1. Introduction
Deep convective clouds (DCCs) impact climate and atmospheric chemistry in important ways. They supply massive amounts of precipitation to tropical regions, are the dominant source of water vapor to the upper tropical troposphere (Worden et al. 2007), and generate radiative heating from high clouds, which affects large-scale circulation (e.g., Sherwood et al. 1994). They produce NOx through lightning and are efficient at transporting chemical compounds to the free troposphere, including ozone, volatile organic compounds, aerosols, and dimethyl sulfide (DMS), a gaseous marine aerosol precursor. Moreover, DCCs help form sulfate compounds, which are key contributors to the cloud condensation nuclei (CCN) budget (Ervens 2015).
Dust is a major aerosol source in the tropical North Atlantic region, particularly below 5 km (Tsamalis et al. 2013). There is ample evidence that tropical deep convective cloud properties are affected by dust aerosols. Convection can ingest large amounts of dust (Corr et al. 2016; Seigel and van den Heever 2012; Twohy et al. 2009, 2017; Twohy 2015), and a variety of previous modeling and remote sensing studies have associated this dust with changes in DCC properties (Dong et al. 2018; Dunion and Velden 2004; van den Heever et al. 2006; Jenkins et al. 2008; Jiang et al. 2018; Koren et al. 2005; Li et al. 2017; Li and Min 2010; Lonitz et al. 2015; Min et al. 2014; Niu and Li 2012; Rosenfeld et al. 2001, 2012; Storer et al. 2014; Zhang et al. 2007, 2009), and with cloud properties in general (e.g., Kishcha et al. 2015). For instance, models indicate that CCN and ice nucleating particle (INP) availability impacts DCC development (Fan et al. 2016; van den Heever et al. 2006), and dust can be a source of both (Bègue et al. 2015; Kanji et al. 2017; Levin et al. 1996; Nenes et al. 2014).
However, observed DCC responses to dust aerosols vary widely, even in sign. DCCs have been associated with both more and less precipitation, or no change in precipitation (Dong et al. 2018; van den Heever et al. 2006; Li and Min 2010; Lonitz et al. 2015; Niu and Li 2012; Rosenfeld et al. 2001). They have also been associated with more, less, and variably intense tropical cyclones (Dunion and Velden 2004; Rosenfeld et al. 2012; Zhang et al. 2007, 2009) and with both DCC invigoration and suppression (Dong et al. 2018; Jiang et al. 2018; Koren et al. 2005; Li et al. 2017; Min et al. 2014; Storer et al. 2014).
One reason for these apparent discrepancies is that aerosol effects on DCCs appear to depend not only on aerosol type but also on spatial/temporal study scale and meteorological conditions (Fan et al. 2016). Another issue is that each study method has its own limitations. Although models provide information on the potential of dust aerosols to affect DCCs, they are constrained by a lack of information about the fraction of cloud-active aerosols and aerosol-mediated freezing processes, and some studies are limited in spatial scale. Aircraft in situ information is not available at large scales, and processes observed at small scales are often not representative of systemwide effects. Remote sensing data can suffer from difficulties with aerosol and cloud collocation, characterization of aerosols near clouds, and the detection of clouds and aerosols below clouds and near the surface.
A variety of confounding factors also make it difficult to isolate dust aerosol microphysical effects from other factors in the system. 1) Meteorology tightly covaries with dust aerosols over the tropics (e.g., Dunion and Velden 2004; Lonitz et al. 2015) and may play a larger role in DCC development than the aerosols themselves (Chakraborty et al. 2016). For example, the intertropical convergence zone (ITCZ) is a major, seasonally variable source of DCCs that also affects regional near-surface winds and Saharan dust transport (Doherty et al. 2012; Carlson and Prospero 1972). 2) Dust direct aerosol-radiation effects and associated aerosol radiative effects on the surface energy budget, the atmospheric profile and cloudiness (i.e., semidirect effects) co-occur (Boucher et al. 2013) and could mask microphysical effects. 3) Marine and combustion aerosols may also occur simultaneously and affect clouds (Jiang et al. 2018). 4) Dust can actively modify the numbers and effectiveness of cloud-active particles of other types, and vice versa (Bègue et al. 2015; Cziczo et al. 2009; Lee et al. 2009; Manktelow et al. 2010). For example, although dust can be a local CCN source, models suggest that over large scales it may actually diminish CCN reservoirs, serving as a condensation/coagulation sink for sulfuric acid vapor and ultrafine particles (Lee et al. 2009; Manktelow et al. 2010). These combined issues highlight why careful assessment of the uncertainties from a variety of platforms is needed to better understand the net effects of dust aerosol–cloud interactions over large spatial scales.
Much effort has been made to characterize aerosol–cloud interaction mechanisms from the bottom up, with the aim of improving model assessments of the net effects, but large uncertainties remain. In the current work, we take a complementary approach, combining cloud, aerosol, and meteorological profiles from satellite and reanalysis products to provide a top-down constraint of net dust aerosol impacts on DCCs over large spatial and temporal scales, directly from measurements. Our goal is to better understand the overall extent to which dust aerosols impact DCC prevalence over the entire tropical North Atlantic region. We use a statistically based method to reduce the above uncertainties and take advantage of the vast amounts of data provided by the frequent, large-scale satellite coverage.
2. Methods
This study takes place over oceanic regions between 0° and 30°N and 81° and 15.5°W (Fig. 1). Satellite data and reanalysis products from 2007 to 2010 are analyzed in 1-km altitude bin increments. Observations are obtained between 1 and 15 km above sea level (ASL), above and below which high-quality data were not fully available from some of the datasets described below. Thus, only the DCC portions between 1 and 15 km ASL are included in the analysis. Unless stated otherwise, the analysis focuses on differences in DCC prevalence aggregated across the entire geographic region but stratified at different altitude levels and in different meteorological bins.
The study region and overlying CloudSat, CALIPSO, and AIRS tracks.
Citation: Journal of Climate 33, 21; 10.1175/JCLI-D-20-0083.1
We define DCC prevalence as the fraction of CloudSat data within a specified group of 1-km profiles belonging to nonanvil portions of a DCC. Unless stated otherwise, DCCs were defined as having a continuous cloud depth of at least 4 km, to ensure cloud tops are tall enough to reach the subzero temperatures required for droplet freezing. They were also required to have a cloud base below 4 km ASL. This is a critical requirement; based on a preliminary analysis, the study region contains many high clouds that are >4 km thick, but do not necessarily reach the near-surface planetary boundary layer (e.g., see Fig. S1 in the online supplemental material). Such clouds are likely to be associated with clean aerosol conditions due to their high altitude, but are not necessarily deep convective clouds, being thinner on average. Thus, failure to exclude them could lead to false evidence for an invigoration effect.
Our goal is to elucidate the DCC microphysical responses to dust. Thus, unless stated otherwise, the analysis focuses on nighttime samples to reduce confounding effects from dust direct aerosol–radiation interactions and associated adjustments to the surface energy budget. Any remaining impact from such confounding effects is further substantially reduced by the meteorological stratification procedure detailed in section 2d.
a. Data sources
1) CloudSat and CALIPSO data
Cloud presence, base and top heights, and precipitation occurrence were obtained from CloudSat Cloud Profiling Radar CLDCLASS-lidar R04 product. These observations are collected twice daily, at 0130 (nighttime) and 1330 (daytime) local time. Cloud geometric thickness was derived from the cloud-base and cloud-top heights. These data were then sorted into 1-km vertical resolution bins for later analysis. CloudSat low-quality-flagged data were excluded (i.e., data with nonzero values in the CloudSat product “Data_quality” field).
Dust and polluted dust aerosol vertical distribution were obtained from Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observation (CALIPSO), version 4.20, level 2, 5 km aerosol-layer products at 532 nm (Winker 2018). These data were obtained only in cloud-free conditions, as determined by both the CALIPSO and CloudSat datasets, and were used only for validation.
2) AIRS data
We used the satellite-derived 12-h Atmospheric Infrared Sounder (AIRS) level 3 gridded standard retrieval product obtained from AIRS infrared and Advanced Microwave Sounding Unit (AMSU) data (Susskind et al. 2014; Kahn et al. 2014; AIRS Science Team and Teixeira 2013). Based on ascending and descending orbits, temperature T profile data are available at multiple vertical pressure levels (1000, 925, 850, 700, 600, 500, 400, 300, 250, 200, 150, 100, 70, and 50 mb; 1 mb = 1 hPa). These data are most trustworthy in clear sky conditions and when there is no heavy precipitation, but they provide useful data under most conditions. We matched these data in time and space to the nearest CloudSat/CALIPSO observations, and most AIRS data are contemporaneous with the cloud data. These data were used to calculate the lower-tropospheric stability (LTS), defined as the difference in potential temperature between 700 and 1000 hPa.
3) MERRA-2 output
Tropospheric modeled dust concentrations were obtained from MERRA-2 3D, 3-hourly, instantaneous, model-level, assimilation, aerosol mixing ratio V5.12.4 (M2I3NVAER_5_12_4) output (GMAO 2015a), after excluding data over terrestrial regions based on the MERRA-2 land–water–ice classification. MERRA-2 obtains aerosol composition by first assimilating satellite aerosol optical depth (AOD) data into the Goddard Earth Observing System Model, version 5 (GEOS-5), from sources such as the Moderate Resolution Imaging Spectroradiometer (MODIS), and simulating its transport using model winds. It then uses the embedded Goddard Chemistry Aerosol Radiation and Transport (GOCART) atmospheric chemistry model (Chin et al. 2002; Colarco et al. 2010; Randles et al. 2017) to estimate concentrations of mineral dust in five particle size bins (0.1–1, 1–1.8, 1.8–3, 3–6, and 6–10 μm) (Colarco et al. 2010), among other chemical compounds, at 0.5° × 0.625° spatial resolution and at 42 standard pressure levels. We obtained modeled dust concentrations at the closest point in time and space to CloudSat observations, and then averaged them into 1-km vertical bins.
Most dust validation in MERRA-2 has been conducted for total dust concentrations at the surface, and for column AOD values. A case study in Barbados indicates that MERRA-2 captures daily surface dust fairly well (r = 0.84) (Buchard et al. 2017). Monthly average surface dust concentrations from MERRA-2 are also very similar (generally within a few micrograms m−3) to those measured at Cayenne, French Guiana (Barkley et al. 2019), and in Barbados and Miami, Florida (Randles et al. 2016). Daily AOD at various dust-dominated locations in the region seem to also be captured well (within a few percent) (Chen et al. 2018), along with seasonal and interannual AOD variability over northwest Africa, the Sahel, and the Caribbean (Buchard et al. 2017; Randles et al. 2017).
MERRA-2 distinguishes between dust sizes, but there is little dust size-resolved validation, although modeled dust values in the two smallest size classes are likely to be most accurate (Kok et al. 2017). In this work, we rely on MERRA-2 to correctly identify dust-free atmospheric profiles (and thus dust-free DCCs). In Fig. S2 we show that CALIPSO profiles offer some indication that this reliance is justified; ~97% of “clean” profiles based on MERRA-2 dust values of <51 ng m−3 were also expected to be dust-free, based on vertically resolved CALIPSO data.
We applied the same MERRA-2 product that we used for dust to estimate the concentrations of sea salt (SS) aerosols (here taken as the mass-based sum per unit volume of all five size classes represented in the model), DMS, and organic carbon (OC), black carbon (BC) (including hydrophobic + hydrophilic components), and sulfate (
MERRA-2 inst3_3 d_asm_Nv: 3D, 3-hourly, instantaneous, model-level, assimilation, assimilated meteorological fields V5.12.4 (M2I3NVASM) (Buchard et al. 2017) were also used to obtain sea level pressure (SLP) and temperature, vertical pressure velocity (ω), specific humidity (Qυ), and relative humidity (RH) at each altitude. Although vertical air mass motion (as proxied here by ω) is a key determinant of cloud properties, it is difficult to constrain (e.g., Kennedy et al. 2011; Nuijens et al. 2015). MERRA-2 ω values are not well validated over this region to our knowledge, but other models have large discrepancies in this variable over the tropical North Atlantic region, particularly during summer and fall (Medeiros and Nuijens 2016). For these reasons, other, better-constrained proxies for large-scale stability, such as SLP and LTS, were also compared (Table 1).
Meteorological parameters used to assess meteorological covariability with dust, the data source from which they were derived, and the ranges and resolutions used to reduce the effects of meteorological covariability. All meteorological parameters are significantly different in clean conditions relative to average conditions, based on a Wilcoxon rank sum test (p values ≤ 0.01).
M2I3NVASM fields were also used to derive a variety of relevant meteorological variables, including horizontal wind speed (derived at each altitude from the MERRA-2 east and northward wind components), wind shear (based on the eastward and northward wind components at 1.5 and 7.5 km ASL), and convective available potential energy (CAPE) (based on MERRA-2 pressure, specific humidity, and temperature values). CAPE values were derived using the R package “aiRthermo” (Sáenz et al. 2018). Those profiles for which CAPE values could not be accurately determined were given a low-quality flag and were excluded from the final analysis in cases where CAPE data were used.
b. Conceptual approach for identifying aerosol impacts on DCCs
The average impact of aerosols on regional DCC prevalence Caero cannot be measured directly. However, Caero can be estimated based on the difference between DCC prevalence in all conditions Call minus that in the clean subset of conditions Cclean (discussed further in section 2c), minus the effects on DCC prevalence from meteorological variables that covary with the aerosols Cmc:
In practice, Caero as estimated in Eq. (1) is approximately equal to the average DCC prevalence response of a system to the different aerosol types and concentrations within the meteorological conditions specific to that system.
Equation (1) is of limited use for predictive purposes, and results from one region may not apply to those in a different region. This is because aerosol impacts on DCC prevalence vary by meteorological and aerosol conditions, and these factors can change in potentially interactive or nonlinear ways over different regional and temporal scales. However, within a given region and time period, an observational method based on Eq. (1) can provide an important constraint on cloud models that has been generally lacking to date, as it enables us to quantify Caero without detailed upfront knowledge about the fraction of cloud-active aerosols and aerosol-mediated freezing processes.
To be meaningful, a method based on Eq. (1) must meet three main requirements: 1) large, representative sample sizes, which can be easily obtained from model output and satellite aerosol remote sensing data gathered over several years; 2) a way to either reduce Cmc to near zero or to otherwise reduce the uncertainties in Cmc so that they do not overwhelm the Caero signal; and 3) the ability to correctly identify clean aerosol conditions.
c. Identifying clean conditions
We identified the subset of conditions that were clean with respect to a given aerosol type. Clean conditions were determined separately for different aerosol types: dust aerosols, combustion aerosols (proxied by BC and OC), sulfate aerosols, and marine aerosols (proxied by sea salt and DMS). The effects of each aerosol type on DCC prevalence were assessed by comparing the DCC prevalence in all conditions with the DCC prevalence in the cleanest 7.5% of conditions [calculated separately for each constituent over the entire study region horizontal domain (0°–30°N and 81°–15.5°W) in the altitude range where constituent types were largest] (Table 2). For example, all conditions with BC ≤ 10 ng m−3 were considered clean with respect to BC (Table 2). This concentration represents the cleanest 7.5% of BC concentrations between 1 and 5 km ASL, above which BC transport over tropical oceans occurs much less frequently (Schwarz et al. 2013). Note that the clean subset for each atmospheric constituent was determined independently of other atmospheric constituent concentrations, so each variable can be constrained separately in subsequent statistical analysis. In certain meteorological conditions, aerosol levels may be predominantly above or below the clean aerosol cutoff levels. We address this issue in section 2d(2). We address potential confounding effects on DCC prevalence from other constituents in section 2d(3).
The atmospheric constituents used to proxy different aerosol types in the study and their associated clean cutoffs (below which conditions are considered clean with respect to that constituent and above which they are not).
A DCC was considered “clean” only if the entire vertical extent of the cloud was free of appreciable levels of the aerosol type of focus. Clouds that were clean with respect to dust, for example, were required to have dust concentrations < 51 ng m−3 for the smallest dust size bin over the entire cloud vertical extent. This criterion allows for taller clouds to be analyzed taking into account any dust microphysical processes operating at lower altitudes. It also avoids biases related to cloud height. For example, if an in-cloud dust average were used instead of a cloud maximum, then all else being equal, clouds with greater vertical extent could be presumed clean more often than shorter clouds, due to their greater exposure to cleaner air aloft. Our conservative requirement that the entire cloud be clean is also the best way we can at present address the fact that in-cloud aerosols can be vertically and horizontally redistributed by entrainment, convection, and precipitation (e.g., Stevens et al. 2016; Yu et al. 2019), to an extent that is difficult to constrain with current satellite or model capabilities. Future studies on this topic will benefit from scientific efforts to improve the understanding of aerosol redistribution within clouds (e.g., Sauter et al. 2019).
d. Using meteorological stratification to separate aerosol and covarying meteorology effects
It is common practice to group atmospheric data by meteorological constraints in order to elucidate cloud responses to external variables (e.g., Barton et al. 2012; Oreopoulos et al. 2016; Taylor et al. 2015). In dust-laden air from the Saharan air layer (SAL), this is particularly important, given strong covarying meteorological effects that could be mis-attributed to aerosols. Others have used the technique of stratifying data by covarying meteorological variables prior to analysis, to isolate aerosol effects from covarying meteorological effects (e.g., Chen et al. 2014; Douglas and L’Ecuyer 2018; Zamora et al. 2018). Our strategy to mitigate the effects of Cmc builds upon these methods, following the three steps outlined in Fig. 2.
Steps taken to assess whether the changes in DCCs are attributable to aerosols (see section 2d for further details).
Citation: Journal of Climate 33, 21; 10.1175/JCLI-D-20-0083.1
1) Step 1: Data stratification
Confounding effects from covarying meteorology are reduced by comparing DCC prevalence within bins defined by 1-km altitude stratification combined with meteorological binning. Meteorological bins are based on three separate meteorological variables (step 1, Fig. 2): RH, ω, and shear. The bin sizes for RH, ω, and shear are listed in Table 1, and Fig. S3 shows relationships between bin sizes and data distributions. By simultaneously stratifying according to altitude and RH, shear, and ω, the dataset is divided into 126 000 separate bins prior to analysis (14 altitude bins × 10 RH bins × 10 shear bins × 90 ω bins), some of which are empty, even as data covering the entire geographical study domain are included. As we only compare 1-km sections of the profile data with other data in the same 1-km range, any meteorological height dependencies are circumvented, and DCC prevalence in each altitude bin is independent of the depth of the individual DCCs outside of that altitude layer.
RH, ω, and shear were used in this step because they were objectively determined to have the highest potential for confounding effects with aerosols in our dataset (see the supplemental material for details). These variables are also meteorologically distinct; one relates to water vapor, one to large-scale vertical air motion, and one to vertical differences in wind directions in the column. The sensitivity of the results to various uncertainties, including to this choice of meteorological stratification variables, is tested in step 2 below. Bin sizes for these variables are chosen to strike a balance between including as many data as possible by using larger bins (and thus having more representative results that are less prone to autocorrelation) and removing as much of the covarying meteorological effect as possible by using smaller bin sizes. We test whether the chosen bin sizes are small enough to adequately reduce meteorological covariability at a later step [see step 3 (Fig. 2), section 2d(3)].
2) Step 2: Determine if aerosols are robustly associated with changes in DCC prevalence
Our next goal is to determine if aerosols are associated with meaningful differences in DCC prevalence within individual bins defined by altitude and meteorological variables, assessed across the entire study region (step 2 in Fig. 2). We first identify the change in DCC prevalence between all minus clean conditions in each bin. In this step, aerosol concentrations are only used to determine whether an observation has aerosol concentrations below the clean aerosol cutoff. A DCC is classified as “clean” when the maximum aerosol level between the base and top of the convective cloud is below the aerosol clean cutoff value. This maximum aerosol level in the DCC may be different than the initial MERRA-2 aerosol value at a particular z level.
To assess whether there is a net overall effect of dust aerosol on DCC prevalence after accounting for meteorological covariability, we need to consider the fact that aerosols are present more commonly in some altitude and meteorological bins than in others. Thus, within each altitude bin z and meteorological bin i, we take the difference in DCC prevalence under all aerosol conditions
Similarly, for each meteorological and altitude bin, we obtain dXzi values (the average differences in the all minus clean values of a given meteorological variable X). Table 1 lists the nine X variables included, and their corresponding ranges. Then, because some meteorological bins are more commonly observed than others, we weight each
Thus, in each elevation bin,
We reduce random error in this step by excluding meteorological bins with low numbers of observations (i.e., those containing <40 clean or <40 nonclean cases, assessed separately for each aerosol type at each 1-km altitude layer). For dust, for example, this removes up to 8% of the data, depending on altitude; the amount removed for other aerosol types is similarly low, except for sea salt, which loses up to 32% of data above 6 km, and sulfate, which loses up to 14% of the data between 1 and 4 km.
Next, we assess how robust the results are with respect to the various uncertainties in the analysis. For example, we have assumed that 1) the dust size range of the smallest dust size bin is best for predicting possible dust impacts on clouds, 2) the minimum dust concentration at which dust can impact clouds is 51 ng m−3 (Table 2), and 3) a DCC height is best defined as >4-km continuous vertical extent. It is important to determine how sensitive the results are to these assumptions, and also to evaluate how robust the results are to the choice of meteorological variables used to reduce the confounding effects of meteorological covariation from step 1. We answer these questions using a random Latin hypercube sampling method (Stein 1987).
The Latin hypercube method works as follows. To test the assumptions mentioned above,
Based on those random selections, we calculate a profile of
The sensitivity of DCC prevalence to other aerosol types is determined in a similar fashion. As before, a Latin hypercube selects one random combination of three meteorological variables by which to subset, and a minimum thickness for which to define a DCC, from values ranging between 4 and 8 km. The
3) Step 3: Signal attribution within meteorologically stratified data
Having established that dust is associated with differences in DCC prevalence, we now aim to determine to which factors this signal can be attributed. So the last step in assessing the dust aerosol impact on DCC prevalence over the study region (step 3 in Fig. 2) is to determine whether the nonzero
(a) Total DCC prevalence observed during night (solid line) and day (dashed line) over the study region shown in Fig. 1. (b) Potential aerosol impacts on DCC prevalence from aerosols of different types, as estimated by
Citation: Journal of Climate 33, 21; 10.1175/JCLI-D-20-0083.1
To answer this question, we identify the factors most predictive of DCC prevalence variability within the RH, ω, and wind shear meteorological bins. In addition to stratifying the data into meteorological bins and altitude in step 1, we now further sort the observations into geographic locations with 5° × 5° resolution, because we need to evaluate DCC prevalence over a range of different dust and other aerosol concentrations within the same meteorological and altitude stratification. The additional stratification provides many different bins within which to assess statistically the relationships between DCC prevalence and different mean values of individual variables under all-minus-clean conditions.
We focus on data in the altitude range with the largest potential aerosol impacts on DCC prevalence after binning by RH, ω, and shear (Fig. 3); this turns out to be between 3 and 8 km (see section 3a). As in step 2, we analyze bins containing at least 40 clean and 40 nonclean observations, to reduce random error. In this third step, we additionally required that the clean and nonclean data were acquired on at least 5 separate days during the study period to reduce autocorrelation. This is needed because some very large DCCs contribute many profiles to the same meteorological bin, which could lead to bias from nonindependent data within meteorological subsets. Last, we only assess data in meteorological bins with >5 locations having nonzero
Then, for each meteorological bin i, and for each location j, at 1 km × 5° × 5° resolution, we calculate
We then use a generalized additive model (GAM; Hastie and Tibshirani 1990) to test whether aerosols are more predictive of the remaining variability (deviance) than the meteorological variables. This allows us to ensure that none of the meteorological variables are significantly predictive of the remaining variability, thus indicating that the meteorological bins are sufficiently small that Cmc [Eq. (1)] has become negligible within them. The GAM also makes it possible to determine the degree to which different aerosol types drive the magnitude of Call − Cclean. By visual inspection of
3. Results
a. Aerosol impacts on the likelihood of DCC prevalence
DCCs occur relatively infrequently over the tropical North Atlantic, being observed only ~4%–5% of the time, based on our definition of at least a 4-km continuous cloud vertical extent starting at below 4 km ASL (Fig. 3a). This estimate avoids capturing the anvil portions of the clouds, or those portions of clouds below 1 km or higher than 15 km.
Based on uniformly positive
Marine aerosols (proxied by DMS and sea salt), and OC are also associated with robust regionwide increases in DCC prevalence at all altitude ranges (Fig. 3b and Fig. S4). Increases are robust for BC and sulfate aerosols above 4 km. Sea salt is associated with the smallest changes in DCC prevalence, with a best estimate
b. The role of specific uncertainties on DCC prevalence in the presence of dust
Before hypothesizing why DCCs were more prevalent in the presence of dust and other aerosols (Fig. 3b), we first investigate the role of uncertainties in the calculation of DCC prevalence. Figure 4 examines the influence of specific uncertainties on the estimated dust impacts on DCC prevalence. As shown in Fig. 4a, the results were not especially sensitive to the meteorological properties used to control for meteorological covariability. This makes sense given that the SAL affects a wide range of meteorological parameters at the same time, as shown by the significant differences in all the available meteorological parameters listed in Table 1 in the presence of dust.
The median of the subset of nighttime
Citation: Journal of Climate 33, 21; 10.1175/JCLI-D-20-0083.1
The DCC height definition also did not have a major influence on the results (Fig. 4b), although DCCs with smaller vertical extents are associated with somewhat more prevalent DCCs. Similarly, the dust size range has only minor impact on the results (Fig. 4c). Smaller size ranges are associated with slightly larger signals, which may be due to models being better able to accurately simulate smaller-sized dust particles than larger size fractions (Kok et al. 2017). The factor with the largest impact on the results is the definition of clean (dust free) conditions (i.e., the minimum dust level at which we assume dust begins to affect regional DCC prevalence). We find that even very low concentrations of dust are associated with robust differences in DCC. As such, higher assumed clean cutoff levels include some of the air masses shown to be impacted by dust as “clean,” resulting in lower
We also assessed how precipitation-related artifacts might affect the Fig. 3
Last, the effect of aerosol redistribution within a DCC is not well constrained by current observations and modeling. This uncertainty might influence our results to an extent that is difficult to quantify. Consistent with the convective nature of DCCs, we assume that if present, aerosols can affect a DCC throughout its vertical column, even if aerosols are not observed externally to the cloud at all vertical levels.
c. Signal attribution
Previously, we observed that dust aerosols are associated with robust reductions in DCC prevalence, even after accounting for confounding meteorological factors. Now in step 3 (Fig. 2), we use a GAM to better understand whether this association is likely to be driven by dust or some other factor. Following section 2d(3), we calculate DCC prevalence differences in all aerosol conditions minus those within the clean (dust free) subset within each meteorological bin i and each 1 km × 5° × 5° resolution bin j (
Figure 5 shows results from this analysis. Only about 17% of the data meet the fairly strict qualifications for this step [see section 2d(3)], although the general trends are still observed with more relaxed data quality requirements (cf. Fig. 5 and Fig. S5). Meteorological factors account for almost none of the remaining variability after the height and meteorological binning step, whereas aerosol proxies explain a substantial amount of the deviance. Thus, we conclude that the meteorological bin sizes are small enough to account for covarying meteorology in step 3 of the analysis (Fig. 5), and that covariation of aerosols with meteorology (e.g., the ITCZ) is unlikely to explain the positive
The factors that best predict
Citation: Journal of Climate 33, 21; 10.1175/JCLI-D-20-0083.1
DCCs are always more prevalent at elevated marine aerosol levels [as indicated by increasingly positive
4. Discussion and conclusions
DCCs loft DMS and other marine gases and small particles into the free troposphere, where they can form new particles that eventually contribute a considerable portion of CCN over the region (Clarke et al. 2013; Sanchez et al. 2018; Williamson et al. 2019). However, in the presence of dust, DMS and other marine-aerosol CCN precursors can also be adsorbed onto dust particles (Howell et al. 2006; Lee et al. 2009; Manktelow et al. 2010). Precipitating clouds can then quickly remove this dust from the system (Prospero et al. 1987; Tulet et al. 2010). Large-scale model studies have predicted that dust adsorption of marine CCN precursors could reduce net CCN levels over large scales (Lee et al. 2009; Manktelow et al. 2010). However, to date this effect has been challenging to verify, and especially to quantify. Models suffer from large uncertainties because they require upfront knowledge about the fraction of cloud-active aerosols and aerosol-mediated freezing processes. These are still poorly understood, and field studies cannot easily quantify the influence of dust absorption on cloud properties due to the large range of temporal and spatial scales involved.
For the first time, we are able to support the occurrence of these effects with remote sensing–based data and offer a preliminary quantification of their effects on DCC prevalence over large regional scales after accounting for meteorological covariation with aerosols and various other uncertainties. Based on four years of nighttime CloudSat/CALIPSO cloud data and MERRA-2 aerosol reanalysis products, we find that dust is associated with a 54% increase in DCC prevalence over the tropical North Atlantic, whereas even more pronounced increases occur in the presence of DMS (a proxy for marine aerosols), and combustion and sulfate aerosols (61%–64%). Given the meteorological stratification procedure that underlies these results, and the similar changes in DCC prevalence during day and night (Fig. 3c), the observations are unlikely to be driven by covarying meteorology or direct aerosol-radiation effects and associated adjustments to the surface energy budget.
Dust, marine, sulfate, and combustion aerosols all show positive relationships with DCC prevalence in this region, but to varying degrees. Model studies suggest that CCN levels can influence factors important to DCC fraction and lifetime, such as cloud phase, precipitation, and the height of freezing and riming processes (Jouan and Milbrandt 2019; Tao and Li 2016; Grabowski and Morrison 2016; Heikenfeld et al. 2019; Storer et al. 2010; van den Heever et al. 2006; Fan et al. 2018; Ilotoviz and Khain 2016). If (as expected) marine aerosols are substantially more effective CCN than dust aerosols, that might explain (i) why DCCs are even more prevalent at elevated DMS levels than at elevated levels of dust and other aerosols; (ii) why DMS and sea salt are 5 times more predictive of DCC prevalence in the presence of dust than all other tested meteorological and aerosol parameters within similar locations and meteorological bins, including the dust concentration itself; and (iii) why marine aerosols are also the most predictive factors for the increases in DCC prevalence associated with marine, combustion and sulfate aerosols. If dust, in addition to only being a weak CCN source, also adsorbs marine CCN precursors, that could contribute to the smaller changes in DCC prevalence in the presence of dust than other aerosols. It would also explain why DCCs are less prevalent at high dust levels compared to low dust levels after controlling for meteorological factors like wind speed, when the opposite is true for marine aerosols. These hypotheses are in line with a previous finding near the North Atlantic tropical region that dust seems to affect precipitation less strongly over land than over ocean, where marine CCN precursors are more abundant (Huang et al. 2009).
We cannot rule out that an INP mechanism (e.g., Twohy et al. 2017) might help explain why dust has smaller impacts on DCC prevalence than most other aerosol types. However, dust aerosols are generally thought to be more ice nucleating active than marine aerosol species (Kanji et al. 2017), and marine aerosols are thought to contribute more to the CCN budget over the study region than dust (Karydis et al. 2011; Quinn et al. 2017). Thus, an INP mechanism would not explain the tighter relationship between DCC prevalence and proxies for marine aerosols than with dust within similar location and meteorological bins. Furthermore, recent work suggests that DCCs in the region are not INP limited but are instead sustained by secondary ice processes (Ladino et al. 2017; Korolev et al. 2019).
Whichever mechanism is at work seems to occur even at relatively low dust concentrations. Thus, better knowledge of low-level dust concentrations will help improve estimates of dust impacts on clouds, as will better understanding of dust precipitation and redistribution within DCCs, the relationships between aerosol-mediated DCC prevalence and large-scale meteorological phenomena, and better data and model validation for atmospheric compounds contributing to the CCN budget in the tropical North Atlantic.
This work is one of several early applications of binning methods for reducing meteorological covariability to estimate aerosol microphysical effects. As such, we cannot rule out the influence of unforeseen factors on the results. We have demonstrated in Fig. 5 why it is unlikely that covariation by the most likely meteorological factors is driving the signal attributed here to aerosol microphysical effects. However, errors in the reanalysis products (e.g., in-cloud dust redistribution) could affect the derived influence of meteorological factors relative to aerosols shown in Fig. 5, and we have no way to account for meteorological “unknown unknowns.”
The methods developed in this study offer one way to quantify the total impacts of aerosol–cloud interactions on cloud properties, and elucidate relevant processes, even when aerosols covary strongly with meteorology. Based on these methods, we can show that aerosols have a major influence on deep convective cloud prevalence over large spatial and temporal scales. This has implications for the fields of climate, precipitation, and atmospheric chemistry. If more widely applied, we believe that future studies adopting the methods developed here can help advance our current understanding of aerosol–cloud interactions, which currently remain among the major uncertainties in climate science.
Acknowledgments
We used R package “aiRthermo” for determining CAPE. We acknowledge the use of imagery from the NASA Worldview application (https://worldview.earthdata.nasa.gov/), part of the NASA Earth Observing System Data and Information System (EOSDIS). We thank P. Colarco, J. Gong, K. Huebert, R. Kramer, D. Lee, R. Levy, A. Loftus, L. Oreopoulos, I. Tan, C. Wang, Y. Yang, and T. Yuan for helpful discussions. The authors were supported in part by the NASA Aerosol-Cloud Modeling and Analysis Program (Grants NNH16ZDA001N-ACMAP16-0045 and NNH18ZDA001N-ACMAP18-0010) under Richard Eckman, and by the Climate and Radiation program under Hal Maring.
Data availability statement
CloudSat data were obtained from the CloudSat Data Processing Center of the Cooperative Institute of Research in the Atmosphere at http://www.cloudsat.cira.colostate.edu. See AIRS Science Team and Teixeira (2013) for the AIRS data, GMAO (2015a,b) for the MERRA-2 output, and Winker (2018) for the CALIPSO data.
REFERENCES
AIRS Science Team, and J. Teixeira, 2013: AIRS/Aqua L3 Daily Standard Physical Retrieval (AIRS+AMSU) 1 degree × 1 degree V006. Goddard Earth Sciences Data and Information Services Center, accessed 11 October 2018, https://doi.org/10.5067/Aqua/AIRS/DATA301.
Barkley, A. E., and Coauthors, 2019: African biomass burning is a substantial source of phosphorus deposition to the Amazon, tropical Atlantic Ocean, and Southern Ocean. Proc. Natl. Acad. Sci. USA, 116, 16 216–16 221, https://doi.org/10.1073/pnas.1906091116.
Barton, N. P., S. A. Klein, J. S. Boyle, and Y. Y. Zhang, 2012: Arctic synoptic regimes: Comparing domain-wide Arctic cloud observations with CAM4 and CAM5 during similar dynamics. J. Geophys. Res., 117, D15205, https://doi.org/10.1029/2012JD017589.
Bègue, N., P. Tulet, J. Pelon, B. Aouizerats, A. Berger, and A. Schwarzenboeck, 2015: Aerosol processing and CCN formation of an intense Saharan dust plume during the EUCAARI 2008 campaign. Atmos. Chem. Phys., 15, 3497–3516, https://doi.org/10.5194/acp-15-3497-2015.
Bian, H., and Coauthors, 2019: Observationally constrained analysis of sea salt aerosol in the marine atmosphere. Atmos. Chem. Phys., 19, 10 773–10 785, https://doi.org/10.5194/acp-19-10773-2019.
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.
Buchard, V., and Coauthors, 2017: The MERRA-2 aerosol reanalysis, 1980 onward. Part II: Evaluation and case studies. J. Climate, 30, 6851–6872, https://doi.org/10.1175/JCLI-D-16-0613.1.
Carlson, T. N., and J. Prospero, 1972: The large-scale movement of Saharan air outbreaks over the northern equatorial Atlantic. J. Appl. Meteor., 11, 283–297, https://doi.org/10.1175/1520-0450(1972)011<0283:TLSMOS>2.0.CO;2.
Chakraborty, S., R. Fu, S. T. Massie, and G. Stephens, 2016: Relative influence of meteorological conditions and aerosols on the lifetime of mesoscale convective systems. Proc. Natl. Acad. Sci. USA, 113, 7426–7431, https://doi.org/10.1073/pnas.1601935113.
Chen, S.-P., C.-H. Lu, J. McQueen, and P. Lee, 2018: Application of satellite observations in conjunction with aerosol reanalysis to characterize long-range transport of African and Asian dust on air quality in the contiguous U.S. Atmos. Environ., 187, 174–195, https://doi.org/10.1016/j.atmosenv.2018.05.038.
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, https://doi.org/10.1038/ngeo2214.
Chin, M., and Coauthors, 2002: Tropospheric aerosol optical thickness from the GOCART model and comparisons with satellite and Sun photometer measurements. J. Atmos. Sci., 59, 461–483, https://doi.org/10.1175/1520-0469(2002)059<0461:TAOTFT>2.0.CO;2.
Clarke, A. D., and Coauthors, 2013: Free troposphere as a major source of CCN for the equatorial Pacific boundary layer: Long-range transport and teleconnections. Atmos. Chem. Phys., 13, 7511–7529, https://doi.org/10.5194/acp-13-7511-2013.
Colarco, P., A. da Silva, M. Chin, and T. Diehl, 2010: Online simulations of global aerosol distributions in the NASA GEOS-4 model and comparisons to satellite and ground-based aerosol optical depth. J. Geophys. Res., 115, D14207, https://doi.org/10.1029/2009JD012820.
Corr, C. A., and Coauthors, 2016: Observational evidence for the convective transport of dust over the central United States. J. Geophys. Res. Atmos., 121, 1306–1319, https://doi.org/10.1002/2015JD023789.
Cziczo, D. J., K. D. Froyd, S. J. Gallavardin, O. Moehler, S. Benz, H. Saathoff, and D. M. Murphy, 2009: Deactivation of ice nuclei due to atmospherically relevant surface coatings. Environ. Res. Lett., 4, 044013, https://doi.org/10.1088/1748-9326/4/4/044013.
Doherty, O. M., N. Riemer, and S. Hameed, 2012: Control of Saharan mineral dust transport to Barbados in winter by the intertropical convergence zone over West Africa. J. Geophys. Res., 117, D19117, https://doi.org/10.1029/2012JD017767.
Dong, X., R. Li, Y. Wang, Y. Fu, and C. Zhao, 2018: Potential impacts of Sahara dust aerosol on rainfall vertical structure over the Atlantic Ocean as identified from EOF analysis. J. Geophys. Res. Atmos., 123, 8850–8868, https://doi.org/10.1029/2018JD028500.
Douglas, A., and T. L’Ecuyer, 2018: Understanding aerosol-cloud-radiation interactions using local meteorology and cloud state constraints. Atmos. Chem. Phys., 19, 6251–6268, https://doi.org/10.5194/acp-19-6251-2019.
Dunion, J. P., and C. S. Velden, 2004: The impact of the Saharan air layer on Atlantic tropical cyclone activity. Bull. Amer. Meteor. Soc., 85, 353–366, https://doi.org/10.1175/BAMS-85-3-353.
Ervens, B., 2015: Modeling the processing of aerosol and trace gases in clouds and fogs. Chem. Rev., 115, 4157–4198, https://doi.org/10.1021/cr5005887.
Fan, J., Y. Wang, D. Rosenfeld, and X. Liu, 2016: Review of aerosol–cloud interactions: Mechanisms, significance, and challenges. J. Atmos. Sci., 73, 4221–4252, https://doi.org/10.1175/JAS-D-16-0037.1.
Fan, J., and Coauthors, 2018: Substantial convection and precipitation enhancements by ultrafine aerosol particles. Science, 359, 411–418, https://doi.org/10.1126/science.aan8461.
GMAO, 2015a: Global Modeling and Assimilation Office (GMAO), MERRA-2 inst3_3d_aer_Nv: 3d, 3-Hourly, Instantaneous, Model-Level, Assimilation, Aerosol Mixing Ratio V5.12.4. Goddard Earth Sciences Data and Information Services Center, accessed 10 October 2018, https://doi.org/10.5067/LTVB4GPCOTK2.
GMAO, 2015b: Global Modeling and Assimilation Office (GMAO), MERRA-2 inst3_3d_asm_Nv: 3d, 3-Hourly, Instantaneous, Model-Level, Assimilation, Assimilated Meteorological Fields V5.12.4. Goddard Earth Sciences Data and Information Services Center, accessed 30 October 2018, https://doi.org/10.5067/WWQSXQ8IVFW8.
Grabowski, W. W., and H. Morrison, 2016: Untangling microphysical impacts on deep convection applying a novel modeling methodology. Part II: Double-moment microphysics. J. Atmos. Sci., 73, 3749–3770, https://doi.org/10.1175/JAS-D-15-0367.1.
Hastie, T. J., and R. J. Tibshirani, 1990: Generalized Additive Models. Chapman and Hall, 352 pp.
Heikenfeld, M., B. White, L. Labbouz, and P. Stier, 2019: Aerosol effects on deep convection: The propagation of aerosol perturbations through convective cloud microphysics. Atmos. Chem. Phys., 19, 2601–2627, https://doi.org/10.5194/acp-19-2601-2019.
Howell, S. G., A. D. Clarke, Y. Shinozuka, V. Kapustin, C. S. McNaughton, B. J. Huebert, S. J. Doherty, and T. L. Anderson, 2006: Influence of relative humidity upon pollution and dust during ACE-Asia: Size distributions and implications for optical properties. J. Geophys. Res., 111, D06205, https://doi.org/10.1029/2004JD005759.
Huang, J., C. Zhang, and J. M. Prospero, 2009: African aerosol and large-scale precipitation variability over West Africa. Environ. Res. Lett., 4, 015006, https://doi.org/10.1088/1748-9326/4/1/015006.
Ilotoviz, E., and A. Khain, 2016: Application of a new scheme of cloud base droplet nucleation in a spectral (bin) microphysics cloud model: Sensitivity to aerosol size distribution. Atmos. Chem. Phys., 16, 14 317–14 329, https://doi.org/10.5194/acp-16-14317-2016.
Jenkins, G. S., A. S. Pratt, and A. Heymsfield, 2008: Possible linkages between Saharan dust and tropical cyclone rain band invigoration in the eastern Atlantic during NAMMA-06. Geophys. Res. Lett., 35, L08815, https://doi.org/10.1029/2008GL034072.
Jiang, J. H., H. Su, L. Huang, Y. Wang, S. Massie, B. Zhao, A. Omar, and Z. Wang, 2018: Contrasting effects on deep convective clouds by different types of aerosols. Nat. Commun., 9, 3874, https://doi.org/10.1038/s41467-018-06280-4.
Jouan, C., and J. A. Milbrandt, 2019: The importance of the ice-phase microphysics parameterization for simulating the effects of changes to CCN concentrations in deep convection. J. Atmos. Sci., 76, 1727–1752, https://doi.org/10.1175/JAS-D-18-0168.1.
Kahn, B. H., and Coauthors, 2014: The Atmospheric Infrared Sounder version 6 cloud products. Atmos. Chem. Phys., 14, 399–426, https://doi.org/10.5194/acp-14-399-2014.
Kanji, Z. A., L. A. Ladino, H. Wex, Y. Boose, M. Burkert-Kohn, D. J. Cziczo, and M. Krämer, 2017: Overview of ice nucleating particles. Ice Formation and Evolution in Clouds and Precipitation: Measurement and Modeling Challenges, Meteor. Monogr., No. 58, Amer. Meteor. Soc., https://doi.org/10.1175/AMSMONOGRAPHS-D-16-0006.1.
Karydis, V. A., P. Kumar, D. Barahona, I. N. Sokolik, and A. Nenes, 2011: On the effect of dust particles on global cloud condensation nuclei and cloud droplet number. J. Geophys. Res., 116, D23204, https://doi.org/10.1029/2011JD016283.
Kennedy, A. D., X. Dong, B. Xi, S. Xie, Y. Zhang, and J. Chen, 2011: A comparison of MERRA and NARR reanalyses with the DOE ARM SGP data. J. Climate, 24, 4541–4557, https://doi.org/10.1175/2011JCLI3978.1.
Kishcha, P., A. da Silva, B. Starobinets, C. Long, O. Kalashnikova, and P. Alpert, 2015: Saharan dust as a causal factor of hemispheric asymmetry in aerosols and cloud cover over the tropical Atlantic Ocean. Int. J. Remote Sens., 36, 3423–3445, https://doi.org/10.1080/01431161.2015.1060646.
Kok, J. F., and Coauthors, 2017: Smaller desert dust cooling effect estimated from analysis of dust size and abundance. Nat. Geosci., 10, 274–278, https://doi.org/10.1038/ngeo2912.
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, https://doi.org/10.1029/2005GL023187.
Korolev, A., and Coauthors, 2019: A new look at the environmental conditions favorable to secondary ice production. Atmos. Chem. Phys., 20, 1391–1429, https://doi.org/10.5194/acp-20-1391-2020.
Ladino, L. A., A. Korolev, I. Heckman, M. Wolde, A. M. Fridlind, and A. S. Ackerman, 2017: On the role of ice-nucleating aerosol in the formation of ice particles in tropical mesoscale convective systems. Geophys. Res. Lett., 44, 1574–1582, https://doi.org/10.1002/2016GL072455.
Lee, Y. H., K. Chen, and P. J. Adams, 2009: Development of a global model of mineral dust aerosol microphysics. Atmos. Chem. Phys., 9, 2441–2458, https://doi.org/10.5194/acp-9-2441-2009.
Levin, Z., E. Ganor, and V. Gladstein, 1996: The effects of desert particles coated with sulfate on rain formation in the eastern Mediterranean. J. Appl. Meteor., 35, 1511–1523, https://doi.org/10.1175/1520-0450(1996)035<1511:TEODPC>2.0.CO;2.
Li, R., and Q.-L. Min, 2010: Impacts of mineral dust on the vertical structure of precipitation. J. Geophys. Res., 115, D09203, https://doi.org/10.1029/2009JD011925.
Li, R., X. Dong, J. Guo, Y. Fu, C. Zhao, Y. Wang, and Q. Min, 2017: The implications of dust ice nuclei effect on cloud top temperature in a complex mesoscale convective system. Sci. Rep., 7, 13826, https://doi.org/10.1038/s41598-017-12681-0.
Lonitz, K., B. Stevens, L. Nuijens, V. Sant, L. Hirsch, and A. Seifert, 2015: The signature of aerosols and meteorology in long-term cloud radar observations of trade wind cumuli. J. Atmos. Sci., 72, 4643–4659, https://doi.org/10.1175/JAS-D-14-0348.1.
Manktelow, P. T., K. S. Carslaw, G. W. Mann, and D. V. Spracklen, 2010: The impact of dust on sulfate aerosol, CN and CCN during an East Asian dust storm. Atmos. Chem. Phys., 10, 365–382, https://doi.org/10.5194/acp-10-365-2010.
Medeiros, B., and L. Nuijens, 2016: Clouds at Barbados are representative of clouds across the trade wind regions in observations and climate models. Proc. Natl. Acad. Sci. USA, 113, E3062–E3070, https://doi.org/10.1073/pnas.1521494113.
Min, Q.-L., R. Li, B. Lin, E. Joseph, V. Morris, Y. Hu, S. W. Li, and S. Wang, 2014: Impacts of mineral dust on ice clouds in tropical deep convection systems. Atmos. Res., 143, 64–72, https://doi.org/10.1016/j.atmosres.2014.01.026.
Nenes, A., B. Murray, and A. Bougiatioti, 2014: Mineral dust and its microphysical interactions with clouds. Mineral Dust: A Key Player in the Earth System, P. Knippertz and J.-B. W. Stuut, Eds., Springer, 287–325.
Niu, F., and Z. Li, 2012: Systematic variations of cloud top temperature and precipitation rate with aerosols over the global tropics. Atmos. Chem. Phys., 12, 8491–8498, https://doi.org/10.5194/acp-12-8491-2012.
Nuijens, L., B. Medeiros, I. Sandu, and M. Ahlgrimm, 2015: The behavior of trade-wind cloudiness in observations and models: The major cloud components and their variability. J. Adv. Model. Earth Syst., 7, 600–616, https://doi.org/10.1002/2014MS000390.
Oreopoulos, L., N. Cho, D. Lee, and S. Kato, 2016: Radiative effects of global MODIS cloud regimes. J. Geophys. Res. Atmos., 121, 2299–2317, https://doi.org/10.1002/2015JD024502.
Prospero, J. M., R. T. Nees, and M. Uematsu, 1987: Deposition rate of particulate and dissolved aluminum derived from Saharan dust in precipitation at Miami, Florida. J. Geophys. Res., 92, 14 723–14 731, https://doi.org/10.1029/JD092iD12p14723.
Quinn, P. K., D. J. Coffman, J. E. Johnson, L. M. Upchurch, and T. S. Bates, 2017: Small fraction of marine cloud condensation nuclei made up of sea spray aerosol. Nat. Geosci., 10, 674–679, https://doi.org/10.1038/ngeo3003.
Randles, C. A., and Coauthors, 2016: Aerosols in MERRA-2. Tech. Rep. Series on Global Modeling and Data Assimilation 45, NASA Global Modeling and Assimilation Office, 140 pp.
Randles, C. A., and Coauthors, 2017: The MERRA-2 aerosol reanalysis, 1980 onward. Part I: System description and data assimilation evaluation. J. Climate, 30, 6823–6850, https://doi.org/10.1175/JCLI-D-16-0609.1.
Rosenfeld, D., Y. Rudich, and R. Lahav, 2001: Desert dust suppressing precipitation: A possible desertification feedback loop. Proc. Natl. Acad. Sci. USA, 98, 5975–5980, https://doi.org/10.1073/pnas.101122798.
Rosenfeld, D., W. L. Woodley, A. Khain, W. R. Cotton, G. Carrió, I. Ginis, and J. H. Golden, 2012: Aerosol effects on microstructure and intensity of tropical cyclones. Bull. Amer. Meteor. Soc., 93, 987–1001, https://doi.org/10.1175/BAMS-D-11-00147.1.
Sáenz, J., S. J. González-Rojí, S. Carreno-Madinabeitia, and G. Ibarra-Berastegi, 2018: aiRthermo: Atmospheric Thermodynamics and Visualization, version 1.2.1. R package, https://cran.r-project.org/web/packages/aiRthermo/index.html.
Sanchez, K. J., and Coauthors, 2018: Substantial seasonal contribution of observed biogenic sulfate particles to cloud condensation nuclei. Sci. Rep., 8, 3235, https://doi.org/10.1038/s41598-018-21590-9.
Sauter, K., T. S. L’Ecuyer, S. C. van den Heever, C. Twohy, A. Heidinger, S. Wanzong, and N. Wood, 2019: The observed influence of tropical convection on the Saharan dust layer. J. Geophys. Res. Atmos., 124, 10 896–10 912, https://doi.org/10.1029/2019JD031365.
Schwarz, J. P., and Coauthors, 2013: Global-scale seasonally resolved black carbon vertical profiles over the Pacific. Geophys. Res. Lett., 40, 5542–5547, https://doi.org/10.1002/2013GL057775.
Seigel, R. B., and S. C. van den Heever, 2012: Dust lofting and ingestion by supercell storms. J. Atmos. Sci., 69, 1453–1473, https://doi.org/10.1175/JAS-D-11-0222.1.
Sherwood, S. C., V. Ramanathan, T. P. Barnett, M. K. Tyree, and E. Roeckner, 1994: Response of an atmospheric general circulation model to radiative forcing of tropical clouds. J. Geophys. Res., 99, 20 829–20 845, https://doi.org/10.1029/94JD01632.
Stein, M., 1987: Large sample properties of simulations using Latin hypercube sampling. Technometrics, 29, 143–151, https://doi.org/10.1080/00401706.1987.10488205.
Stevens, B., and Coauthors, 2016: The Barbados cloud observatory: Anchoring investigations of clouds and circulation on the edge of the ITCZ. Bull. Amer. Meteor. Soc., 97, 787–801, https://doi.org/10.1175/BAMS-D-14-00247.1.
Storer, R. L., S. C. van den Heever, and G. L. Stephens, 2010: Modeling aerosol impacts on convective storms in different environments. J. Atmos. Sci., 67, 3904–3915, https://doi.org/10.1175/2010JAS3363.1.
Storer, R. L., S. C. van den Heever, and T. S. L’Ecuyer, 2014: Observations of aerosol-induced convective invigoration in the tropical east Atlantic. J. Geophys. Res. Atmos., 119, 3963–3975, https://doi.org/10.1002/2013JD020272.
Susskind, J., J. M. Blaisdell, and L. Iredell, 2014: Improved methodology for surface and atmospheric soundings, error estimates, and quality control procedures: The Atmospheric Infrared Sounder Science Team version-6 retrieval algorithm. J. Appl. Remote Sens., 8, 084994, https://doi.org/10.1117/1.JRS.8.084994.
Tao, W.-K., and X. Li, 2016: The relationship between latent heating, vertical velocity, and precipitation processes: The impact of aerosols on precipitation in organized deep convective systems. J. Geophys. Res. Atmos., 121, 6299–6320, https://doi.org/10.1002/2015JD024267.
Taylor, P. C., S. Kato, K.-M. Xu, and M. Cai, 2015: Covariance between Arctic sea ice and clouds within atmospheric state regimes at the satellite footprint level. J. Geophys. Res. Atmos., 120, 12 656–12 678, https://doi.org/10.1002/2015JD023520.
Tsamalis, C., A. Chédin, J. Pelon, and V. Capelle, 2013: The seasonal vertical distribution of the Saharan air layer and its modulation by the wind. Atmos. Chem. Phys., 13, 11 235–11 257, https://doi.org/10.5194/acp-13-11235-2013.
Tulet, P., K. Crahan-Kaku, M. Leriche, B. Aouizerats, and S. Crumeyrolle, 2010: Mixing of dust aerosols into a mesoscale convective system: Generation, filtering and possible feedbacks on ice anvils. Atmos. Res., 96, 302–314, https://doi.org/10.1016/j.atmosres.2009.09.011.
Twohy, C. H., 2015: Measurements of Saharan dust in convective clouds over the tropical eastern Atlantic Ocean. J. Atmos. Sci., 72, 75–81, https://doi.org/10.1175/JAS-D-14-0133.1.
Twohy, C. H., and Coauthors, 2009: Saharan dust particles nucleate droplets in eastern Atlantic clouds. Geophys. Res. Lett., 36, L01807, https://doi.org/10.1029/2008GL035846.
Twohy, C. H., and Coauthors, 2017: Saharan dust, convective lofting, aerosol enhancement zones, and potential impacts on ice nucleation in the tropical upper troposphere. J. Geophys. Res. Atmos., 122, 8833–8851, https://doi.org/10.1002/2017JD026933.
van den Heever, S. C., G. G. Carrió, W. R. Cotton, P. J. DeMott, and A. J. Prenni, 2006: Impacts of nucleating aerosol on Florida storms. Part I: Mesoscale simulations. J. Atmos. Sci., 63, 1752–1775, https://doi.org/10.1175/JAS3713.1.
Varble, A., 2018: Erroneous attribution of deep convective invigoration to aerosol concentration. J. Atmos. Sci., 75, 1351–1368, https://doi.org/10.1175/JAS-D-17-0217.1.
Williamson, C. J., and Coauthors, 2019: A large source of cloud condensation nuclei from new particle formation in the tropics. Nature, 574, 399–403, https://doi.org/10.1038/s41586-019-1638-9.
Winker, D. M., 2018: CALIPSO Lidar Level 2 5 km Aerosol Layer Data V4-20. NASA Langley Research Center Atmospheric Science Data Center DAAC, accessed 15 November 2019, https://doi.org/10.5067/CALIOP/CALIPSO/LID_L2_05kmALay-Standard-V4-20.
Wood, S. N., 2011: Fast stable restricted maximum likelihood and marginal likelihood estimation of semiparametric generalized linear models. J. Roy. Stat. Soc., 73B, 3–36, https://doi.org/10.1111/j.1467-9868.2010.00749.x.
Wood, S. N., N. Pya, and B. Säfken, 2016: Smoothing parameter and model selection for general smooth models. J. Amer. Stat. Assoc., 111, 1548–1563, https://doi.org/10.1080/01621459.2016.1180986.
Worden, J., D. Noone, and K. Bowman, 2007: Importance of rain evaporation and continental convection in the tropical water cycle. Nature, 445, 528–532, https://doi.org/10.1038/nature05508.
Yu, P., and Coauthors, 2019: Efficient in-cloud removal of aerosols by deep convection. Geophys. Res. Lett., 46, 1061–1069, https://doi.org/10.1029/2018GL080544.
Zamora, L. M., R. A. Khan, K. B. Huebert, A. Stohl, and S. Eckhardt, 2018: A satellite-based estimate of combustion aerosol cloud microphysical effects over the Arctic Ocean. Atmos. Chem. Phys., 18, 14 949–14 964, https://doi.org/10.5194/acp-18-14949-2018.
Zhang, H., G. M. McFarquhar, S. M. Saleeby, and W. R. Cotton, 2007: Impacts of Saharan dust as CCN on the evolution of an idealized tropical cyclone. Geophys. Res. Lett., 34, L14812, https://doi.org/10.1029/2007GL029876.
Zhang, H., G. M. McFarquhar, W. R. Cotton, and Y. Deng, 2009: Direct and indirect impacts of Saharan dust acting as cloud condensation nuclei on tropical cyclone eyewall development. Geophys. Res. Lett., 36, L06802, https://doi.org/10.1029/2009GL037276.
