Research Article

Phytoplankton and Cloudiness in the Southern Ocean

See allHide authors and affiliations

Science  01 Dec 2006:
Vol. 314, Issue 5804, pp. 1419-1423
DOI: 10.1126/science.1131779

This article has a correction. Please see:


The effect of ocean biological productivity on marine clouds is explored over a large phytoplankton bloom in the Southern Ocean with the use of remotely sensed data. Cloud droplet number concentration over the bloom was twice what it was away from the bloom, and cloud effective radius was reduced by 30%. The resulting change in the short-wave radiative flux at the top of the atmosphere was –15 watts per square meter, comparable to the aerosol indirect effect over highly polluted regions. This observed impact of phytoplankton on clouds is attributed to changes in the size distribution and chemical composition of cloud condensation nuclei. We propose that secondary organic aerosol, formed from the oxidation of phytoplankton-produced isoprene, can affect chemical composition of marine cloud condensation nuclei and influence cloud droplet number. Model simulations support this hypothesis, indicating that 100% of the observed changes in cloud properties can be attributed to the isoprene secondary organic aerosol.

Marine aerosols strongly affect properties and lifetime of stratiform clouds, influencing Earth's radiation budget and climate. The role of oceanic biota in modifying chemical composition and size distribution of marine cloud condensation nuclei (CCN) has been one of the most intriguing questions in climate studies. Production of sulfate from the oxidation of dimethylsulfide (DMS), proposed by Shaw (1) and explored by Charlson et al. (2)[the CLAW hypothesis, named after the authors of the paper, Charlson, Lovelock, Andreae, and Warren (2)] and primary emissions of biogenic organic matter from wave breaking (3, 4) have been suggested as possible mechanisms by which phytoplankton can modulate properties of marine clouds. In this work, remotely sensed data from the Moderate Resolution Imaging Spectroradiometer (MODIS) and Sea-viewing Wide Field-of-view Sensor (SeaWiFS) were combined with the National Center for Environmental Prediction (NCEP)–generated meteorological fields to examine the effects of ocean productivity on cloud microphysical and radiative properties and explore an alternative pathway by which phytoplankton may affect marine CCN. The data analysis is carried out in the Southern Ocean (SO) near South Georgia Island. Because of its unique spatial location and topography, waters near the island are a natural laboratory for investigating the effects of marine productivity on clouds. Waters in this area can support massive phytoplankton blooms (5, 6), with chlorophyll a concentrations ([Chl a]) more than an order of magnitude higher than the background (7, 8). Because the surface [Chl a] in this area can be used as a reliable proxy for the primary production (5, 9), satellite retrievals can provide the link between ocean productivity and clouds. Strong and persistent westerlies (10) make it possible to examine cloud properties upwind and downwind of the bloom, and the periodic nature of the bloom is ideal for exploring the temporal relationship between ocean productivity and regional clouds.

Temporal correlation between [Chl a] and Reff. We restrict our analysis to liquid-water clouds. Figure 1 shows surface [Chl a] and effective radius (Reff) in the SO for the 6 years of available data. Figure 1A demonstrates that the observed enhancement of primary productivity near South Georgia Island is a localized phenomenon that typically occurs between September and February. Despite its regularity, Fig. 1 shows substantial annual variations in the bloom's temporal appearance, spatial extent, and strength, with strong anticorrelation between [Chl a] and Reff. In 2001 and 2002, the smallest Reff coincided with the largest enhancement in [Chl a], whereas the largest summertime Reff was observed during the austral summer of 2000 and 2001 with negligible phytoplankton levels. Such systematically significant temporal anticorrelation of [Chl a] and Reff suggests a link between ocean biological productivity and marine cloud properties; however, the results shown in Fig. 1 alone cannot be used to ascertain a causal relationship, given that other factors such as variation in background aerosol size distribution and cloud dynamics may also affect Reff. Phytoplankton productivity and clouds could also be influenced by large-scale atmospheric circulation; such a mechanism, if it exists, would exhibit a high correlation between [Chl a] and regional cloud properties even if ocean productivity had no effect on clouds.

Fig. 1.

The 8-day averaged (A) SeaWiFS-observed chlorophyll a and (B) MODIS-retrieved cloud effective radius. Data for [Chl a] is gridded at a resolution of 9 by 9 km and zonally averaged between 49°S and 54°S; data for Reff is gridded at a resolution of 1° by 1° and averaged in the area of 49° to 54°S and 35° to 41°W. White areas in (A) indicate missing data.

Effect of phytoplankton on clouds. To constrain the effect of phytoplankton on clouds, we adopt an approach similar to Chylek et al. (11). A rectangular geographical area was selected in the SO from 55°W to 21°W and 42°S to 60°S, then divided into 153 cells of 2° by 2°. The background colors in Fig. 2A correspond to the monthly averaged surface [Chl a] for the year with the largest bloom on record (Fig. 1). Increasing grid box number on Fig. 2B traverses through a row of cells in an eastward direction. Sharply increased [Chl a] between 48°S and 56°S indicates enhanced marine productivity near South Georgia Island.

Fig. 2.

(A) Monthly averaged (11 December 2001 to 8 January 2002) 4-km resolution SeaWiFS-observed surface [Chl a]. The black color over the ocean denotes the missing data due to clouds. The South Georgia Island boundary (54.3° to 55.0°S, 35.3° to 38.3°W) is located between cells 111 and 112. (B) The 2° by 2° square monthly averaged SeaWiFS surface chlorophyll concentration (green), MODIS cloud-top effective radius of liquid droplets (blue), estimated cloud droplet number concentration (red), aerosol optical thickness (black), and NCEP reanalysis–generated surface wind speed (purple) as a function of the grid cell number. Tick markers at every 17 cells correspond to the starting point of the next west-to-east row in (A). Broken line indicates the missing data.

Figure 2B suggests that average Reff for background clouds in this area is ∼14 μm, with a sharp decrease (∼10 μm) in the vicinity of the bloom. Although MODIS-retrieved Reff may be biased to larger sizes compared with in situ measurements, it is reasonable to expect that errors in observed relative changes of Reff are small (12). This figure demonstrates that, on average, water clouds near the bloom region have effective droplet radii 30% smaller than those of background clouds over the SO.

Cloud droplet number concentration (CNDC) can be used as a direct microphysical link between the biology and cloud properties. From the remote sensing data, CDNC (cm–3) can be estimated as (13, 14): Embedded Image(1) Embedded Image where LWC is cloud liquid water content; ρw is the density of liquid water; H is the cloud thickness; τ and Reff are MODIS-observed cloud optical depth and effective droplet radius, respectively; and k is the constant ∼0.8 (14). H was estimated as the distance between the cloud lifting condensation level, zLCL (used as a cloud base proxy), and the cloud-top height. The zLCL is a strong function of relative humidity and temperature and is calculated using NCEP reanalysis surface data (15). Cloud-top height was computed by matching adiabatic liquid water path (LWP) with the MODIS-observed LWP. Because the average Reff in the study area was typically less than the threshold effective radius for precipitation (∼14 to 15 μm) (16), we assumed that precipitation did not cause substantial deviation of observed LWP from the adiabatic value. The calculated liquid water cloud thickness was between 250 and 400 m, consistent with observations (17, 18). The uncertainty in H does not contribute considerably to the Reff variability.

Figure 2B shows that the calculated monthly averaged CDNC outside the bloom area is ∼55 cm–3, whereas in the bloom region, CDNC increases sharply approaching twice the background level. Comparison of these values with the average summertime CDNCs for the Southern Ocean Cloud Experiment (SOCEX) shows that the estimated CDNC outside the bloom area compares well with 57 cm–3 reported for the SOCEX baseline conditions, whereas calculated CDNCs over the bloom are close to ones reported for the clouds affected by anthropogenic emissions (109 cm–3) (19). This comparison suggests that the magnitude of variation of the SO marine cloud microphysics over the bloom may be comparable to anthropogenic indirect aerosol effects.

Other factors that may potentially affect CDNCs near the bloom are long-range transported Patagonian dust and sea salt. Because winds in the SO typically flow eastward (10), the presence of mineral dust will be manifested by noticeable decrease in aerosol optical thickness (AOT) from west to the east. Figure 2B shows that this is not the case. The enhanced productivity near South Georgia Island is primarily controlled by ocean upwelling, not by dust-Fe fertilization (5, 6, 9). Therefore, dust is not responsible for changes in either ocean productivity or Reff. An increase in submicrometer-sized sea-salt particles can certainly affect CDNC, but this must be accompanied by an increase in surface wind speed over the bloom, which we did not observe (Fig. 2B). Although the general trend of increase in CDNC between 42°S and 54°S and decrease further southward can be associated with the variation in the surface wind speed (Fig. 2B), there is no clear relationship between the two near the bloom region.

Analysis of Fig. 2 indicates strong coupling between observed changes in marine biological productivity and microphysical properties of warm clouds over the bloom. We examined the possibility that both processes are driven by the same large-scale influence.

Role of meteorology. Strong winds (associated with cyclonic circulation) can cause vertical mixing and upwelling of nutrient-rich waters from below the mixed-layer depth, fueling photosynthesis and causing large-scale phytoplankton blooms. Such deep water entrainments may also be associated with a depression in sea-surface temperatures (SST) (20, 21) that may last up to 2 weeks and considerably influence properties of marine stratocumulus clouds (22). Cyclones would therefore generate a correlation (but no causality) between [Chl a] and cloud properties. The results of linear multiple regression analysis of satellite-retrieved and model-generated parameters shown in Table 1 suggest that cloud properties over the bloom are not influenced by cyclonic winds. To quantify the possible influence of phytoplankton on clouds, the analysis was carried out separately for the region with enhanced productivity (48°S to 56°S) near South Georgia Island and the areas with relatively low [Chl a] (42°S to 48°S and 56°S to 60°S), hereafter referred to as the inside and outside regions, respectively.

Table 1.

The influence of meteorological parameters and [Chl a] on the Reff in the Study area of the Southern Ocean. Meteorological parameters used as independent variables were selected according to Kaufman et al. (56). We analyzed dependence of Reff on (i) MODIS-retrieved AOT, SST, cloud-top temperature (indicator of cloud height), and total precipitable water vapor (indicator of convergence); (ii) SeaWiFS-observed [Chl a]; and (iii) NCEP and National Center for Atmospheric Research (NCAR) reanalysis–generated surface wind speed, equivalent potential temperature difference between 500 and 925 hPa, and the broad-scale vertical motion at 850 hPa. The logarithm of the AOT is used to reduce nonlinearity in the regression (56). The parameters are ranked by order of importance based on the correlation with Reff in the outside region. Columns to the right of correlation coefficients show the change in Reff associated by the multiple regression with the changes in meteorological parameters and [Chl a]. To compare multiple regression coefficients of variables of different magnitudes and dispersions, we standardized all variables by subtracting the mean and dividing the result by the standard deviation. Therefore, coefficients given in the “Change in Reff” columns show the average amount of change in Reff when each meteorological parameter and [Chl a] change by one standard deviation, while keeping others constant. The range around the sample regression coefficients was determined for 95% confidence interval.

Outside the bloom area (42°S to 48°S and 56°S to 60°S)Inside the bloom area (48°S to 56°S)
Correlation to ReffChange in ReffCorrelation to ReffChange in Reff
Total column precipitable water vapor 0.67 ± 0.08 0.48 ± 0.10 -0.02 ± 0.12 -0.02 ± 0.10
Sea-surface temperature 0.65 ± 0.08 0.63 ± 0.09 0.01 ± 0.12 0.01 ± 0.11
Surface wind speed -0.52 ± 0.09 -0.50 ± 0.11 -0.14 ± 0.12 -0.15 ± 0.10
ln(AOT) -0.43 ± 0.10 -0.41 ± 0.09 -0.01 ± 0.12 -0.01 ± 0.13
Potential temperature difference 0.34 ± 0.10 -0.40 ± 0.11 -0.09 ± 0.12 -0.08 ± 0.11
Vertical velocity -0.27 ± 0.10 0.23 ± 0.09 -0.10 ± 0.12 -0.12 ± 0.10
Eastern wind at 850 hPa 0.22 ± 0.10 0.27 ± 0.12 0.10 ± 0.12 0.01 ± 0.11
Cloud-top temperature -0.02 ± 0.11 -0.01 ± 0.09 -0.02 ± 0.12 -0.02 ± 0.11
[Chl a] 0.18 ± 0.14 0.19 ± 0.10 -0.48 ± 0.11 -0.49 ± 0.09

The analysis addresses two main questions: (i) Do meteorological parameters and [Chl a] affect Reff differently in the inside and the outside regions? Table 1 shows a strong difference in the relationship of meteorological parameters and [Chl a] with Reff in the inside and outside regions. Outside, Reff is mainly controlled by large-scale atmospheric parameters (i.e., column precipitable water vapor, SST, surface wind speed, and AOT), and correlation between Reff and [Chl a] is minor. The relationship between [Chl a] and Reff in the inside region is markedly different from that of the outside region; inside, the effect of [Chl a] on Reff has by far the strongest impact of all parameters examined. (ii) Can changes in meteorological parameters affect Reff while changing [Chl a]? Linear multiple regression analysis suggests that in the outside region, the change in Reff is primarily associated with variability in meteorological parameters, whereas in the inside region, [Chl a] is the single most important parameter controlling the Reff (Table 1).

As a result of this analysis, we concluded that over the bloom, the relationship between the ocean productivity and Reff of warm clouds is unique to these two variables and does not extend to large-scale meteorological parameters—i.e., biological productivity is the prime cause for changes in cloud microphysical and radiative parameters.

Radiative forcing. The perturbation in short-wave radiation (ΔF) at the top-of-the atmosphere (TOA) within our study area is estimated as (23) Embedded Image(2) where Fin is the monthly averaged solar flux at the top of marine liquid clouds calculated with the NASA Global Modeling Initiative (GMI) with implemented shortwave radiative transfer code (24); Rc and Ac are monthly averaged MODIS-observed cloud albedo and cloud fraction, respectively; and ΔlnNdb is relative change in calculated CDNC [ΔlnNdb =(NdNb)/Nd], where Nd and Nb are average droplet numbers over the bloom and in the background air, respectively. Because of variability of background CDNC in zonal direction (Fig. 2B), Nb was estimated separately for each longitude using a linear fit between the locations farthest from the bloom (square numbers 1 and 17, 18 and 34, and so on, in Fig. 2A).

Figure 3 shows considerable variation in TOA mean short-wave forcing in the study area resulting from changes in properties of liquid clouds. However, the remarkable feature of Fig. 3 is the very strong cooling near the bloom, reaching ∼–15 W m–2. Such a large change in TOA short-wave radiation is comparable in magnitude with the aerosol indirect effect in highly polluted regions (2426), highlighting the need for improved quantification of interactions between marine biota, aerosols, clouds, and climate. Clearly, the link between ocean productivity and change in cloud properties is in the modification of CCN. We next examined whether production of secondary organic aerosol (SOA) from the oxidation of phytoplankton-produced isoprene can lead to considerable changes in marine CCN.

Fig. 3.

Change in TOA short-wave radiation. The radiative effect was evaluated for the change in albedo of warm marine clouds. Calculations are carried out using monthly averaged MODIS-observed data at 1° by 1° resolution and the GMI-supplied monthly averaged solar flux at 4° by 5° resolution.

Phytoplankton isoprene SOA and its effect on CCN. Although organosulfur emissions and the transfer of surface active organic matter from the oceanic surface layer to the atmosphere have been studied in detail, little is known about the effect of phytoplankton-produced nonmethane hydrocarbons (NMHC) on marine aerosol. Oceans are known to be a potential source of NMHC and particularly isoprene (2729); observed atmospheric concentrations of isoprene in remote SO are high (∼0.25 parts per billion by volume) (30)—about one-fifth the amount of typical boundary-layer isoprene concentration over the Amazon (31). Atmospheric oxidation of isoprene may lead to formation of SOA (3234). Because SOA concentrations in remote marine regions are very small (35), ocean-emitted isoprene could contribute considerably to the organic fraction of marine CCN. We propose that isoprene SOA can affect CCN composition and contribute to the observed changes in cloud properties. To evaluate the plausibility of this hypothesis, we estimated atmospheric concentrations of isoprene, the resulting SOA, and its subsequent effect on CCN. Table 2 summarizes [Chl a] inside the bloom, estimated sea-air fluxes, and concentrations of isoprene in the marine boundary layer (MBL). Seawater-dissolved isoprene in the bloom region (Embedded Image) was estimated with the approach of Palmer and Shaw (36).

Table 2.

Ocean chlorophyll a, fluxes, and atmospheric concentrations of isoprene. [Chl a] inside the bloom is based on SeaWiFS observed chlorophyll a data retrieved at 9-km resolution. Maximum and minimum concentrations correspond to the middle and the edge of the bloom. Embedded Image was estimated according to Palmer and Shaw (36). The isoprene production rate was calculated by multiplying suggested rates of 1.8 ± 0.7 μmol isoprene produced (grams phytoplankton chlorophyll a)–1 day–1 by SeaWiFS [Chl a]. The range of dissolved isoprene is from Wingenter (57). Embedded Image was estimated using SOFeX-N measured surface values and scaled with SeaWiFS [Chl a]. The sea-air isoprene flux F was parameterized as F = kI(CswCBL/H), where kI is the piston velocity, Csw is the estimated seawater concentration of isoprene in the bloom, CBL is marine boundary layer isoprene concentration, and H is the Henry's law constant for isoprene (58). For the typical range of atmospheric and oceanic isoprene concentrations, the second term in the equation is at least an order of magnitude smaller (38) and therefore is ignored here. The Embedded Image (59), where U10 is NCEP reanalysis–generated monthly averaged wind speed at 10-m height, and Sc is the Schmidt number calculated using isoprene molar volume and MODIS-observed ocean temperatures. Amazon fluxes are the median fluxes over the tropical forest site of the Peruvian Amazon (31). Assuming that isoprene oxidation had no significant impact on OH levels in the MBL (60), the average MBL isoprene concentration above the bloom was calculated with the use of FB, the average isoprene lifetime of 2 hours and boundary layer height of 600 to 1000 m. The SOA concentration was estimated with the use of global isoprene SOA yield of 3% (35, 46).

[Chl a] (mg m-3)Dissolved isoprene concentration (nM)Isoprene flux (108 molecules cm-2 s-1)Estimated MBL concentration (ng m-3)
BloomSOFeXEmbedded Image SOFeXEmbedded Image F A F B AmazonIsopreneSOA
Average 3.0 2.4 0.03 31.4 36.3 1.8 2370 18200 1920 50
Max 12.7 2.6 0.13 >40 145 8.6 9470 20000 7700 230
Min 0.1 0.1 0.003 <10 6.1 0.2 395 7000 320 5

For comparison, in Table 2 we include observed [Chl a] and dissolved isoprene concentrations for the iron-enriched North Patch of the Southern Ocean Iron Enrichment Experiment (SOFeX-N) (37). The average surface ocean [Chl a] measured for SOFeX-N and for the bloom in our study area are comparable (Table 2), yet SOFeX-N isoprene concentration was more than three orders of magnitude higher than estimated for the bloom. The chlorophyll content of seawater, as sensed by the satellite, is related to the rate of isoprene production (38). Given that both blooms had similar average [Chl a], were located in the SO at comparable latitudes, had similar SSTs and mixed-layer depths, and occurred in the same season, such large discrepancies in dissolved isoprene concentrations could arise from the difference in phytoplankton species. Laboratory studies show that isoprene production rates between diatoms, dinoflagellates, and cocolithophores can vary over orders of magnitude (28, 39, 40). Production rates suggested in (36) and used in Table 2 for Embedded Image are applicable to species typical for the oligotrophic oceans (cyanobacteria, picoeucariotes, and coccolithophores) and are likely not representative of the blooms in the SO where microphytoplankton typically contributes >70% of total cell counts, followed by nano- and picophytoplankton (4143). Inside the bloom, diatoms were five times more abundant than the next contributor, dinoflagellates; species measured (in order of their abundance) were Thallasiosira sp. 1, Nitzschia spp., and Chaetoceros spp. (41). In the absence of any local data and good similarity with SOFeX-N (42), dissolved isoprene in the bloom was estimated with SOFeX-N measured isoprene concentration and scaled with [Chl a]. Estimated sea-air fluxes of isoprene are orders of magnitude higher than the monthly averaged fluxes (0.4 × 108 to 8 × 108 molecules cm–2 s–1) suggested for the SO (36), indicating that the global marine isoprene flux of ∼0.1 Tg C year–1 (36) should be viewed as the low-end estimate. In Table 2, we also include estimated MBL isoprene and SOA concentrations over the bloom.

To evaluate the role of this potentially important organic aerosol source on cloud microphysics, we calculated CDNC using a cloud parcel model (44). The simulations were performed for different concentrations of organic aerosol, non–sea-salt (nss) sulfate, and updraft velocities. The range of SOA concentrations examined (0 to 250 ng m–3) corresponds to the estimated SOA over the background air and at the center of the bloom (Table 2). In all simulations, aerosol number remains constant; therefore, addition of SOA corresponds to the condensation growth and aging of marine aerosols (44). Ambient measurements and chamber experiments showed that 2-methyltetrols and C5 alkene triols are some of the main particulate-phase oxidation products of isoprene under low-NOx conditions (32, 4548). Because the reaction pathways leading to production of isoprene SOA and chemical properties of the oligomers with high molecular weight remain little understood (45), we assumed that ∼20% of organic particulate mass is water soluble with chemical properties corresponding to 2-methyltetrols (44). Cloud-base updraft velocity is constrained by estimated CDNCs outside the bloom region. The background CDNC (the low left corner on Fig. 4, A and B) is well represented by trimodal marine aerosol size distribution and chemical composition (44) with updraft velocities between 0.35 to 0.5 ms–1, typical of marine stratiform clouds (15, 49). Figure 4 shows that the excess amount of organic mass increases CDNC; modification of the ambient size and CCN activity of marine aerosols due to addition of organic mass can explain up to 60% of the droplet number concentration over the bloom.

Fig. 4.

(A and B) Contours of cloud droplet number concentration (cm–3) as a function of chemical composition and updraft velocity.

In addition to enhanced concentration of dissolved isoprene, ocean waters in blooms have commonly been characterized by elevated levels of DMS, atmospheric oxidation of which is a major source of nss sulfate in remote marine air. Model simulations show (left side of Fig. 4) that for the range of nss sulfate measured over the SO (50) sea-salt CCN and nss CCN may account for the enhanced CDNC over the bloom. However, Fig. 4 also shows that when mixed with SOA, even the minimal concentration of nss sulfate can fully account for the observed enhancement of CDNC. This is important, considering that regions with high isoprene productivity may not coincide with elevated levels of DMS (30) or under conditions with biological net DMS consumption in a bloom (51).

Our model results suggest that phytoplankton isoprene emissions could contribute to the organic fraction of marine CCN and be a viable mechanism by which ocean biota may affect properties of shallow marine clouds. We propose that SOA of marine origin can act synergistically with the established mechanisms (14) and lead to changes in marine CCN chemical composition and number concentration.

Discussion and conclusions. Analysis of remotely sensed data indicates that over the enhanced biological productivity region of the SO, cloud droplet number was doubled and the effective radius was decreased by more than 30%. Analysis of data revealed that changes in the properties of warm clouds over the bloom were primarily associated with the enhanced ocean biological productivity. These changes can lead to a TOA short-wave radiative forcing of –15 W m–2, comparable to the aerosol indirect effect over highly polluted regions of the globe. We propose that SOA formed from the oxidation of ocean-emitted isoprene can account for the observed change in properties of shallow marine clouds over the bloom. Model simulations presented support this hypothesis, making ocean isoprene emissions a viable mechanism by which marine biota may affect properties of shallow clouds. Considering that isoprene SOA can be an important source of marine aerosol organic mass, this unaccounted SOA may partly reconcile the large organic aerosol source missing from current global models (52). Cooperative efforts of researchers from different fields are required to provide accurate estimates of sea-air fluxes of biogenic volatile organic compounds (VOCs) in different parts of the ocean. Work is also needed to constrain the chemical composition of SOA in marine environments and its effect on aerosol activation. Future campaigns may provide the evidence for the importance of this new source of organics in the SO and the viability of the proposed mechanism. Given that the evolution of microalgae can be affected by anthropogenic air pollutants (53, 54) and environmental changes (55), the proposed mechanism of SOA formation in remote marine air may need to be included in global models. Because the average concentration of [Chl a] in the bloom was similar to that of SOFeX-N, which is thought to be representative of the glacial era concentrations of Fe in the SO (37), we propose that SOA from phytoplankton-produced isoprene may have played a considerable role in climate transition, perhaps amplifying the negative feedback loop suggested by the CLAW hypothesis.

Supporting Online Material

Materials and Methods


References and Notes

View Abstract

Stay Connected to Science

Navigate This Article