Satellite-observed relationships between aerosol and trade-wind cumulus cloud properties over the Indian Ocean

Authors


Abstract

[1] We examine the relationships between simultaneous observations of aerosol properties from the Multiangle Imaging SpectroRadiometer (MISR) and trade-wind cumulus macrophysical properties including cloud fraction (fc), cloud size and cloud top height (hct) distribution from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) over the Indian Ocean during the dry season (Nov–Apr) of 2006–2007. Mean fc increases from 0.029 to 0.089 with an increase in aerosol optical depth (τa) from 0.1 to 0.3, plateaus between 0.3 < τa < 0.45 and decreases by >30% at τa > 0.45. The hct displays insensitive response to increasing τa for τa < 0.4, but reduces at τa > 0.4. Cloud size distribution shifts significantly to large cloud size at τa > 0.45, possibly due to more efficient desiccation of small clouds under semi-direct effect. The observed changes in cloud properties in relation to increasing τa suggest a very complex interaction between trade-wind cumuli and absorbing aerosols that cannot be explained by potential remote sensing artifacts or synoptic meteorology. Further examination of the potential role of mesoscale meteorology on such relationships using multi-year data sets is warranted.

1. Introduction

[2] Quantifying the impact of aerosols on cloud properties remains a large source of uncertainty in estimating anthropogenic climate change [Intergovernmental Panel on Climate Change, 2007]. Aerosols can either enhance cloud albedo and lifetime through indirect effects [Heymsfield and McFarquhar, 2001] or reduce cloud lifetime through the semi-direct effect [McFarquhar and Wang, 2006]. Koren et al. [2008] have hypothesized a smooth transition between these two aerosol effects through dynamic and radiative feedback processes, supported by observations over the Amazon. However, the combined influence of these two competing effects of aerosols on cloud properties may be region specific depending upon the distribution of aerosol optical and microphysical properties, cloud types and meteorology. Due to this complex and uncertain interaction between aerosol and clouds, Stevens and Feingold [2009] called for efforts to intensify research aimed at specific cloud regimes to understand their responses to the changing environment.

[3] One such cloud regime is the shallow, warm cumulus clouds found ubiquitously over the tropical trade wind regions. These clouds are known to play an integral part in the maintenance of tropical circulations through their influence on energy and moisture budgets [e.g., Betts, 1997], yet their interplay with aerosols are poorly understood. Observations from the Indian Ocean Experiment (INDOEX) first addressed the issue of aerosol effects on trade-wind cumuli [Heymsfield and McFarquhar, 2001]. More recent efforts to understand the evolution of these small clouds have included the Rain in Cumulus over the Ocean (RICO) field campaign [Rauber et al., 2007]. In quantifying aerosol effects on these clouds based on relationships between satellite-retrieved aerosol and cloud properties, we must recognize that these relationships may be influenced by physical factors such as the coupling of aerosol and cloud properties to the meteorology and artificial factors caused by the varied sources of systematic errors within our remotely sensed data products [Loeb and Schuster, 2008].

[4] In this study, we take advantage of a unique dataset, namely the high-resolution (15 m) Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data that has been tasked over the Indian Ocean as part of the validation efforts for Multiangle Imaging SpectroRadiometer (MISR) data products [Zhao et al., 2009]. Both ASTER and MISR are on the EOS-Terra satellite, thus offering coincident data from both instruments. ASTER has also been tasked over the Caribbean Sea as part of RICO [Zhao and Di Girolamo, 2007]. However, we focus only on the ASTER data over the Indian Ocean, where higher variation of aerosol optical depth (τa) relative to the Caribbean Sea dataset allows for an examination of relationships between trade-wind cumuli properties and aerosol characteristics, and where aircraft measurements during INDOEX suggest a strong influence of aerosols on cumulus clouds [McFarquhar et al., 2004]. Because of the low cloud fractions (fc) of trade-wind cumuli, the cloud macrophysical properties derived from ASTER can be directly correlated with the simultaneous retrievals of aerosol properties from MISR, in which the clouds are embedded. The objective of this study is to examine and interpret these correlations in terms of the plausible aerosol effects on clouds, with careful consideration of meteorological and potential artificial remote sensing influences.

2. Data and Methodology

[5] A detailed description of the ASTER dataset collected over the Indian Ocean is given by Zhao et al. [2009]. In brief, ASTER archives data collected over 60 km × 60 km regions when tasked to do so. The ASTER instrument was tasked to collect 1235 scenes over part of the Indian Ocean (12°N–5°S, 68°–78°E, Figure 1) during Nov 2006–Apr 2007. Each ASTER scene was visually inspected and any scene containing any amount of visible cirrus and stratiform clouds, and/or impeded by sun glint or land was discarded from the analysis. A cloud mask for each of the remaining 277 scenes (12, 41, 84, 99, 24 and 17 scenes for the consecutive six months from Nov to Apr) has been developed using a single, scene-dependent and manually chosen threshold applied to the 0.67 μm channel at 15 m resolution [Zhao et al., 2009].

Figure 1.

Normalized frequency distributions of τa, fc and α derived from 3641 samples over the study region in the Indian Ocean (shown by the bold rectangle with the spatial distribution of τa from MISR on an outline map of the Indian Subcontinent in inset, where the ‘star’ and ‘circle’ show the locations of the radiosonde and AERONET station, respectively) during Nov 2006–Apr 2007. The last (first) bin of τa and fc (α) histogram represents the normalized frequency of τa and fc (α) >0.55 and >0.5 (<0.8). The latitudinal distribution of the samples (#) used in the statistics is also shown in inset (‘black’ represents no data).

[6] Version 22 of the MISR level-2 aerosol products, reported at 17.6 km × 17.6 km resolution [Kahn et al., 2009], were collected over the overlapped region for each ASTER scene. We focus only on τa at 558 nm and Angstrom Exponent (α) that is derived by MISR as the negative of the slope of the fitted line of the natural logarithm of τa vs. the natural logarithm of wavelength using all four MISR wavelengths. This version of the τa product has reached a mature stage of validation relative to α as summarized by Kahn et al. [2009]. In total, 3641 MISR pixels, whose spatial distribution is shown in Figure 1, have a valid aerosol retrieval overlapping the 277 ASTER scenes.

[7] Cloud macrophysical properties (fc, cloud size and cloud top height (hct) distribution) were generated from the ASTER scenes at a 17.6 km × 17.6 km domain size overlapping each of these 3641 MISR pixels. fc is derived as the ratio of the total number of cloudy pixels to the total number of pixels in a 17.6 km domain. The cloud size distribution is represented by λ (obtained by single least square fit to the center of each size bin of 100 m width) following Zhao and Di Girolamo [2007]. The hct are retrieved at 90 m spatial resolution using the ASTER 12 μm channel (see Zhao and Di Girolamo [2007] for a detailed methodology). The radiance of each 90 m fully cloudy pixel is converted into a brightness temperature, which is then converted into hct using the temperature profiles measured by the radiosonde at 0Z (closest to Terra's overpass time) on the same day at Minicoy Island (Figure 1). Mean hct were derived by averaging 90 m hct within each of the MISR aerosol pixels. The ASTER hct retrieval may be biased due to a north–south gradient in the vertical temperature profiles and the boundary layer diurnal cycle [McFarquhar and Wang, 2006] because the 0Z sounding was measured at a single site ∼5 hour prior to the Terra overpass. To address this issue, version 17 of MISR-hct, retrieved by stereo technique using the “best quality winds” and independent of temperature profiles [cf. Genkova et al., 2007] was also analyzed. Median hct were calculated from all 1.1 km MISR-hct overlapping each MISR aerosol pixel. The cloud properties were grouped into narrow τa bins of 0.05 width to examine their variability as a function of τa. The large number of samples, even at high τa conditions, allows the observed dependence of cloud properties on τa to be statistically robust as demonstrated below. Because domain size can impact cloud macrophysical properties [Dey et al., 2008], cloud and aerosol properties were regenerated at a 60 km domain corresponding to one ASTER scene and their relations were examined for both domain sizes. We focus our discussion on the 17.6 km domain unless stated otherwise, since aerosol properties from MISR are reported at the 17.6 km domain and we have ten times the sample density for our statistics at 17.6 km domain compared to 60 km domain.

3. Results

[8] The mean (±standard deviation, σ) fc, τa and α of the valid 3641 samples over the Indian Ocean during the study period were 0.074 ± 0.069, 0.232 ± 0.102 and 1.12 ± 0.25, respectively with a wide frequency distribution as shown in Figure 1. τa varies from 0.006 to 1.09 with a strong north (high τa) to south (low τa) gradient (Figure 1). The fc histogram shows a peak (52%) at fc < 0.05 with a gradual decrease in the normalized frequencies at higher fc values. α > 1 for 68% of the time implies the relative dominance of small particles having a significant fraction transported from the South Asian landmass with anthropogenic origins and light absorption properties [Ramanathan et al., 2007]. This is supported by a low mean (±σ) single scattering albedo of 0.82 ± 0.1 at 440 nm retrieved from Level 1.5 cloud-screened AERONET data from the Hanimaadhoo site (location shown in Figure 1).

[9] Figure 2a shows the mean (±σ) fc increasing from 0.029 ± 0.01 to 0.089 ± 0.03 with an increase in τa from 0.1 to 0.3. The increase is significant at 99% confidence level according to two-tailed t-test, and the respective increases in mean fc at each τa bin are also statistically significant at 99% confidence level according to the Mann-Whitney test. There is a statistically insignificant variation of mean fc with τa in the range 0.3 < τa < 0.45 according to two-tailed t-test, beyond which the mean fc decreases by >30% in a statistically significant manner. The mean fc shows a similar response (increasing from 0.026 ± 0.01 at τa = 0.1 to 0.107 ± 0.04 at τa = 0.35 followed by a decrease) to increasing τa for the 60 km domain (Figure S1 of the auxiliary material). The mean fc increases with increasing α for α < 1 (indicating relative dominance of natural aerosols) and subsequently decreases by a rate of 10% per unit α for larger α (inset in Figure 2a). λ decreases from 3.53 ± 0.1 (2.99 ± 0.1) to 3.42 ± 0.1 (2.89 ± 0.07) in the first two τa bins and varies within the uncertainty levels until τa reaches 0.3 (0.25) at 17.6 km (60 km) domain as shown in Figure 2a (Figure S1). The variation of λ at 0.3 < τa < 0.45 resembles the fca variation. A decrease of λ at τa > 0.45, significant at 99% confidence level according to the Mann-Whitney test, indicates a decrease in the normalized frequencies of the small clouds. However, at the 60 km domain size there is a significant and continuous decrease in λ for τa > 0.25 even when fc increases with τa at 0.25 < τa < 0.35. These differences in the variations of fc and λ to increasing τa are attributed to a well-understood domain size effect on the distribution of fc and λ [Dey et al., 2008]. Nevertheless, the trends (i.e., an increase in fc along with insignificant variation in λ followed by a decrease in both fc and λ with increasing τa) are similar at two different domain sizes.

Figure 2.

Variations of (a) mean fc and λ from ASTER (in left Y-axis) and number of samples (in right Y-axis), and (b) the median and 90th percentile of the hct distribution from MISR and ASTER with increasing τa over the Indian Ocean at 17.6 km × 17.6 km domain. The solid vertical bars through mean λ and the dotted vertical bars through mean fc for each τa bin are the uncertainty of the linear fits to the corresponding cloud size distribution and ±1σ, respectively. The values in the X-axis are the upper limits of the τa bins. The variation of mean fc (±1σ shown as vertical bars) with increasing α is displayed in inset of Figure 2a.

[10] The changes in fc in relation to increasing τa and α (a proxy for relative amount of anthropogenic aerosol in this region) are similar to the trends observed over the Amazon [Koren et al., 2008]. However, the increasing and decreasing cloud cover over Amazon were attributed to invigoration of deeper clouds and inhibition of shallow clouds respectively. In our case, the positive correlation between fc and τa at τa < 0.3 may be in part explained by an increase in number of individual clouds. Since λ (representing the cloud size distribution) does not change significantly, the growth of the individual clouds may be retarded by the evaporation-entrainment feedback, hypothesized by Jiang et al. [2006], as aerosol concentration increases. Because invigoration of shallow cumulus clouds is unlikely, the peak of the fca curve over the Indian Ocean is much lower compared to the Amazon. High relative humidity (RH) near cloud edges [Loeb and Schuster, 2008] can also contribute to a positive correlation between fc and τa. The negative correlation between fc and τa for higher τa may be explained by aerosol semi-direct effect [McFarquhar et al., 2004]. Simultaneous decrease in λ at high τa suggests a more efficient desiccation of small clouds relative to large clouds [Koren et al., 2008]. The stabilization of the lower atmosphere due to aerosol heating further contributes to this decrease in fc. Since drizzle did not occur frequently in either clean or polluted clouds during INDOEX [Heymsfield and McFarquhar, 2001], the cloud-lifetime effect may not be an important factor here.

[11] ASTER-hct shows insignificant variation with τa for τa < 0.4. However, the ASTER infrared retrieval of hct may be impacted by the use of a temperature profile from a single site, unlike the MISR stereo retrieval of hct. The median hct from MISR displayed in Figure 2b also shows relative insensitivity to τa for τa < 0.4, but reduces significantly to 541 m at τa > 0.5. The 90th percentile of the MISR-hct distribution increases from 893 m to 1638 m with τa between 0.1 to 0.4 and decreases significantly according to the two-tailed t-test for τa > 0.4. The reduction in hct in both ASTER and MISR datasets at large τa is consistent with previous INDOEX observations [McFarquhar et al., 2004] and the hypothesis of stronger downward motion at cloud tops due to enhanced cooling by evaporation in polluted conditions [Xue and Feingold, 2006].

[12] McFarquhar and Wang [2006] showed that fc and hct are reduced due to the semi-direct effect when aerosols are present within the cloud layer; however, only fc is reduced when aerosols with similar absorption characteristics are present below the cloud layer. The hct increases in the latter case because warmer air has to rise further to reach the lifting condensation level due to enhanced radiative heating below clouds relative to the air above clouds. This implies a larger variability in the response of hct compared to fc to increasing aerosol concentration when aerosols are distributed above, within and below the cumulus cloud layers. Although coincident aerosol vertical profiles are not available, aerosols were distributed within 0.5–3 km during this period [Ramanathan et al., 2007] compared to hct mostly confined to within 2 km from surface. This may explain the insensitive response of hct at τa < 0.4 compared to fc observed here in relation to increasing τa [McFarquhar and Wang, 2006].

4. Discussion and Conclusions

[13] Although undetected clouds at the sub-pixel level can increase τa and decrease α, the observed relationships are not influenced by cloud contamination because direct cumulus contamination is negligible on MISR aerosol retrievals [Zhao et al., 2009]. The entire regime of the ‘umbrella’ shaped relations of fca and hct distribution-τa and decreases in relative frequencies of small clouds cannot be explained by radiative cloud adjacency effects because such effects would result in an increase in τa and α (due to a larger increase of τa at shorter wavelengths than longer wavelength near the clouds) [Marshak et al., 2008] with fc. The fc, λ and hct may increase even without any influence of aerosols, as the boundary layer deepens and the air mass traveling over the Indian Ocean towards the equator moistens. But, the cloud macrophysical properties do not show any latitudinal gradient (not shown here) in contrast to τa decreasing from north to south (Figure 1), and hence, the variations of cloud properties with increasing τa are consistent with the hypothesized aerosol effect.

[14] However, a big challenge in quantifying aerosol effects on clouds is to disentangle them from meteorological influences. Here, variations of lower tropospheric potential temperature difference (Δθ = θ700 − θ1000), sea surface temperature (SST), RH below700 hPa (RH>700), vertical wind at 850 hPa (ω850) and precipitable water (PW), derived from National Centers for Environment Prediction (NCEP)-reanalysis data, with τa (Figure S2) were considered to test their potential influence on the variability of observed cloud properties. RH>700, Δθ and SST show no statistical dependence on τa, while PW shows a decrease (statistically significant at τa > 0.35) and ω850 shows an increase with increasing τa for τa < 0.35. Their variability with τa suggests that the synoptic-scale meteorology cannot explain the changes in the cloud properties across the entire spectrum of τa. For example, the reduction in fc at high τa may partially be influenced by the decrease in PW, but then the three-fold increase in fc with an increase in τa from 0.1 to 0.3 cannot be explained by a decrease in PW or by an increase in ω850 (suggesting increasing subsidence). In the absence of synoptic scale changes in meteorological forcing, the lack of statistical significance in the variation of RH>700 with τa is not surprising given the expected small hygroscopic effect on τa in our observed range of RH>700 relative to the very large variability in τa. Moreover, although the turn-over point varies, the persistence of ‘umbrella’ shaped fca curves over a wide range of RH>700 (Figure S3) indicates the dominance of a strong aerosol forcing over other meteorological forcings in influencing the trade-wind cumulus properties sampled. Nevertheless, the potential influence of mesoscale meteorology cannot be ruled out, as data at finer resolution than NCEP are not available to carry out the analysis.

[15] The observed relationships between aerosol and shallow cumulus cloud properties, consistent with a transition from increasing to decreasing fc with τa, implies that investigations of such relationships in ‘polluted’ and ‘clean’ regimes [McFarquhar et al., 2004] may produce varying results depending on the threshold between ‘clean’ and ‘polluted’ conditions. The mean fc from ASTER over the Indian Ocean (0.074) and the RICO domain (0.086 [Zhao and Di Girolamo, 2007]) are comparable. But, the Indian Ocean has more small clouds (mean λ of 3.01 ± 0.05) than RICO (mean λ of 2.85 ± 0.05) at the same 60 km × 60 km domain, similar to the simulations of more small cumuli in polluted conditions compared to clean conditions based on RICO soundings [Jiang et al., 2009]. Quantitative differences in the variation of cloud properties with increasing τa over the Indian Ocean and RICO region are expected, given the larger τa and higher aerosol absorption over the Indian Ocean. In fact, this contrast emphasizes the importance of regional comparisons of aerosol-cloud interaction under varying environmental conditions with models that can resolve the changes in cloud properties in response to changes in aerosol properties, in an effort to test our current understanding of aerosol-cloud interactions and their role in weather and climate.

[16] Due to the limitation in collecting ASTER data over the oceans the observations presented here were restricted to a six month period and do not account for inter-annual variability. To account for this variability, a dedicated high-resolution satellite imager for collecting cloud and aerosol properties over the ocean is required. Until then, fc data from more common sensors (e.g., MISR and MODerate resolution Imaging Spectrometer, MODIS) must be used to cover multi-season, multi-year observations to improve the understanding of aerosol impacts on trade-wind cumuli (and other clouds). However, the true observed relationships between aerosol properties and fc may be difficult to assess because MISR and MODIS fc calculated from the respective cloud masks reported at 1.1 and 1 km resolutions over the trade-wind cumuli dominated regions tend to be larger than the true fc. The amount of overestimated fc is not linearly related to the true fc and varies with the spatial distribution of clouds [Zhao and Di Girolamo, 2006]. The observed changes in cloud size distribution in response to increasing τa may also be impacted because of the shifting of cloud size distribution to larger cloud size due to amalgamation of many small clouds into fewer large clouds at these coarse resolutions [Dey et al., 2008]. Hence, a strategy to correct the biases in MISR and MODIS cloud masks is needed before using those data to examine such complex relations.

Acknowledgments

[17] This research is partially supported by Jet Propulsion Laboratory of the California Institute of Technology under contract 1260125, the National Oceanic and Atmospheric Administration under grant NA06OAR4310003 and the National Aeronautic and Space Administration under NNG06GB92G. The MISR, ASTER, meteorological and sounding data were obtained from the Atmospheric Sciences Data Center at the NASA Langley Research Center, the Land Process Distributed Active Data Center, NCEP-reanalysis data set and University of Wyoming website, respectively. We acknowledge two anonymous reviewers whose comments helped to improve the manuscript.

Ancillary