significant at the 95% level, and low correlations (below 0.2) are gray shaded.

Figure 3, analogous to Fig. 2, shows the variability in AOD and its correlation with total albedo, in models and observations. Here too, the Aqua and Terra estimates agree well, but the agreement between models and observations is poorer. The models to a varying degree show a positive correlation (up to 0.5) between albedo and AOD, whereas in the observations these quantities are less well, or even negatively, correlated. The observed standard deviations are close to the mean of the relatively large range displayed by the models.

Fig. 3.

As in Fig. 2, but for AOD.

Fig. 3.

As in Fig. 2, but for AOD.

We note that most of the studied models (all except BNU-ESM, GFDL-ESM2G, GFDL-ESM2M, GISS-E2-R, IPSL-CM5A-LR, IPSL-CM5A-MR, and IPSL-CM5B-LR; Collins et al. 2013) implement a cloud lifetime effect (Albrecht 1989), yielding a relationship between AOD and LWP. For these models, the correlation between LWP and albedo could result in a strengthened correlation between AOD and albedo. However, no such systematic difference between models with and without a cloud lifetime effect is seen.

2) Spatial and temporal variability

The albedo variability may further originate from variations in both the spatial and temporal dimensions. Figure 4 compares temporal and spatial variability, and their relative contribution to the total variability, in the satellite observations. For the Californian, Peruvian, Australian, and Namibian regions, it is clear that averaging out the temporal variability by taking the mean over the whole time period removes much of the variability and results in a close linear relation between albedo and cloud fraction. This supports the notion of a homogeneous cloud type, in terms of similar cloud reflectivity across the full region. When on the other hand the spatial variability is averaged out, by taking the mean of the deseasonalized data over the whole domain, the remaining temporal variability is comparatively large, resulting in a greater spread around the linear relation, as well as a larger range of cloud fractions. For the Canarian region, the tendency for the temporal variability to be the larger component is still clear, but here the remaining relation when the temporal variability is averaged out shows a narrow range of cloud fractions, and a relatively constant total albedo. This, in turn, indicates that cloud albedo is lower at higher cloud fraction; that is, there is a negative correlation between cloud fraction and cloud albedo.

Fig. 4.

Albedo as a function of cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Averaging over the whole region yields the residual temporal variability (red) and averaging over the whole time period yields the residual spatial variability (blue). Gray shading indicates the full distribution. The cross on the y axis indicates the regional mean CERES clear-sky albedo.

Fig. 4.

Albedo as a function of cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Averaging over the whole region yields the residual temporal variability (red) and averaging over the whole time period yields the residual spatial variability (blue). Gray shading indicates the full distribution. The cross on the y axis indicates the regional mean CERES clear-sky albedo.

This can be compared with the climate models, which similar to the observations in most cases show a larger residual temporal than spatial variability. The residual temporal variability, however, is in general larger than that seen in the observations, and the residual spatial variation is in many cases not as linearly distributed in albedo–cloud fraction space as it is in the observations. Figure 5 shows the residual spatial and temporal variability in one model (NorESM1-M), as an example.

Fig. 5.

As in Fig. 4, but for one CMIP5 model, NorESM1-M.

Fig. 5.

As in Fig. 4, but for one CMIP5 model, NorESM1-M.

Figure 4 further shows a very good agreement between the reported clear-sky albedo and that estimated from the relation between albedo and cloud fraction, from CERES and MODIS. For the NorESM1-M, on the other hand (Fig. 5), the intercept is lower than the reported clear-sky albedo in all regions except the Namibian. The NorESM1-M overestimates the clear-sky albedo compared to the satellite observations in all regions, which is also true for the multimodel mean.

b. Patterns

1) LWP

Being a main driver for cloud albedo variability, LWP as expected causes a spread around the near-linear relation between albedo and cloud fraction, and is also distributed in albedo–cloud fraction space in such a way that high values of LWP for any given cloud fraction correspond to a higher albedo. This is manifested as a positive gradient of LWP in albedo–cloud fraction space at each cloud fraction. Figures 6 and 7 show the relationship between albedo, cloud fraction, and LWP for satellite observations and the ensemble mean of the climate models respectively. It is hereby clear that the variations in LWP are related to cloud-albedo variations in similar ways in the models and the observations. The LWP values are presented as deviations from the mean value in each cloud fraction bin, rather than absolute values, as the mean LWP varies widely among models (see also Bender 2008; Jiang et al. 2012).

Fig. 6.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Each point is color-coded by the corresponding LWP anomaly (i.e., deviation from mean in each cloud fraction bin).

Fig. 6.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Each point is color-coded by the corresponding LWP anomaly (i.e., deviation from mean in each cloud fraction bin).

Fig. 7.

As in Fig. 6, but for the CMIP5 ensemble mean.

Fig. 7.

As in Fig. 6, but for the CMIP5 ensemble mean.

2) AOD

(i) Idealized model

According to the cloud-albedo or Twomey effect, an increase in the number of aerosols that can act as cloud condensation nuclei leads to an increase in cloud albedo, given that the liquid water content remains constant. The idealized cloud-albedo effect is illustrated in Fig. 8, showing results from the simple model described in section 2e. In the model the cloud albedo increases with increasing aerosol loading, here represented by AOD to facilitate the comparison with satellite data (see section 2). A positive gradient in AOD along the albedo axis at any given cloud fraction is evident, implying that the simulated polluted clouds are associated with higher albedos.

Fig. 8.

Total albedo vs cloud fraction with points color coded by their AOD from an idealized model simulation. The upper color distribution corresponds to the absolute AOD and lower one to the AOD anomaly, i.e., deviation from mean value for each given cloud fraction. The top and bottom dashed lines indicate linear regression of albedo onto cloud fraction from subsets of data representing polluted and clean conditions, respectively. The bottom of the figure shows the standard deviation of AOD for each cloud fraction.

Fig. 8.

Total albedo vs cloud fraction with points color coded by their AOD from an idealized model simulation. The upper color distribution corresponds to the absolute AOD and lower one to the AOD anomaly, i.e., deviation from mean value for each given cloud fraction. The top and bottom dashed lines indicate linear regression of albedo onto cloud fraction from subsets of data representing polluted and clean conditions, respectively. The bottom of the figure shows the standard deviation of AOD for each cloud fraction.

In Fig. 8, as a quantification of the cloud-albedo effect, the difference in cloud albedo between cases representing clean and polluted conditions is estimated by extrapolating linear regression lines to 100% cloud fraction, in two contrasting subsets of data. We let the clean and polluted cases correspond to values of cloud droplet number concentrations of 75 ± 50 and 150 ± 50 cm−3 respectively, representing a moderate estimate of the range from clean to polluted conditions (cf. Wood 2012; Zeng et al. 2014), and resulting in an estimated difference in albedo Δα of 0.04. From this Δα a local radiative effect ΔR can be calculated, as ΔRlocal = SΔαf, and with a solar mean insolation of Slocal = 400 W m−2 and a mean cloud fraction of f = 50% this ΔRlocal is estimated to be approximately 8 W m−2. Assuming that the same albedo difference is valid for all marine stratocumulus clouds (covering q = 23% of the earth’s surface; Wood 2012) a corresponding global ocean radiative effect can be calculated (with Sglobal = 340 W m−2) as ΔRglobal = qSΔα (i.e., 3 W m−2).

If the cloud-albedo effect dominates, a signal resembling the gradient seen in Fig. 8 may be expected to appear in climate models as well as in observational data. To test this we examine model and satellite estimates of albedo, cloud fraction, and AOD.

(ii) Climate models and satellite observations

When the satellite data are plotted in albedo–cloud fraction space in an attempt to reproduce the results from the idealized model in Fig. 8, a positive correlation between AOD and cloud fraction appears, and in some regions creates a dominating pattern (see Fig. 9).

Fig. 9.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Each point is color-coded by the corresponding absolute value of AOD, showing an emerging correlation in the satellite observations between cloud fraction and AOD. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Fig. 9.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a)–(e) the five regions. Each point is color-coded by the corresponding absolute value of AOD, showing an emerging correlation in the satellite observations between cloud fraction and AOD. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Such a correlation is a well-documented feature in satellite estimates of these quantities and although many physical mechanisms have been proposed to explain the observed relations—including aerosol–cloud interactions and covariation in AOD and cloud fraction driven by relative humidity—the primary causes remain to be identified (Kaufman et al. 2005; Myhre et al. 2007; Quaas et al. 2009; Loeb and Schuster 2008; Grandey et al. 2013). Owing to the observed correlations between AOD and cloud fraction, the mean AOD is subtracted for each given cloud fraction range, to obtain the deviations from the mean for each cloud fraction. Then, for each cloud fraction range, variations are sought in the total albedo, in response to deviations of the AOD values from their means.

In the climate model output, although for example humidity swelling is represented, the correlation between AOD and cloud fraction is not present nearly to the same extent as in the satellite data, but for consistency these results too are shown as anomalies. In the idealized model results, displayed in Fig. 8, no such relation between AOD and cloud fraction is imposed, and the results may be expressed either in absolute values or anomalies, with no effect on the pattern created by AOD in albedo–cloud fraction space.

Figures 10 and 11 consequently display the relation between total albedo and cloud fraction for the five regions, with each point color-coded by its AOD anomaly in models and observations respectively. The CMIP5 models are shown as a multimodel ensemble mean (see section 2b), which is found to be representative for the ensemble of models. As in Fig. 8, the color scales are chosen to correspond to ±1 mean standard deviation of all AOD values in each cloud fraction bin, also displayed in Figs. 10 and 11. We note that for the Californian, Peruvian, and Australian regions, the data and model output show comparable standard deviations, and also corroborate the range produced by the idealized model (Fig. 8). In the Namibian and Canarian regions the idealized model underestimates the AOD variability.

Fig. 10.

Total albedo vs cloud fraction in the CMIP5 ensemble mean for (a)–(e) the five regions. Each point is color-coded by the corresponding AOD anomaly. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Fig. 10.

Total albedo vs cloud fraction in the CMIP5 ensemble mean for (a)–(e) the five regions. Each point is color-coded by the corresponding AOD anomaly. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Fig. 11.

As in Fig. 10, but for satellite observations (CERES and MODIS).

Fig. 11.

As in Fig. 10, but for satellite observations (CERES and MODIS).

The CMIP5 ensemble clearly confirms the idealized model results, showing higher albedos for higher AOD, for all five regions. Contrarily for the satellite observations, the distribution of AOD values in albedo–cloud fraction space does not agree with the expected theoretical results. The Californian, Peruvian, and Australian regions all display weak and, to a varying degree, unorganized patterns in AOD. The Namibian and Canarian regions both display gradients in AOD nearly inverse to that expected from theory and seen in the CMIP5 models.

4. Discussion

In the following we discuss possible reasons for the discrepancy between models and observations in their depiction of the relationship between albedo, cloud fraction, and AOD.

a. Aerosol differences

It must be noted that the five regions studied have differing aerosol signatures both in terms of dominating aerosol type, and altitude of the aerosol relative to the cloud layer. The region off the coast of Namibia is strongly influenced by absorbing soot aerosols from biomass burning on the African continent, aerosols that are generally overlying the cloud layer and therefore will largely not interact with the clouds (Kaufman et al. 2002; Wilcox 2010). Furthermore, their hydrophobic properties may make them inefficient as cloud condensation nuclei (Kaufman et al. 2002). Therefore, instead of an influential cloud-albedo effect, other aerosol effects may be hypothesized. These include 1) a direct effect, by which the absorption of incident sunlight by the dark aerosol will lower the scene albedo (Wilcox 2012); 2) a positive semidirect effect in cases when the absorbing aerosol is mixed with the cloud, causing heating and subsequent cloud clearing and thinning and thereby reducing the scene albedo (Hansen et al. 1997; Johnson et al. 2004); and 3) a negative semidirect effect, by which aerosol overlying the cloud causes a cloud thickening through a dynamical feedback related to the stability of the atmosphere, increasing the scene albedo (Wilcox 2010). A direct effect and a positive semidirect effect are both consistent with the results shown in Fig. 11, namely that polluted cloud scenes appear to be darker than cleaner cases, whereas a negative semidirect effect would rather create or reinforce a Twomey-like gradient. Similar reasoning applies to the Canarian region, where outflow of weakly absorbing Saharan dust aloft contributes strongly to the AOD (Kaufman et al. 2002), making direct and semidirect effects potentially more important than the cloud-albedo effect. Apparently, this is not captured by the CMIP5 model ensemble (Fig. 10).

For the remaining regions, aerosols may be assumed to be more vertically co-occurring with the low-level clouds and more efficient as cloud condensation nuclei, as supported by the satellite-based aerosol climatologies presented by, for example, Devasthale and Thomas (2011), Waquet et al. (2013), and Winker et al. (2013). The aerosols in these regions are therefore more prone to give rise to a cloud-albedo effect, but still Twomey-consistent patterns are nonexistent in the satellite data, and cloud albedos for clean and polluted cases cannot be separated.

b. Cloud cover and vertical cloud distribution

The five studied regions are chosen in agreement with the observed local maxima in low cloud cover, corresponding to persistent and extensive stratocumulus decks. In section 3a it was demonstrated that in the observations there is spatial homogeneity in the cloud radiative properties, particularly in the Californian, Peruvian, Namibian, and Australian regions, although this is somewhat less pronounced in the models. As an indication that the chosen regions cover areas of extensive clouds in the models as well, Fig. 12 shows the relative difference between modeled and observed total cloud fraction, by region. For the Californian, Peruvian, Namibian, and Australian regions the relative deviation between regional climatological mean cloud fraction, comparing models to observations, is less than 0.3 in more than 75% of the models, and all models have relative errors below 0.5. For the Canarian region, however, only 20% of the models have a relative deviation from the observations of less than 0.3. Therefore only 10% of the models have relative errors below 0.3 for all regions and only 25% of the models have a relative error that is smaller than 0.6 in all regions.

Fig. 12.

Number of CMIP5 models with relative error in cloud fraction compared to MODIS below a given value for the five regions.

Fig. 12.

Number of CMIP5 models with relative error in cloud fraction compared to MODIS below a given value for the five regions.

Although the regions studied are dominated by low-level clouds there are cases of inhomogeneous cloud scenes (e.g., high clouds overlying the stratocumulus deck). Overlying high clouds could potentially be masking the aerosol signal in the satellite data, but this appears not to be the case. To identify scenes with high clouds we use the MODIS cloud-top temperature. Cases with a daily mean cloud-top temperature below 273 K are screened out and the remaining data are used to create new monthly means for albedo, cloud fraction, and AOD, with high cloud cases excluded. These data are displayed in Fig. 13. The vast majority (~80%) of the scenes are found to have lower cloud tops, and the similarity of the results including all clouds and excluding high clouds (Figs. 11 and 13) indicates that it is adequate to analyze the full cloud scene.

Fig. 13.

As in Fig. 11, but excluding scenes with high clouds (MODIS cloud-top temperature <273 K).

Fig. 13.

As in Fig. 11, but excluding scenes with high clouds (MODIS cloud-top temperature <273 K).

c. Seasonal variation

The seasonal variation of aerosol loading and composition may be significant, particularly in the Namibian and Canarian regions, where the presence of absorbing aerosol is presented as a possible explanation for the reversed gradient in AOD in albedo–cloud fraction space (section 4a).

While aerosols from industry, traffic, and domestic burning are emitted throughout the year, open biomass burning aerosols over southern Africa are produced primarily during the dry season (i.e., the Southern Hemisphere winter months), and dust aerosols over northern Africa are expected to peak during Northern Hemisphere spring and summer, but with some emissions throughout the year (see, e.g., Prospero et al. 2002; Engelstaedter et al. 2006; Liousse et al. 2010).

The seasonal variation of the AOD distribution in albedo–cloud fraction space, for these regions, is illustrated in Fig. 14. For the Namibian region there is a clear seasonality in the AOD pattern, with a stronger gradient in the Southern Hemisphere winter months, including the peak biomass burning season in southern Africa. For the Canarian region the difference between summer and winter is not as large, but it is clear that the major contribution to the reversed AOD gradient comes from the Northern Hemisphere summer months, including the peak dust season in northern Africa. In the Californian, Peruvian, and Australian regions the lack of organized pattern of AOD anomalies in aerosol–cloud fraction space is seen for all seasons (not shown).

Fig. 14.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a),(b) the Namibian and (c),(d) Canarian regions in (left) April–September (AMJJAS) and (right) October–March (ONDJFM). Each point is color-coded by the corresponding AOD anomaly.

Fig. 14.

Total albedo vs cloud fraction in satellite observations (CERES and MODIS) for (a),(b) the Namibian and (c),(d) Canarian regions in (left) April–September (AMJJAS) and (right) October–March (ONDJFM). Each point is color-coded by the corresponding AOD anomaly.

d. Time scale

In this study, we focus on the monthly mean time scale, for which a comparison between satellite data and CMIP5 models can be performed. We note, however, that the results may to some degree be time scale dependent. Daily data do not exhibit as well-defined linear behavior as the monthly mean data when plotted in albedo–cloud fraction space, and therefore the methods based on Bender et al. (2011) are not directly applicable, but segregating the daily resolved data by AOD reveals that a reversed gradient is present in the Canarian and Namibian regions, for moderate to large cloud fractions. For the Californian, Peruvian, and Australian regions, the daily data display a positive gradient that is partly consistent with what the models display on the monthly mean time scale, but is not seen in monthly mean satellite data. For LWP there are positive gradients in all five regions, as is also seen in both models and data on the monthly mean scale. Daily data are shown in  appendix B.

5. Summary and conclusions

In this study, five subtropical stratocumulus cloud regions are studied. The relation between the quantities albedo, cloud fraction, liquid water path, and aerosol optical depth, including their variability and covariation, are investigated, comparing satellite observations from CERES and MODIS with output from a number of CMIP5 climate models, on the monthly mean scale.

Although the variability neither in AOD nor in LWP is found to be systematically greater in models than in observations, most models display a positive correlation between AOD and albedo that is not seen in the observations. This could contribute to the previous findings that climate models in general display greater variability in albedo for a given cloud fraction (Engström et al. 2014). We also find that the largest contribution to the variability originates from temporal rather than spatial variability, both in observations and models, although individual models display different patterns of variability in different regions.

Studying the distribution of the LWP and AOD values around the near-linear relation between albedo and cloud fraction we find that in both models and observations, the LWP is closely related and positively correlated with the cloud albedo; values of relatively high LWP at a given cloud fraction are related to relatively high albedo, indicating higher cloud albedo at higher LWP. The similarity in the distribution of the LWP in albedo–cloud fraction space between models and observations is reassuring.

On the other hand, the contrast between models and observations in the distribution of AOD in albedo–cloud fraction space is more troublesome. The climate models for all five studied regions indicate that relatively high AOD is related to relatively high albedo, for a given cloud fraction, in accordance with an idealized model representation of the cloud-albedo or Twomey effect. For the satellite observations, on the other hand, three regions show no organized pattern of AOD in albedo–cloud fraction space, and two regions display reversed gradients, indicating that high AOD is related to low cloud albedo. The results shown for the multimodel mean are found to be representative for the ensemble of models.

We investigate possible reasons for these discrepancies, but we find that neither differences in total cloud cover in the studied regions nor differences in altitude distribution of clouds offer a plausible explanation. In general, the agreement in estimated cloud fraction is good between models and observations, although in the Canarian region (one of the regions displaying a reversed gradient) the models tend to wrongly estimate the cloud fraction to a greater extent than in the other four regions. Limiting the analysis to low clouds only, by screening out cases with high clouds, does not significantly influence the results presented.

The difference between the Californian, Peruvian, and Australian regions, on the one hand, with no discernible gradient in AOD in the albedo–cloud fraction space, and the Canarian and Namibian regions on the other hand, with a reversed gradient in AOD, may originate in differences in prevailing aerosol conditions, with larger contribution from cloud-overlying absorbing aerosol in the latter two regions. It is found that the seasonality in the biomass burning aerosol and dust loading respectively is reflected in the strength of the reversed AOD gradient in the Namibian and Canarian regions, making contributions from these aerosols a plausible explanation for the patterns. Underestimation or misrepresentation of the vertical distribution of absorbing aerosols in climate models may thereby partly explain the distinct discrepancy between models and observations in these regions.

Despite the striking differences between observations and models we emphasize that our results by no means disprove the existence of a microphysical Twomey effect. In fact, the daily averaged satellite data present a signature coherent with a cloud brightening in several regions. However, while the influence of the parameterized Twomey effect in the models propagates to monthly mean scale, the observations indicate that its impact on cloud albedo and total scene albedo may not dominate on this scale. Aerosol influence on clouds is complex and uncertain, and the difficulty in the determination of aerosol effects on cloud properties is well known, and can be illustrated as the “cloud problem” (Stevens and Brenguier 2009); aerosols are only one factor of importance for cloud properties, and it is virtually impossible to separate their effects from effects of perturbations in meteorology and dynamical or microphysical feedbacks that may exert higher-order influence. In the CMIP5 models, the parameterized microphysical effect of aerosols on cloud droplet number has larger-scale impacts, whereas the satellite data indicate that in the monthly mean such an effect is likely to be hidden or overriden by other processes. This corroborates the previously discussed compensational or buffering effects in the system (Rosenfeld et al. 2014; Stevens and Feingold 2009). Hence, the discrepancy between models and observations may be a consequence of models failing to account for all the processes of relevance for determining the cloud albedo, further illustrating that leaving some processes out may give too much weight to those represented.

We acknowledge the importance of temporal scale to the presented results. Previous studies (Quaas et al. 2009) have found positive relations between AOD and albedo over large areas, for daily satellite data, in contradiction with the AOD patterns in albedo–cloud fraction space shown here for the monthly mean scale. We find that this relation disappears on monthly scale for three of the studied regions and is reversed on both daily and monthly scales for two regions, making the observations disagree with the climate models on the climatologically relevant monthly mean scale. The relation between albedo and LWP seems to be more persistent across time scales. Finally, we note that the observational analysis presented here is based on one particular dataset (CERES–MODIS as described in section 2b). Performing the corresponding analysis with complementing satellite data from other sensors [e.g., CALIPSO lidar reflectance (Charlson et al. 2007), cloud fraction, and aerosol optical depth] could add more confidence to the conclusions, and may also be a way to investigate the dependence on temporal and spatial scale further.

Acknowledgments

The authors thank H. Rodhe for insightful comments, and the NASA Langley Atmospheric Science Data Center for satellite data distribution. We acknowledge the World Climate Research Programme’s Working Group on Coupled Modelling, which is responsible for CMIP, and we thank the climate modeling groups (listed in Table 1 of this paper) for producing and making available their model output. For CMIP the U.S. Department of Energy’s Program for Climate Model Diagnosis and Intercomparison provides coordinating support and led development of software infrastructure in partnership with the Global Organization for Earth System Science Portals.

APPENDIX A

Quantification of Studied Variables

There are differences among models, and between models and observations, in their estimates of the studied variables. Figures A1A4 show the distributions of albedo, cloud fraction, liquid water path (LWP), and aerosol optical depth (AOD) in the CMIP5 models and in satellite observations from CERES and MODIS on Aqua, for the five regions. For albedo, despite the spread among models, the multimodel mean agrees well with the observed distribution in all regions except the Canarian region, where almost all models overestimate the albedo values. The cloud fraction distributions in the models show a larger spread, but still the multimodel mean is in fair agreement with the observations. The largest discrepancy is again seen in the Canarian regions, where most models overestimate the cloud fraction values. For AOD the modeled distributions are typically broader than the observed, but the mean values are similar in the multimodel mean and the observations. The largest spread among models is seen in the Canarian region. The LWP distributions differ widely among models, in several cases (the Peruvian, Canarian, and Australian regions) resulting in a bimodal multimodel mean distribution. The models tend to underestimate the observed LWP distributions, and in all regions the mean values are higher in the observations than the multimodel mean.

Fig. A1.

Distribution of monthly and regional mean albedo in the five regions in the individual CMIP5 models (colors) with the multimodel mean (thick black lines) and the CERES observations (thick red lines).

Fig. A1.

Distribution of monthly and regional mean albedo in the five regions in the individual CMIP5 models (colors) with the multimodel mean (thick black lines) and the CERES observations (thick red lines).

Fig. A2.

As in Fig. A1, but for cloud fraction and with MODIS observations (red).

Fig. A2.

As in Fig. A1, but for cloud fraction and with MODIS observations (red).

Fig. A3.

As in Fig. A2, but for AOD.

Fig. A3.

As in Fig. A2, but for AOD.

Fig. A4.

As in Fig. A2, but for LWP.

Fig. A4.

As in Fig. A2, but for LWP.

APPENDIX B

Daily Satellite Data

Figures B1 and B2 show daily satellite data, in albedo–cloud fraction space, segregated by AOD and LWP respectively. On daily time scale, the relation between albedo and cloud fraction shows a larger spread than on monthly mean scale, and a constant regional cloud-albedo is not well defined (cf. Bender et al. 2011). However, the patterns of AOD and LWP on daily scale can be compared qualitatively to those on monthly mean scale. For AOD, the Californian, Peruvian, and Australian regions all show a positive gradient in albedo–cloud fraction space, consistent with the theoretical model (Fig. 8) and the CMIP5 model data on monthly mean scale (Fig. 10). This signal indicating higher scene albedo at higher AOD is seen for all cloud fractions (i.e., seems to be an effect both in clear and cloud-covered sky) but apparently it does not propagate to the monthly mean scale in the satellite observations (Fig. 11). In the Californian region, and to lesser extent in the Australian region, there is indication of a reversed gradient at high cloud fractions and low albedos (i.e., for cases of thin but extensive cloud cover). For the Namibian and Canarian regions, the reversed signal of lower scene albedo at higher AOD, also seen in satellite observations on the monthly mean scale (Fig. 11), is present for cloud fractions above ~30%. At lower cloud fractions, including completely clear sky, the gradient indicates higher scene albedo at higher AOD. Hence, the only signal in AOD that propagates from daily to monthly mean scale in the data is the reversed signal found in the Namibian and Canarian regions. For LWP, all regions show a positive gradient, indicating that on the daily scale high values of LWP also correspond to high values of scene albedo, consistent with what is found for the monthly mean scale in both models and observations (Figs. 7 and 6).

Fig. B1.

Total albedo vs cloud fraction with points color coded by their AOD anomaly from daily satellite observations (CERES and MODIS) in (a)–(e) the five regions. Dashed lines indicate linear regression of albedo onto cloud fraction. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Fig. B1.

Total albedo vs cloud fraction with points color coded by their AOD anomaly from daily satellite observations (CERES and MODIS) in (a)–(e) the five regions. Dashed lines indicate linear regression of albedo onto cloud fraction. The bottom of the panels shows the standard deviation of AOD for each cloud fraction.

Fig. B2.

Total albedo vs. cloud fraction with points color-coded by their LWP value from daily satellite observations (CERES and MODIS), (a)–(e) the five regions. Dashed lines indicate linear regression of albedo onto cloud fraction.

Fig. B2.

Total albedo vs. cloud fraction with points color-coded by their LWP value from daily satellite observations (CERES and MODIS), (a)–(e) the five regions. Dashed lines indicate linear regression of albedo onto cloud fraction.

REFERENCES

REFERENCES
Albrecht
,
B. A.
,
1989
:
Aerosols, cloud microphysics, and fractional cloudiness
.
Science
,
245
,
1227
1230
, doi:.
Andreae
,
M. O.
,
2009
:
Correlation between cloud condensation nuclei concentration and aerosol optical thickness in remote and polluted regions
.
Atmos. Chem. Phys.
,
9
,
543
556
, doi:.
Bender
,
F. A.-M.
,
2008
:
A note on the effect of GCM tuning on climate sensitivity
.
Environ. Res. Lett.
,
3
,
014001
, doi:.
Bender
,
F. A.-M.
,
H.
Rodhe
,
R. J.
Charlson
,
A. M.-L.
Ekman
, and
N.
Loeb
,
2006
:
22 views of the global albedo—Comparison between 20 GCMs and two satellites
.
Tellus
,
58A
,
320
330
, doi:.
Bender
,
F. A.-M.
,
R. J.
Charlson
,
A. M.-L.
Ekman
, and
L.
Leahy
,
2011
:
Quantification of monthly mean regional-scale albedo of marine stratiform clouds in satellite observations and GCMs
.
J. Appl. Meteor. Climatol.
,
50
,
2139
2148
, doi:.
Bony
,
S.
, and
J.-L.
Dufresne
,
2005
:
Marine boundary layer clouds at the heart of tropical cloud feedback uncertainties in climate models
.
Geophys. Res. Lett.
,
32
,
L20806
, doi:.
Boucher
,
O.
, and Coauthors
,
2013
: Clouds and aerosols. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 571–657.
Caldwell
,
P. M.
,
Y.
Zhang
, and
S. A.
Klein
,
2013
:
CMIP3 subtropical stratocumulus cloud feedback interpreted through a mixed-layer model
.
J. Climate
,
26
,
1607
1625
, doi:.
Carslaw
,
K. S.
, and Coauthors
,
2013
:
Large contribution of natural aerosols to uncertainty in indirect forcing
.
Nature
,
507
,
67
71
, doi:.
Charlson
,
R. J.
,
A. S.
Ackerman
,
F. A.-M.
Bender
,
T. L.
Anderson
, and
Z.
Liu
,
2007
:
On the climate forcing consequences of the albedo continuum between cloudy and clear air
.
Tellus
,
59B
,
715
727
, doi:.
Chen
,
Y.-C.
,
M. W.
Christensen
,
G. L.
Stephens
, and
J. H.
Seinfeld
,
2014
:
Satellite-based estimate of global aerosol–cloud radiative forcing by marine warm clouds
.
Nat. Geosci.
,
7
,
643
646
, doi:.
Clement
,
A. C.
,
R.
Burgman
, and
J. R.
Norris
,
2009
:
Observational and model evidence for positive low-level cloud feedback
.
Science
,
325
,
460
464
, doi:.
Coakley
,
J. A.
,
R. L.
Bernstein
, and
P. A.
Durkee
,
1987
:
Effect of ship tracks on cloud reflectivity
.
Science
,
237
,
1020
1022
, doi:.
Collins
,
M.
, and Coauthors
,
2013
: Long-term climate change: Projections, commitments and irreversibility. Climate Change 2013: The Physical Science Basis, T. F. Stocker et al., Eds., Cambridge University Press, 1029–1136.
Conover
,
J. H.
,
1966
:
Anomalous cloud lines
.
J. Atmos. Sci.
,
23
,
778
785
, doi:.
Corti
,
T.
, and
T.
Peter
,
2009
:
A simple model for cloud radiative forcing
.
Atmos. Chem. Phys.
,
9
,
5751
5758
, doi:.
Costantino
,
L.
, and
F.-M.
Bréon
,
2010
:
Analysis of aerosol–cloud interaction from multi-sensor satellite observations
.
Geophys. Res. Lett.
,
37
,
L11801
, doi:.
Devasthale
,
A.
, and
M. A.
Thomas
,
2011
:
A global survey of aerosol–liquid water overlap based on four years of CALIPSO-CALIOP data
.
Atmos. Chem. Phys.
,
11
,
1143
1154
, doi:.
Ekman
,
A. M. L.
,
2014
:
Do sophisticated parameterizations of aerosol–cloud interactions in CMIP5 models improve the representation of recent observed temperature trends?
J. Geophys. Res. Atmos.
,
119
,
817
832
, doi:.
Engelstaedter
,
S.
,
I.
Tegen
, and
R.
Washington
,
2006
:
North African dust emissions and transport
.
Earth Sci. Rev.
,
79
,
73
100
, doi:.
Engström
,
A.
,
F. A.-M.
Bender
, and
J.
Karlsson
,
2014
:
Improved representation of marine stratocumulus cloud shortwave radiative properties in the CMIP5 climate models
.
J. Climate
,
27
,
6175
6188
, doi:.
Feng
,
Y.
, and
V.
Ramanathan
,
2010
:
Investigation of aerosol–cloud interactions using a chemical transport model constrained by satellite observations
.
Tellus
,
62B
,
69
86
, doi:.
George
,
R. C.
, and
R.
Wood
,
2010
:
Subseasonal variability of low cloud radiative properties over the southeast Pacific Ocean
.
Atmos. Chem. Phys.
,
10
,
4047
4063
, doi:.
Grandey
,
B. S.
,
P.
Stier
, and
T. M.
Wagner
,
2013
:
Investigating relationships between aerosol optical depth and cloud fraction using satellite, aerosol reanalysis and general circulation model data
.
Atmos. Chem. Phys.
,
13
,
3177
3184
, doi:.
Han
,
Q. Y.
,
W. B.
Rossow
, and
A. A.
Lacis
,
1994
:
Near-global survey of effective droplet radii in liquid water clouds using ISCCP data
.
J. Climate
,
7
,
465
497
, doi:.
Hansen
,
J.
,
M.
Sato
, and
R.
Ruedy
,
1997
:
Radiative forcing and climate response
.
J. Geophys. Res.
,
102
(
D6
),
6831
6864
, doi:.
Haywood
,
J. M.
,
V.
Ramaswamy
, and
L. J.
Donner
,
1997
:
A limited-area-model case study of the effects of sub-grid scale variations in relative humidity and cloud upon the direct radiative forcing of sulfate aerosol
.
Geophys. Res. Lett.
,
24
,
143
146
, doi:.
Heus
,
T.
, and
H. J. J.
Jonker
,
2008
:
Subsiding shells around shallow cumulus clouds
.
J. Atmos. Sci.
,
65
,
1003
1018
, doi:.
Hudson
,
J. G.
, and
S.
Noble
,
2014
:
CCN and vertical velocity influences on droplet concentrations and supersaturations in clean and polluted stratus clouds
.
J. Atmos. Sci.
,
71
,
312
331
, doi:.
Jiang
,
J. H.
, and Coauthors
,
2012
:
Evaluation of cloud and water vapor simulations in CMIP5 climate models using NASA “A-Train” satellite observations
.
J. Geophys. Res.
,
117
,
D14105
, doi:.
Johnson
,
B. T.
,
K. P.
Shine
, and
P. M.
Forster
,
2004
:
The semi-direct aerosol effect: Impact of absorbing aerosols on marine stratocumulus
.
Quart. J. Roy. Meteor. Soc.
,
130
,
1407
1422
, doi:.
Jones
,
A.
,
D. L.
Roberts
,
M. J.
Woodage
, and
C. E.
Johnson
,
2001
:
Indirect sulphate aerosol forcing in a climate model with an interactive sulphur cycle
.
J. Geophys. Res.
,
106
,
20 293
20 310
, doi:.
Karlsson
,
J.
,
G.
Svensson
, and
H.
Rodhe
,
2008
:
Cloud radiative forcing of subtropical low level clouds in global models
.
Climate Dyn.
,
30
,
779
788
, doi:.
Kaufman
,
Y. J.
,
D.
Tanré
, and
O.
Bocuher
,
2002
:
A satellite view of aerosols in the climate system
.
Nature
,
419
,
215
223
, doi:.
Kaufman
,
Y. J.
,
I.
Koren
,
L. A.
Remer
,
D.
Rosenfeld
, and
Y.
Rudich
,
2005
:
The effect of smoke, dust, and pollution aerosol on shallow cloud development over the Atlantic Ocean
.
Proc. Natl. Acad. Sci. USA
,
102
,
11 207
11 212
, doi:.
King
,
M. D.
, and Coauthors
,
2003
:
Cloud and aerosol properties, precipitable water, and profiles of temperature and water vapor from MODIS
.
IEEE Trans. Geosci. Remote Sens.
,
41
,
442
458
, doi:.
Kirkevåg
,
A.
, and Coauthors
,
2013
:
Aerosol–climate interactions in the Norwegian Earth System Model—NorESM1-M
.
Geosci. Model Dev.
,
6
,
207
244
, doi:.
Klein
,
S.
, and
D. L.
Hartmann
,
1993
:
The seasonal cycle of low stratiform clouds
.
J. Climate
,
6
,
1587
1606
, doi:.
Knight
,
C. G.
, and Coauthors
,
2007
:
Association of parameter, software, and hardware variation with large-scale behavior across 57,000 climate models
.
Proc. Natl. Acad. Sci. USA
,
104
,
12 259
12 264
, doi:.
Koren
,
I.
,
Y. J.
Kaufman
,
D.
Rosenfeld
,
L. A.
Remer
, and
Y.
Rudich
,
2005
:
Aerosol invigoration and restructuring of Atlantic convective clouds
.
Geophys. Res. Lett.
,
32
,
L14828
, doi:.
Koren
,
I.
,
L. A.
Remer
,
Y. J.
Kaufman
,
Y.
Rudich
, and
J. V.
Martins
,
2007
:
On the twilight zone between clouds and aerosols
.
Geophys. Res. Lett.
,
34
,
L08805
, doi:.
Krüger
,
O.
, and
H.
Graßl
,
2002
:
The indirect aerosol effect over Europe
.
Geophys. Res. Lett.
,
29
,
1925
, doi:.
Latham
,
J.
, and Coauthors
,
2008
:
Global temperature stabilization via controlled albedo enhancement of low-level maritime clouds
.
Philos. Trans. Roy. Soc.
,
366A
,
3969
3987
, doi:.
Liousse
,
C.
, and Coauthors
,
2010
:
Updated African biomass burning emission inventories in the framework of the AMMA-IDAF program, with an evaluation of combustion aerosols
.
Atmos. Chem. Phys.
,
10
,
9631
9646
, doi:.
Loeb
,
N. G.
, and
G. L.
Schuster
,
2008
:
An observational study of the relationship between cloud, aerosol and meteorology in broken low-level cloud conditions
.
J. Geophys. Res.
,
113
,
D14214
, doi:.
Martin
,
G. M.
,
D. W.
Johnson
, and
A.
Spice
,
1994
:
The measurement and parameterization of effective radius of droplets in warm stratocumulus clouds
.
J. Atmos. Sci.
,
51
,
1823
1842
, doi:.
Matsui
,
T.
,
H.
Masunaga
,
S. M.
Kreidenweis
,
R. A.
Pielke
Sr.
,
W.-K.
Tao
,
M.
Chin
, and
Y. J.
Kaufman
,
2006
:
Satellite-based assessment of marine low cloud variability associated with aerosol, atmospheric stability, and the diurnal cycle
.
J. Geophys. Res.
,
111
,
D17204
, doi:.
Menon
,
S.
,
A. D.
Del Genio
,
Y.
Kaufman
,
R.
Bennartz
,
D.
Koch
,
N.
Loeb
, and
D.
Orlikowski
,
2008
:
Analyzing signatures of aerosol–cloud interactions from satellite retrievals and the GISS GCM to constrain the aerosol indirect effect
.
J. Geophys. Res.
,
113
,
D14S22
, doi:.
Myers
,
T. A.
, and
J. R.
Norris
,
2013
:
Observational evidence that enhanced subsidence reduces subtropical marine boundary layer cloudiness
.
J. Climate
,
26
,
7507
7524
, doi:.
Myhre
,
G.
, and Coauthors
,
2007
:
Aerosol–cloud interaction inferred from MODIS satellite data and global aerosol models
.
Atmos. Chem. Phys.
,
7
,
3081
3101
, doi:.
Nakajima
,
T.
,
A.
Higurashi
,
K.
Kawamoto
, and
J. E.
Penner
,
2001
:
A possible correlation between satellite-derived cloud and aerosol microphysical parameters
.
Geophys. Res. Lett.
,
28
,
1171
1174
, doi:.
Noda
,
A. T.
, and
M.
Satoh
,
2014
:
Intermodel variances of subtropical stratocumulus environments simulated in CMIP5 models
.
Geophys. Res. Lett.
,
41
,
7754
7761
, doi:.
Ockert-Bell
,
M. E.
, and
D. L.
Hartmann
,
1992
:
The effect of cloud type on Earth’s energy balance: Results for selected regions
.
J. Climate
,
5
,
1157
1171
, doi:.
Penner
,
J. E.
,
L.
Xu
, and
M.
Wang
,
2011
:
Satellite methods underestimate indirect climate forcing by aerosols
.
Proc. Natl. Acad. Sci. USA
,
108
,
13 404
13 408
, doi:.
Peters
,
K.
,
J.
Quaas
, and
H.
Gral
,
2011
:
A search for large-scale effects of ship emissions on clouds and radiation in satellite data
.
J. Geophys. Res.
,
116
,
D24205
, doi:.
Peters
,
K.
,
J.
Quaas
,
P.
Stier
, and
H.
Graßl
,
2014
:
Processes limiting the emergence of detectable aerosol indirect effects on tropical warm clouds in global aerosol–climate models and satellite data
.
Tellus
,
66B
,
24054
, doi:.
Platnick
,
S.
,
M. D.
King
,
S. A.
Ackerman
,
W. P.
Menzel
,
B. A.
Baum
,
J. C.
Riedi
, and
R. A.
Frey
,
2003
:
The MODIS cloud products: Algorithms and examples from Terra
.
IEEE Trans. Geosci. Remote Sens.
,
41
,
459
473
, doi:.
Prospero
,
J. M.
,
P.
Ginoux
,
O.
Torres
,
S. E.
Nicholson
, and
T. E.
Gill
,
2002
:
Environmental characterization of global sources of atmospheric soil dust identified with the Nimbus 7 Total Ozone Mapping Spectrometer (TOMS) absorbing aerosol product
.
Rev. Geophys.
,
40
,
1002
, doi:.
Quaas
,
J.
, and Coauthors
,
2009
:
Aerosol indirect effects general circulation model intercomparison and evaluation with satellite data
.
Atmos. Chem. Phys.
,
9
,
8697
8717
, doi:.
Ramanathan
,
V.
,
R. D.
Cess
,
E. F.
Harrison
,
P.
Minnis
,
B. R.
Barkstrom
,
E.
Ahmad
, and
D.
Hartmann
,
1989
:
Cloud-radiative forcing and climate: Results from the Earth Radiation Budget Experiment
.
Science
,
56
,
57
63
, doi:.
Remer
,
L. A.
, and Coauthors
,
2005
:
The MODIS aerosol algorithm, products and validation
.
J. Atmos. Sci.
,
62
,
947
973
, doi:.
Remer
,
L. A.
, and Coauthors
,
2008
:
Global aerosol climatology from the MODIS satellite sensors
.
J. Geophys. Res.
,
113
,
D14S07
, doi:.
Roberts
,
G. C.
,
M. V.
Ramana
,
C.
Corrigan
,
D.
Kim
, and
V.
Ramanathan
,
2008
:
Simultaneous observations of aerosol–cloud–albedo interactions with three stacked unmanned aerial vehicles
.
Proc. Natl. Acad. Sci. USA
,
105
,
7370
7375
, doi:.
Rosenfeld
,
D.
, and
G.
Feingold
,
2003
:
Explanation of discrepancies among satellite observations of the aerosol indirect effects
.
Geophys. Res. Lett.
,
30
,
1776
, doi:.
Rosenfeld
,
D.
,
S.
Sherwood
,
R.
Wood
, and
L.
Donner
,
2014
:
Climate effects of aerosol–cloud interactions
.
Science
,
343
,
379
380
, doi:.
Schwartz
,
S. E.
,
1988
:
Are global cloud albedo and climate controlled by marine phytoplankton?
Nature
,
336
,
441
445
, doi:.
Shindell
,
D.
, and Coauthors
,
2013
:
Radiative forcing in the ACCMIP historical and future climate simulations
.
Atmos. Chem. Phys.
,
13
,
2939
2974
, doi:.
Stevens
,
B.
, and
J.-L.
Brenguier
,
2009
: Cloud-controlling factors: Low clouds. Clouds in the Perturbed Climate System, J. Heintzenberg and R. Charlson, Eds., MIT Press, 173–196.
Stevens
,
B.
, and
G.
Feingold
,
2009
:
Untangling aerosol effects on clouds and precipitation in a buffered system
.
Nature
,
461
,
607
613
, doi:.
Stubenrauch
,
C. J.
, and Coauthors
,
2013
:
Assessment of global cloud datasets from satellites: Project and database initiated by the GEWEX radiation panel
.
Bull. Amer. Meteor. Soc.
,
94
,
1031
1049
, xxxx.
Taylor
,
K. E.
,
R. J.
Stouffer
, and
G. A.
Meehl
,
2012
:
An overview of CMIP5 and the experiment design
.
Bull. Amer. Meteor. Soc.
,
93
,
485
498
, doi:.
Twomey
,
S. A.
,
1974
:
Pollution and the planetary albedo
.
Atmos. Environ.
,
8
,
1251
1256
, doi:.
Twomey
,
S. A.
,
1977
:
The influence of pollution on the shortwave albedo of clouds
.
J. Atmos. Sci.
,
34
,
1149
1152
, doi:.
Waquet
,
F.
,
F.
Peers
,
F.
Ducos
,
P.
Goloub
,
S.
Platnick
,
J.
Riedi
,
D.
Tanré
, and
F.
Thieuleux
,
2013
:
Global analysis of aerosol properties above clouds
.
Geophys. Res. Lett.
,
40
,
5809
5814
, doi:.
Wen
,
G.
,
A.
Marshak
,
R. F.
Cahalan
,
L. A.
Remer
, and
R. G.
Kleidman
,
2007
:
3-D aerosol–cloud radiative interaction observed in collocated MODIS and ASTER images of cumulus cloud fields
.
J. Geophys. Res.
,
112
,
D13204
, doi:.
Wielicki
,
B. A.
,
B. R.
Barkstrom
,
E. F.
Harrison
,
R. B.
Lee
III
,
G. L.
Smith
, and
J. E.
Cooper
,
1996
:
Clouds and the Earth’s Radiant Energy System (CERES): An Earth Observing System Experiment
.
Bull. Amer. Meteor. Soc.
,
77
,
853
868
, doi:.
Wilcox
,
E. M.
,
2010
:
Stratocumulus cloud thickening beneath layers of absorbing smoke aerosol
.
Atmos. Chem. Phys.
,
10
,
11 769
11 777
, doi:.
Wilcox
,
E. M.
,
2012
:
Direct and semi-direct radiative forcing of smoke aerosols over clouds
.
Atmos. Chem. Phys.
,
12
,
139
149
, doi:.
Winker
,
D. M.
,
J. L.
Tackett
,
B. J.
Getzewich
,
Z.
Liu
,
M. A.
Vaughan
, and
R. R.
Rogers
,
2013
:
The global 3-D distribution of tropospheric aerosols as characterized by CALIOP
.
Atmos. Chem. Phys.
,
13
,
3345
3361
, doi:.
Wood
,
R.
,
2012
:
Stratocumulus clouds
.
Mon. Wea. Rev.
,
140
,
2373
2423
, doi:.
Zeng
,
S.
,
J.
Riedi
,
C. R.
Trepte
,
D. M.
Winker
, and
Y.-X.
Hu
,
2014
:
Study of global cloud droplet number concentration with A-Train satellites
.
Atmos. Chem. Phys.
,
14
,
7125
7134
, doi:.