The effect of the environment on individual clouds makes it difficult to isolate the signal of the aerosol indirect effect, particularly at larger spatial and temporal scales. This study uses observations from the Tropical Rainfall Measuring Mission (TRMM), CloudSat, and Aqua satellites to identify convective cloud systems in clean and dirty environments. The Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol index is collocated with radar precipitation features (RPFs) from TRMM and congestus cloud features (CFs) from CloudSat. The Interim ECMWF Re-Analysis (ERA-Interim) is interpolated to identify the environmental profile surrounding each feature. Regions in Africa, the Amazon, the Atlantic Ocean, and the southwestern United States are examined. TRMM features in the Africa and Amazon regions are more intense and have higher lightning flash rates under dirty background conditions. RPFs in the southwestern United States are more intense under clean background conditions. The Atlantic region shows little difference in intensity. The differences found in the mean thermodynamic profile for RPFs forming in clean and dirty environments could explain these differences in convective intensity.
Congestus identified with CloudSat show smaller differences between clouds forming in clean and dirty environments in Africa and the Amazon. Congestus in clean environments have higher reflectivities and generally larger widths, but no trend is seen in cloud-top height. The signal of the aerosol indirect effect is so small that it is very difficult to detect confidently using these methods. The environment must be considered in any study of the aerosol indirect effect, because important environmental changes can occur as aerosols are introduced to an air mass.
The impact of aerosols on the microphysical and dynamic properties of clouds has been much debated since the first and second aerosol indirect effects were introduced by Twomey (1977) and Albrecht (1989). Many studies have examined both the modeling and observational sides of aerosol–cloud interactions. So far, the overwhelming result can be summed up in this statement from Khain (2009, p. 2): “The effect of aerosols on precipitation is a very complicated multi-scale problem.”
This problem can be addressed in a multitude of ways. Aerosols are known to affect the climate through the absorption and scattering of solar radiation and also through the modification of cloud properties such as cloud drop size distribution (DSD), which can increase the reflectance of the cloud (Twomey 1977; van den Heever et al. 2011). The effects of aerosols on microphysical and dynamical processes that affect precipitation production are equally important, particularly when attempting to determine possible changes in the hydrologic cycle caused by increasing amounts of aerosols and particulates in the atmosphere. The effects of aerosols on warm rain, or rain forming from coalescence as all parts of the cloud are warmer than 0°C (Glickman 2000), and shallow clouds have been studied considerably, through both observations and modeling (Gunn and Phillips 1957; Rosenfeld 1999). In most cases, aerosols suppress warm rain (Rosenfeld 1999; Andreae et al. 2004; Lebsock et al. 2008; Berg et al. 2008). This phenomenon has been observed in the Amazon due to smoke from biomass burning (Andreae et al. 2004). Unfortunately, the effect of aerosols on deep convection is much more complicated. When ice processes are involved, clouds can be invigorated by aerosols (Koren et al. 2008, 2010; Yuan et al. 2011a; Heiblum et al. 2012; Niu and Li 2011; Storer and van den Heever 2013; Storer et al. 2013, manuscript submitted to J. Geophys. Res.). The objective of this study is to quantify the indirect effect of aerosols on convective clouds using satellite observations, if such an aerosol signal can be observed. We will accomplish this by looking at deep, precipitating convective clouds using the Tropical Rainfall Measuring Mission (TRMM). To investigate the effects of aerosols on shallower clouds with less ice, we will use CloudSat to identify cumulus congestus clouds.
The microphysical changes that occur due to increased aerosol concentration produce changes in storm dynamics. Polluted clouds have a larger number of cloud droplets (Sekiguchi et al. 2003). The larger number of smaller droplets allows greater release of latent heat of fusion as the droplets accrete onto ice crystals, and this can strengthen the updraft (Zipser 2003) and cause higher cloud tops (Yuan et al. 2011a,b; Storer et al. 2013, manuscript submitted to J. Geophys. Res.) and a greater likelihood for overshooting tops (Andreae et al. 2004, Rosenfeld et al. 2007). This dynamic response has been simulated in computational models as increases in both updraft speed and frequency (Khain and Pokrovsky 2004; van den Heever et al. 2006, 2011; Storer and van den Heever 2013). Morrison and Grabowski (2011) find higher anvils from polluted clouds, as smaller ice particles have a slower sedimentation velocity.
Several observational studies have looked at the effects of aerosols in individual regions of the world, typically during specific time periods. Lin et al. (2006) found that an increase in Moderate Resolution Imaging Spectroradiometer (MODIS) aerosol optical depth was correlated with an increase in precipitation, increased occurrence of intense rain events, enhanced cloud cover, elevated cloud-top heights, increased cloud water path, and greater formation of ice in the Brazilian Amazon during the burning seasons (August–October) of 2000 and 2003. Rosenfeld (1999) also saw suppression of warm rain in observations of smoky clouds and an abundance of precipitation in observations of relatively smoke-free clouds over Indonesia due to biomass burning. Jiang et al. (2008) linked polluted clouds during the dry season in South America to suppressed precipitation and reduced effective radius of cloud droplets. Sekiguchi et al. (2003) found that, while summer pollution in China significantly suppresses the growth of cloud particles, the correlation between cloud-top temperature and aerosol number concentration can be positive or negative depending on season and location. Heiblum et al. (2012) use the TRMM satellite to find increases in the vertical distribution of rain with increasing MODIS aerosol optical depth. Berg et al. (2006, 2008) attributed differences between TRMM’s passive microwave- and radar-based estimates of rainfall to the effect of aerosols. Huang et al. (2009) found significant negative correlations between aerosols and precipitation in both observations and model simulations in Africa. Koren et al. (2008, 2010) found that higher aerosol concentrations overall influence the distribution of clouds in the tropical Atlantic toward higher clouds and larger cloud fractions. Niu and Li (2011) use 1 year of the MODIS (King et al. 2003) aerosol index (AI; Nakajima et al. 2001) and CloudSat and find conditional correlations between AI and cloud-top temperature and height, cloud ice water path, and precipitation rate in the tropics. Williams et al. (1999), Altaratz et al. (2010), and Yuan et al. (2011a) found an increase in lightning activity with increased aerosol concentrations.
In all of the studies that have examined the relationship between convection and increased aerosol concentrations, many use numerical models, which rely on various microphysical parameterizations including bin schemes, convective parameterizations, and other assumptions with varying degrees of uncertainty. These results are invaluable in developing a conceptual model of the effects of aerosols, but observational studies are critical to support the ideas that originate or are tested with model runs. Obvious problems arise when designing such a study. First, there are no clouds (or cloud systems) in nature that differ only by aerosol amount (Khain 2009). Isolating the effects of aerosols from those of the thermodynamic environment is very difficult, particularly with satellite observations that only see convective features for one moment in time. Causality is extremely challenging to determine from sensors that see a snapshot, and we assume that convective features form in the environment observed by MODIS. Additionally, the sensitivity of clouds to aerosol effects can appear weaker due to “buffering:” changes in an isolated cloud due to aerosols may be cancelled out by opposing changes from other clouds when a large area is looked at on longer temporal scales (Stevens and Feingold 2009). Yet, determining the influences of aerosols on convective clouds is an important task.
Many past studies have focused on very specific regions for limited time periods of only a few years. The temporal and spatial scales of the TRMM satellite and the A-Train constellation of satellites allow a climatology of over 14 years of convection in the tropics. This study uses convective features from 10 years of TRMM data and 5 years of CloudSat data to examine multiple regions over a longer time period. Additionally, we integrate reanalysis to analyze the background environment, which is very important to separate out the aerosol indirect effect. Using these satellites, can we detect the aerosol indirect effect, and if so can we determine how aerosols affect the intensity of convective clouds in specific regions of the tropics?
2. Data and methods
a. TRMM satellite
The Precipitation Radar (PR) on TRMM is the first of its kind in space, and it allows the opportunity to see precipitation in three dimensions in the tropics (Kummerow et al. 1998). It has a 250-km swath, and its range extends from 36°S to 36°N. The TRMM Microwave Imager (TMI) is a passive microwave radiometer with nine channels. TRMM data covering 14 years (1998–2011) help compensate for infrequent sampling and allow for the construction of robust climatological statistics. To investigate the properties of convective clouds, snapshots of precipitation systems observed by TRMM are summarized in radar precipitation features (RPFs), which are identified by contiguous near-surface raining pixels (Iguchi et al. 2000). Some of the properties defined for RPFs include maximum height of the 20-, 30-, and 40-dBZ echoes; volumetric rainfall from PR (Iguchi et al. 2009); minimum polarization-corrected temperature (PCT), which is brightness temperature corrected for background scattering (Spencer et al. 1989) for 37 and 85 GHz as seen by the TMI; area of the RPF; and the geocenter location of the RPF. Mean rain rate was calculated by dividing volumetric rainfall by area, and convective percentage was calculated using the TRMM 2A23 algorithm (Awaka et al. 1998, 2009). Lightning flash count was determined using the Lightning Imaging Sensor (LIS), which is a staring optical imager that identifies changes in radiances in the field of view (Christian et al. 2000). Some of these parameters can be used as proxies for convective intensity and make TRMM an excellent tool for showing changes in storm strength. All TRMM data were obtained from the University of Utah TRMM database (Liu et al. 2008). In this study, we require RPFs to have a minimum area of four pixels, or around 80 km2 to remove the noise signals. Only RPFs with collocated MODIS AI (see section 2d) data during 2002–11 are used.
b. CloudSat satellite
CloudSat is part of the A-Train constellation, which includes five satellites flying in formation so that they follow closely behind one another (Stephens et al. 2002). CloudSat carries the Cloud Profiling Radar (CPR), a 94-GHz near-nadir-pointing cloud radar, which measures the vertical structure of clouds, producing a two-dimensional cross section along the satellite track (Marchand et al. 2008). Cloud features have been identified using the CloudSat geometric profile product (2B-GEOPROF), which contains cloud mask and reflectivity at a resolution of 1.1 km along track × 1.3 km across track with a minimum detectable signal of −28 dBZ (Stephens et al. 2002; Mace et al. 2007). All CloudSat data products were downloaded from the CloudSat Data Processing Center (http://www.cloudsat.cira.colostate.edu/) for the years 2006–11. A cloud feature (CF) is defined as 10 or more contiguous pixels with reflectivity greater than −28 dBZ and a cloud mask value of at least 20. Cloud mask is a product that identifies whether the radar return is likely to be cloud or noise (Marchand et al. 2008). A value of 20 has a lesser chance of false detection of a cloud than a lower number. The reflectivity and cloud mask criteria identify clouds, while the area restriction helps filter out noisy pixels.
We have selected a subset of these CloudSat CFs from 2006 to 2011 to represent congestus clouds. Our congestus have tops between 5 and 8 km, a cloud echo base less than 1.5 km above the terrain height, a maximum reflectivity of at least −5 dBZ, cloud thickness of at least 4 km, and an along-track width of less than 30 CloudSat pixels (33 km). Cloud width is the maximum distance that the echo extends along the satellite track. It should be noted that these congestus are not individual turrets, or even individual clouds. The 1.1-km resolution of CloudSat smears congestus together. Therefore, we refer to these clouds as “congestus groups” in recognition of their size. Additional details about the population of congestus groups utilized here can be found in Wall et al. (2013).
To add meteorological context to the RPFs and CFs defined using TRMM and CloudSat, data from the Interim European Centre for Medium-Range Weather Forecasts (ECMWF) Re-Analysis (ERA-Interim; Dee et al. 2011) has been collocated to each feature. Three-hourly data with a resolution of 1.5° were obtained from the ECMWF website. The mean latitude and longitude of each cloud feature were used to find the nearest ERA-Interim grid point, and the data from this point were linearly interpolated to match the time of the TRMM or CloudSat overpass of the cloud feature. Geopotential height, relative humidity, horizontal and vertical winds, and temperature are considered at 10 levels (1000, 975, 925, 850, 700, 500, 400, 300, 200, 100 hPa). Mean thermodynamic soundings were created using temperature and relative humidity from all 37 available levels from the ERA-Interim.
d. MODIS aerosol products
AI was used as a measure of the background aerosol concentration around our RPFs and CFs and was retrieved from MODIS, aboard the Aqua satellite, which was launched in 2002 (King et al. 2003). It provides daily coverage on a global level, with overpass times around 0130 and 1330 local time (LT). The corrected aerosol optical depth (AOD) for land, mean effective optical depth for ocean, Angstrom exponents for land and ocean, and number of pixels used in the calculation of each level-3 1° bin were obtained from the Level 1 and Atmosphere Archive and Distribution System (LAADS) (Kaufman et al. 1997; Tanré et al. 1997; Levy et al. 2003, Remer et al. 2008). AOD has been correlated to cloud condensation nuclei (CCN) concentration on a first order (Nakajima et al. 2001; Rosenfeld et al. 2008). The MODIS AI is the product of AOD and the Angstrom exponent and better reflects aerosol characteristics because it takes into account particle size (Nakajima et al. 2001). Seasonal values of MODIS AI are shown in Figs. 1a–d. Note the maxima due to biomass burning in South America and Africa in September–November. The East Coast of the United States and China have higher values of AI in Northern Hemisphere spring and summer. Land areas tend to be more polluted than ocean areas, but the Atlantic has higher values of AI during June–August. The southern Pacific and Indian Oceans also have higher values during Southern Hemisphere spring and summer.
AOD and AI are proxies for CCN, and the reliability of assuming that the two are related depends on the uniformity of the aerosol size, composition, and vertical distribution (Koren et al. 2010; Tao et al. 2012). MODIS provides no information about aerosol composition, and yet different aerosol types are expected to produce different aerosol effects (Wen et al. 2006; Yuan et al. 2008; van den Heever et al. 2006). Further uncertainties are derived from undetected clouds, aerosol humidification, and so-called 3D cloud effects, which can yield a larger AOD when measured from space (Koren et al. 2010). Deliquescence of aerosol in regions of high relative humidity makes it difficult to determine how much of the scattering contribution to AOD comes from aerosol and how much comes from water (Gassó et al. 2000). Brennan et al. (2005) found that cloud contamination begins with AOD values of 0.6, so RPFs/CFs with AOD larger than 0.6 were excluded from this study.
MODIS cannot measure AOD in or near cloud, so we are forced to make the assumption that aerosol properties on a larger scale are similar enough to aerosol properties near the cloud to compute useful relationships between aerosols and cloud properties (Jones and Christopher 2010). Another assumption involves the timing of satellite overpasses. The values of AOD/AI come from Aqua MODIS, which has an overpass time of 1330 LT (AOD is only calculated for daytime overpasses). TRMM does not have a sun-synchronous orbit, so RPFs are observed at any hour, leading to a time gap between the TRMM observation of an RPF and the MODIS overpass. Using a wide area to average values of AOD and investigating regions known for high values of AI on a large scale should help mitigate this problem.
The collocated value of MODIS AI was calculated for each 1° bin by multiplying the AOD by the Angstrom exponent. A mean background AI was calculated for each RPF/CF by averaging all good values within 2° of the centroid of the feature. At least 8 of 16 possible data points must exist within the 4° box. Because we are focusing on regions of large-scale aerosol transport, we assume that the mean value of AI within this region is representative of the large-scale environment.
e. Regions of interest
The environment significantly influences convective clouds of the sort that we examine in this study and in many cases plays a larger role in cloud dynamics than aerosols (Khain et al. 2005; Khain 2009; Jones and Christopher 2010; van den Heever et al. 2011). Properties of convective clouds are different in different regions and for different seasons (Liu and Zipser 2008). Additionally, the effects of aerosols are expected to be different in different regions (Stevens and Feingold 2009; Koren et al. 2012). For these reasons, we have selected four specific regions for detailed study. These regions are shown in Fig. 1e. We look at September–November in the Amazon and Africa regions and June–August in the Atlantic and North American monsoon (NAM) regions. The Amazon, Africa, and Atlantic regions were selected because they have been examined in previous studies (Andreae et al. 2004; Huang et al. 2009; Koren et al. 2010). The NAM region was selected because a large number of RPFs occur in background environments with high AI during the monsoon. The origin of these aerosols is unknown. The seasons were selected because of a greater presence of aerosols, which results in an increased sample size of features occurring in “dirty” conditions.
Biomass burning in the Amazon releases fine-mode sulfur particles, black carbon, mercury, iron, and other particulate matter, and during biomass burning season, values of aerosol optical thickness increase drastically (Artaxo et al. 2009). A significant fraction (40%–60%) of these aerosol particles can act as CCN (Gunthe et al. 2009). In Africa, anthropogenic-biomass-burning particulate matter and natural mineral dust can become mixed to varying degrees. Studies have shown that during the fire season, an elevated layer of smoke is present above a dust-rich surface layer (Liousse et al. 2010). During the summer, mineral dust and anthropogenic pollution from Europe is advected over the central Atlantic Ocean from the Sahara (Garrett et al. 2003). The source of aerosols in the NAM region remains unclear.
Within these regions, we set limits to define “clean” and “dirty” regimes based on the background value of AI retrieved with MODIS. These limits are shown in Table 1. Figure 2 shows a histogram of the percentage of the population having a certain AI value and a box plot of the numbers of features for a given AI. Several methods were tested to define “clean” and “dirty.” Using percentages of the population, such as the upper and lower quartile or 10%, is problematic because of the large number of RPFs with very clean background AI values over the ocean. Using these cutoffs, features that are dirty over the ocean would be considered clean over land. The histogram of AI over the Atlantic in Fig. 2a shows dramatic differences when compared to the land regions—there are fewer aerosol sources over the ocean. The NAM region also tends to have lower values of AI than the Amazon and Africa regions. To make comparisons between regions and to search for background AI values that are consistent with the idea that “clean” refers to a background with a small number of available aerosols that can act as CCN and that “dirty” refers to an environment with a multitude of available CCN, we have selected our own limits for clean and dirty regimes. Sensitivity tests of our AI limits as well as tests of using AOD rather than AI show that our results are not especially sensitive to the limits used for these definitions.
a. Convective intensity differences in TRMM RPFs under clean and dirty environments
Because we want to determine whether aerosols invigorate deep convection, we look at proxies for convective intensity: flash count, brightness temperatures, maximum reflectivity, rain rate, and volumetric rainfall. Table 2 shows the differences in many properties of RPFs in clean and dirty regimes for each region. The response of RPFs to increased AI varies for each region. For the Africa and Amazon regions, RPFs in a dirty environment have a greater number of mean lightning flashes, a greater maximum number of lightning flashes, and a greater percentage of RPFs with at least one and at least five flashes. The difference in mean flash rates between clean and dirty RPFs for these regions is statistically significant at the 99% level. This is consistent with past studies by Yuan et al. (2011a), who found an increase in lightning activity with increased aerosol loading, and Williams et al. (1999) found that larger CCN concentrations lead to higher maximum flash rates. However, the NAM region shows the opposite trend from Africa and the Amazon—clean RPFs have more flashes.
The 85-GHz PCT provides information about the amount of ice scattering within the cloud. In general, the greater the brightness temperature depression at 85 and 37 GHz, the greater the ice water path (Spencer et al. 1989; Vivekanandan et al. 1991) and the greater the convective intensity (Zipser et al. 2006). In the Amazon region, RPFs occurring in a dirty environment have lower 85-GHz PCTs, indicating more ice scattering. In the Africa region, dirty RPFs have lower minimum and mean 85-GHz PCTs, while in the Atlantic and NAM regions, clean RPFs have lower 85-GHz PCT and more ice scattering. At 37 GHz, similar trends are observed, although in general the numbers for clean and dirty RPFs are much closer.
Aerosols have the potential to produce larger rain rates later in a cloud’s life cycle because of the delay in the development of raindrops by suppressed collision–coalescence. TRMM only sees a snapshot of each RPF, so we cannot tell at what stage in the life cycle they may be, but over a large number of RPFs, Table 2 shows that in the Amazon, Africa, and Atlantic regions, dirty RPFs have higher mean rain rates. Differences in mean rain rates for clean and dirty RPFs are statistically significant at the 99% level for every region but the NAM region. Volumetric rainfall, which is the product of rain rate and area, does not show any conclusive trend. In Africa and the Amazon, RPFs forming in a clean environment have larger minimum and mean volumetric rainfalls, but smaller maximum volumetric rainfall amounts. In the NAM region, clean RPFs have larger minimum, mean, and maximum volumetric rainfall.
Overall, the results from Table 2 indicate that in both Africa and the Amazon, RPFs forming in dirty environments could be considered more intense. They have more lightning flashes, a greater amount of ice scattering, and larger rain rates. In the Atlantic, there is very little change from clean to dirty, although dirty RPFs have larger mean rain rates. In the NAM region, clean RPFs are more intense by each proxy utilized here.
What about convective intensity shown by radar reflectivity? Figure 3 shows profiles of maximum reflectivity at each level for each region. The solid dots on the profile indicate levels at which distributions of clean and dirty reflectivities are statistically different at the 99% level using a Komolgorov–Smirnov test. Amazon and Africa (Figs. 3a and 3b) show large differences between clean and dirty RPFs. Even at the median maximum reflectivity, these regions have at least 0.5 km in difference in echo-top height. Difference in maximum reflectivity reaches 5 dBZ at some levels. These results for Amazon and Africa are consistent with previous studies that have found higher cloud tops and enhanced reflectivities in dirty clouds (Koren et al. 2010; Yuan et al. 2011a; Koren et al. 2012; Storer and van den Heever 2013; Storer et al. 2013, manuscript submitted to J. Geophys. Res.).
The Atlantic and NAM regions, Figs. 3c and 3d, respectively, show smaller differences in reflectivity profiles. In the Atlantic, clean RPFs have slightly taller echo-top heights, although dirty RPFs have higher maximum reflectivities within the top 10% of the profiles. In the NAM region, the median profile shows that clean RPFs have higher echo-top heights and slightly higher reflectivities, but the differences remain much less than those for the Amazon and Africa regions.
Because we selected different criteria for clean and dirty regimes over different regions, the above-mentioned results cannot be fairly compared among regions. Therefore, we demonstrate how properties of the RPFs change against their absolute AI values for all regions in Fig. 4. Note that as AI increases, sample size decreases (see Fig. 2b). Both the Amazon and Africa show increasing trends—the mean 20-, 30-, and 40-dBZ echo-top height increases with increasing AI. Both the NAM and Atlantic regions show a decreasing trend and then an anomalous jump in echo-top height that likely result from the limited sample size. The correlation between AI and echo-top height for these regions remains quite low, with the highest value of 0.157 occurring between the AI and 40-dBZ echo top in the Amazon. The Amazon region has the highest correlation between the AI and echo-top height, followed by Africa.
b. Large-scale meteorological environment of RPFs under clean and dirty conditions
Thus far, the differences between clean and dirty RPFs show that in some regions, dirty RPFs are more intense, but each region has different characteristics. We have not yet made any attempt to account for environment, which plays a large role in determining the intensity of convective clouds. If aerosols are advected into a region along with a different air mass, or if the aerosols modify the air mass through absorption of radiation (the semidirect effect), then differences in convective intensity could be more closely related to those differences in the environment rather than to the aerosol indirect effect. Figure 5 shows the mean soundings for clean (blue) and dirty (red) RPFs in each region. The dots indicate the significant levels at which permutation tests showed both temperature and relative humidity distributions to be statistically significant at the 99% level.
The Africa (Fig. 5b) and NAM (Fig. 5d) regions exhibit the most substantial differences between the clean and dirty profiles. In Africa the mean profile for dirty RPFs shows warmer surface temperatures and drier air in the midtroposphere. Such an environment would inhibit weak convection. Perhaps RPFs in this environment are more intense because they must have strong forcing to overcome convective inhibition. Also, there are differences in low-level winds between clean and dirty RPFs, which suggest different air masses at low levels for clean and dirty RPFs.
The Amazon region (Fig. 5a) does show small differences between the profiles—dirty RPFs tend to have drier air in the midtroposphere. Again, this dry air in the midtroposphere is a significant barrier to weak convection (Malkus and Riehl 1964). Dry air over moist surface air could lead to increased lapse rates, meaning that once convection is able to overcome the convective inhibition, it will be more intense. Surface conditions remain similar for both profiles, but the dry air in the midtroposphere in dirty soundings for both the Amazon and Africa could contribute to more intense convection occurring in these environments.
Differences between clean and dirty profiles in the Atlantic region (Fig. 5c) are very small, although the distributions are still statistically different at almost all significant levels.
The NAM area is quite different from the other regions examined here. The mean sounding for dirty conditions is cooler and moister throughout the lower part of the troposphere. The clean and dirty soundings look remarkably like mean monsoon break and burst soundings, respectively (Wall 2009). This could explain why clean RPFs are more intense. During a burst period, monsoon-induced convection is widespread throughout the region. Outflow from intense thunderstorms could transport dust, causing high values of AI and creating a dirty background (Seigel and van den Heever 2012). Widespread weak convection during burst periods could cause a large number of weak RPFs to influence and reduce the mean values for lightning flashes and brightness temperatures. During break periods convection is less widespread, causing less dust lofted by outflow, and resulting in a greater population of clean RPFs and a smaller number of weak clouds. This idea is purely speculative. This region needs additional study to determine the composition and source of the aerosols detected by MODIS.
In addition to examining the mean soundings shown in Fig. 5, we investigated the differences between more extreme environments producing clean and dirty RPFs. For each region, soundings were plotted for cases with the top and bottom 10% of CAPE values as well as the top and bottom 10% of mean 700–500-hPa relative humidity (not shown). For the Atlantic, Amazon, and NAM regions, the differences between these clean and dirty “extreme” soundings were consistent with the changes seen in Fig. 5. For the Amazon and Africa, dirty environments showed drier air at midlevels. For the NAM region, clean environments are again warmer and drier. The Atlantic region was slightly different—for the driest 10% of soundings, clean environments had drier air. For the top 10% of midlevel relative humidity cases, the soundings look very much like those shown in Fig. 5.
Figure 5 shows differences in mean soundings for clean and dirty RPFs that make it impossible to attribute increases in convective intensity solely to the aerosol indirect effect. The environment surrounding the RPFs changes enough that it is likely causing differences in convective intensity (more lightning flashes and more ice scattering) in at least some instances. To account for these environmental differences, we must group RPFs by environmental factors.
Figure 6 shows how mean midlevel (700–500 hPa) relative humidity and convective available potential energy (CAPE) affect convective intensity using a modification parameter (MP). This modification parameter is created using volumetric rainfall (VR), area, convective percentage (CP), 20-dBZ echo top (ETH), and 85-GHz PCT (PCT). The modification parameter is calculated using the following equation:
The difference in means for clean and dirty RPFs is found and normalized by the standard deviation σ. The 85-GHz PCT term is multiplied by −1 since lower brightness temperatures indicate a more intense cloud. Higher (lower) values are shown in blue (red) and indicate that clean (dirty) RPFs are more intense. Note that at least 10 clean and 10 dirty RPFs must be present in each bin.
In the Amazon and Africa regions (Figs. 6a and 6b), dirty RPFs tend to be more intense under most environments. However, when CAPE is around 1800–2000 J kg−1, clean RPFs are more intense over the Amazon. When midlevel relative humidity is around 70%–80%, many clean RPFs are more intense over Africa. In the Atlantic (Fig. 6c), dirty RPFs are more intense in many environments, but the differences are much smaller. In the NAM region (Fig. 6d), clean RPFs tend to be more intense, but for some values of relative humidity and CAPE, dirty RPFs are stronger. Modeling studies have shown that CAPE and relative humidity can impact the amount that aerosols invigorate convective clouds (Khain et al. 2005; Lee et al. 2008; Lee 2011; Fan et al. 2009). Our results do not show a preferred environment for invigoration, but they do support the generalizations that in the Amazon and Africa, adding aerosols to the background concentration could invigorate convective clouds, while in the NAM region RPFs in clean environments tend to be more intense. Aerosols affect cloud microphysics and dynamics, and these changes manifest themselves in different ways depending on many other environmental factors, including the type of cloud involved (van den Heever et al. 2011). These factors, such as instability, humidity, shear, and cloud type, make it very difficult to predict what effect aerosols may have on a cloud.
c. Differences in CloudSat congestus under clean and dirty environments
Congestus clouds are smaller than their deep convective counterparts, which we have observed with TRMM. Aerosols are known to affect warm clouds. The extent of aerosol influences on clouds with some ice is uncertain. The results of Rosenfeld and Lensky (1998) show development of a mixed-phase region in their satellite-observed clouds between −5° and −10°C in most cases. Observations from western Pacific convective clouds show that by −10°C, congestus certainly contain ice, but that clouds that only briefly reach −5°C before falling back to lower altitudes continue to be composed of water (Rangno and Hobbs 2005). In the vicinity of the Kwajalein Atoll, penetration of deep convective clouds revealed that it was rare to observe supercooled liquid water at temperatures less than −12°C, suggesting that liquid water freezes at warmer temperatures (Stith et al. 2002).
Figure 7 shows the histograms of AI for CloudSat congestus in the Amazon and Africa regions (shown in Fig. 1e). Africa has a higher maximum of AI, but both regions look similar. Again, clean congestus are defined as having a MODIS background AI of 0.0–0.2, and dirty congestus are defined as having a MODIS background AI of 0.6–1.5.
Table 3 shows the differences in several properties of CloudSat congestus groups in clean and dirty backgrounds. The results for Africa and the Amazon are different. In Africa, clean congestus groups have higher cloud tops, while in the Amazon, congestus in dirty environments have higher tops. In both regions dirty congestus tend to have lower bases, but the difference is slight in the Amazon. In looking at cloud echo base, we must consider that CloudSat cannot differentiate between cloud and rain. Cloud echo base is therefore the lowest detectable echo. Dirty congestus tend to be thicker in the Amazon region, while clean congestus tend to be thicker in Africa.
Results for cloud width and maximum reflectivity are similar for both regions—mean cloud widths and maximum cloud widths are larger for dirty congestus groups. Additionally, minimum and mean values for maximum reflectivity are smaller for dirty congestus groups. This result corresponds with the idea that adding aerosols can suppress collision–coalescence, resulting in smaller raindrops for clouds forming in dirty environments. A distribution of smaller raindrops would result in smaller maximum reflectivities, as we observe. Fan et al. (2007) found higher radar reflectivities in simulated maritime cumulus clouds compared to continental cumulus clouds produced under identical dynamic and thermodynamic conditions.
Figure 8 shows profiles of maximum reflectivity for clean and dirty congestus groups in the Amazon and Africa regions. For both regions clean congestus have higher reflectivities on almost every vertical level. The means of maximum reflectivity for clean and dirty congestus are statistically significant at the 99% level using a Kolmogorov–Smirnov test below 4 km for the Amazon region and only at 4–5 km for the Africa region. Thus, even though reflectivity is the parameter that shows the greatest differences between congestus forming in clean and dirty environments, the differences found are not statistically significant.
What about the environment? Figure 9 shows the mean soundings for clean and dirty congestus in the Amazon and Africa regions. The temperature profiles are nearly identical, and the distributions of temperature are only statistically different at 700 and 150 hPa in the Amazon and none in Africa. The only differences in the soundings are in moisture in the midtroposphere. At these levels, the distributions and means of relative humidity are statistically significantly different at the 99% level. Dirty congestus occur in environments with drier air at midlevels. Dry air in the midtroposphere could inhibit the growth of congestus into deeper clouds, as clouds grow and begin to entrain drier air in dirty environments. In the Amazon region (Fig. 9a), drier air begins by 700 hPa. This dry air would lead us to expect lower cloud tops in dirty conditions. We do see lower cloud tops in dirty environments in Africa, and both regions have lower 0-dBZ echo tops in dirty environments (Table 3).
The environmental variations between clean and dirty congestus again could explain some of the intensity differences. However, of the proxies for convective intensity listed in Table 3, only cloud width in both regions and maximum reflectivity in the Amazon have statistically different means for clean and dirty congestus.
4. Discussion and conclusions
The indirect effects of aerosols on convective clouds have been well studied using a variety of models, but observations of clouds in clean and dirty environments are critical to verifying these effects. However, separating the indirect effects of aerosols from that of the environment is not a task to be taken lightly. In this study, TRMM, CloudSat, and MODIS data have been utilized in an attempt to isolate aerosol indirect effects on convective clouds. MODIS AI is used to estimate background aerosol concentration. Because the MODIS sensor cannot detect aerosols within cloud, values from a rather large area are averaged to obtain AI. These values for AI are utilized as a best guess for background aerosol concentration, although much uncertainty exists within both the MODIS aerosol retrievals (including assumptions of spherically shaped aerosols, assumptions of aerosol composition, humidification effects, etc.) and with the method of matching MODIS AI to our RPFs and congestus (differences in overpass times, inability of satellite to resolve small scales, inability of satellite to measure aerosols within or close to the cloud).
Assuming that the MODIS aerosol estimates are reasonable, and despite difficulties in collocation, differences are observed in TRMM RPFs in clean and dirty environments. Our deeper-convection RPFs consistently have higher reflectivities, taller echo tops, and more lightning flashes in dirty regimes in the Amazon and Africa regions. These results are consistent with previous observational studies (Williams et al. 2002; Andreae et al. 2004; Altaratz et al. 2010; Koren et al. 2010; Yuan et al. 2011a; Heiblum et al. 2012). In the NAM region, RPFs in clean environments actually have more flashes, leading us to believe that differences in AI in this region are a symptom of different environments.
The Atlantic regions show the fewest differences between RPFs occurring in clean and dirty environments. This finding appears to contradict previous studies, which have observed aerosol-induced invigoration of raining convective clouds (Heiblum et al. 2012), increases in cloud-top height and cloud fraction with increasing AOD (Koren et al. 2005, 2010), increasing rainfall rates with increasing AOD (Koren et al. 2012), and increased cloud top, rain top, and ice water path with increased aerosol loading (Storer et al. 2013, manuscript submitted to J. Geophys. Res.). Yuan et al. (2012) suggest that oceanic convection would have more lightning if more aerosols were present. Why do we not see these differences in our Atlantic region? The answer could lie in our definition of “clean” and “dirty.” Dirty features over the ocean still have low values of AI and lower concentrations of aerosols than continental air masses. Perhaps the dirty category examined here still lacks enough aerosols to produce visible manifestations of the indirect effect on the scale observed. Still, there must be an additional explanation, as previous studies have examined a similar region over similar time periods.
The Atlantic region has some other important differences from the Amazon and Africa. While the values of AI are in the same ranges, mean values of the Angstrom exponent are much smaller in the Atlantic than in any other region. This indicates the presence of larger aerosol particles, possibly including giant cloud condensation nuclei (GCCN). GCCN activate easier than smaller CCN and can reduce the number of cloud droplets, even in polluted conditions (O’Dowd et al. 1999; Yuan et al. 2008). Additionally, it must be noted that our RPFs in the Atlantic have lower cloud tops and are likely a mixture of features with warm rain and deeper convection with ice processes. The suppression of warm rain by aerosols and the invigoration of deeper features could be canceling each other out.
Environment plays a more important role in determining convective intensity than aerosols and therefore must be considered when searching for the aerosol indirect effect (Khain 2009; Jones and Christopher 2010; Williams et al. 2002). When comparing clean and dirty thermodynamic soundings, each region shows statistically significant differences between the two profiles. In the Amazon and Africa regions, the dirty soundings are drier at midlevels. In the NAM region, the dirty sounding is much moister. Differences in the soundings for the Atlantic region are difficult to discern, but the lower levels in the dirty environment are warmer and moister. Because of these statistically significant environmental differences, we cannot attribute the changes in clean and dirty RPFs observed in this study to aerosol indirect effect alone.
Separating the RPFs into different environments based on CAPE and midlevel relative humidity shows that, while generalizations can be made about the effects of aerosols on convective clouds in individual regions, there are no patterns that can be applied to the entire globe, or even to specific environments. For example, in most environments in the Amazon and Africa regions, we see that dirty RPFs are more convectively intense than clean RPFs. We cannot say dirty RPFs are more intense than clean RPFs for low values of CAPE. Unlike previous modeling studies (Lee et al. 2008; Fan et al. 2009; Lee 2011), we are not able to find an environment in which differences between clean and dirty RPFs are maximized.
Congestus clouds are examined to determine the effects of aerosols on shorter clouds without a significant amount of ice. Most of the congestus in this study are, in reality, small groups of congestus because individual clouds are comparable in size to the CloudSat resolution. In the Amazon and Africa regions observed, clean congestus have higher reflectivities, as would be expected in a cleaner cloud with fewer CCN and larger cloud droplets and rain. Congestus width also changes—the maximum width of congestus groups in dirty environments is larger than that in clean environments. Congestus in dirty backgrounds also have higher mean cloud tops than those in clean environments, but the difference is not statistically significant. This is doubtless a result of the smaller sample size of CloudSat congestus, and further investigation is warranted. The effects of aerosols on these congestus, which fall somewhere in between warm rain and deeper convective clouds with considerable ice processes, remain unknown.
Previous studies tell us that the effects of aerosols on smaller clouds, such as the congestus features identified by CloudSat, and larger, deep clouds with ice processes, such as the RPFs observed using TRMM, are expected to be different. Aerosols are known to suppress warm rain in shallow clouds (Rosenfeld 1999; Khain et al. 2005; Berg et al. 2008; Khain et al. 2008; Rosenfeld et al. 2008; Koren et al. 2012). For deeper clouds that contain significant quantities of ice, an increase in convective intensity is expected (Rosenfeld 1999; Andreae et al. 2004; Khain and Pokrovsky 2004; van den Heever et al. 2006; Rosenfeld et al. 2007; van den Heever et al. 2011; Storer and van den Heever 2013). At what tipping point do convective clouds move from suppression by aerosols to invigoration by aerosols? This question was not answered by this study but is worth further investigation.
Aerosols must have some effects on convective clouds. The indirect effect has been observed in case studies (Rosenfeld 1999; Williams et al. 1999). Other satellite-based studies show invigoration by aerosols (Lin et al. 2006; Jiang et al. 2008; Koren et al. 2010; Heiblum et al. 2012; Storer et al. 2013, manuscript submitted to J. Geophys. Res.). Some of these studies do not include an in-depth look at the environment, which is critical to isolating the indirect effect. The signal of the aerosol indirect effect could be too small to be observed conclusively using a large satellite-based dataset and these methods, or it could be hiding in the assumptions of the aerosol algorithms, in the technique used to identify mean background aerosol index, or in the time lag between satellite overpasses. The atmosphere could be compensating for aerosols in ways that cannot be detected with a single satellite overpass of a cloud. The true magnitude of the aerosol indirect effect remains elusive on a global scale.
This study appears to point to the inadequacies of current satellite instruments and algorithms to observe aerosols at the level needed to convincingly determine aerosol indirect effects on convective clouds. At this point, further investigation by satellites would not be as useful as improving ground-based measurements of aerosols in the vicinity of a radar with the ability to obtain full life cycles of convective clouds or implementing a larger-scale field study involving the collocation of ground-based radars with aircraft equipped with aerosol sensors. The satellites used in this study only view a snapshot of these clouds, making causality impossible to ascertain. Modeling studies point to aerosol effects on cloud lifetime. Observations are critical to verifying these modeling studies, but current observational networks are inadequate to answer these important questions. The initial suppression of rainfall may be countered by heavier rainfall later in the life of a storm, so observing the full life cycle of these clouds is critical to determining the total aerosol indirect effect (Rosenfeld et al. 2008; Stevens and Feingold 2009; Koren et al. 2012). Collocation of aerosol sensors that can resolve aerosol size and composition with cloud and precipitation radars are necessary to make strides in this field. Additional measurements of cloud droplet size distribution and drop size distribution would also help verify the microphysical effects of aerosols on clouds.
CloudSat data were obtained from the CloudSat Data Processing Center (http://www.cloudsat.cira.colostate.edu/). This work was funded by NASA Grant NNX11AG31G under the direction of Dr. Ramesh Kakar and NASA Grant NNX08AK28G under the direction of Dr. Erich Stocker. Thanks also go to Drs. Erich Stocker and John Kwiatkowski and the rest of the Precipitation Processing System (PPS) team at NASA Goddard Space Flight Center, Greenbelt, Maryland, for data processing assistance.