Submerged oceanic bubbles, which have a much longer life span than whitecaps
or bubble rafts, have been hypothesized to increase the water-leaving
radiance and thus affect satellite-based estimates of water-leaving radiance
to non-trivial levels. This study explores this effect further to determine
whether such bubbles are of sufficient magnitude to impact satellite aerosol
optical depth (AOD) retrievals through perturbation of the lower boundary
conditions. There has been significant discussion in the community
regarding the high positive biases in retrieved AODs in many remote ocean
regions. In this study, for the first time, the effects of oceanic bubbles
on satellite retrievals of AOD are studied by using a linked Second Simulation
of a Satellite Signal in the Solar Spectrum (6S) atmospheric and HydroLight
oceanic radiative transfer models. The results suggest an insignificant
impact on AOD retrievals in regions with near-surface wind speeds of less
than 12 m s
The remote sensing community has long noticed anomalously high aerosol optical depth (AOD) retrievals over high wind belts of the southern oceans, North Pacific, and North Atlantic (e.g., Myhre et al., 2005; Zhang and Reid, 2006, 2010; Shi et al., 2011a, b; Smirnov et al., 2011; Toth et al., 2013; Kalashnikova et al., 2013; Chin et al., 2014). Some passive retrievals of AOD from satellites observe a belt of high AOD over the southern oceans known as the enhanced southern oceans anomaly (ESOA) that is especially biased when compared with ship-based measurements of AOD. These anomalously high values are thought to be in part due to a combination of cloud contamination and enhanced radiance from the ocean surface from whitecaps and bubble rafts. Given the size of the oceans, even small but consistent biases can have a significant influence in the overall estimated top-of-atmosphere (TOA) radiative forcing by aerosol particles.
The University of North Dakota and the Naval Research Laboratory have been systematically investigating persistent oceanic biases in satellite AOD estimates. Early studies first verified that the high oceanic AOD belts were in fact highly biased (Smirnov et al., 2011; Toth et al., 2013). This was then followed by the most logical factor, cloud contamination. Indeed, a series of studies suggests that most of the high bias is related to clouds. However, there is a clear lower boundary condition signal as well, with increasing positive AOD bias with wind speed (e.g., Zhang and Reid, 2006; Shi et al., 2011a). Given that sea salt aerosol production, specular reflection (sun glint), and whitecapping all covary with wind speed, AOD retrievals are a potentially confounded system.
An underwater image of bubbles generated by
plunging waves. The picture was taken using an underwater bubble camera
system designed to measure the number density of bubbles over a size range
of 40–800
Some Level 3 products (e.g., Zhang and Reid, 2008; Shi et al., 2011a) include an empirical correction for wind-speed-related bias to retrieved AOD. Some Level 2 satellite retrievals (e.g., Sayer et al., 2010, 2012; Jackson et al., 2013; Levy et al., 2013; Limbacher and Kahn, 2014) also incorporate wind speed data into the radiative transfer calculations using parameterizations of wind effects on whitecaps and bubble rafts. The current study uses a unique combination of data sets to further investigate the mechanics of the ocean lower boundary condition.
Whitecaps and the resulting bubble rafts form an easily identifiable
perturbation to the ocean surface reflectivity. However, there is another
consideration: subsurface bubbles (e.g., Fig. 1). While whitecaps last for
only seconds, subsurface bubbles can have a much longer lifetime (e.g.,
Johnson and Cooke, 1981). Theoretically, an air bubble in pure water would
either rise to the surface under buoyancy (Harper, 1972) or dissolve
under surface tension pressure (Epstein and Plesset, 1950). In
open ocean environments, bubbles are found to be coated with organic and
surfactant materials (Fox and Herzfeld, 1954; Yount, 1979). The
coating process prevents gas diffusion and stabilizes the bubbles against
buoyancy (Fox and Herzfeld, 1954; Yount, 1979). While rising
bubbles burst at the surface and form whitecaps and bubble rafts, stabilized
bubbles can stay in water for hundreds to thousands of seconds (Johnson and
Cooke, 1981). Under moderate wind conditions (> 3 m s
The goal of this study is to evaluate whether subsurface bubbles pose a lower boundary condition problem for aerosol remote sensing. Already these stabilized bubbles have been recognized as a complicating term in retrievals of water-leaving radiance (Zhang, 2001; Zhang and Lewis, 2002; Flatau et al., 2000). While previous efforts to empirically correct the lower boundary condition of aerosol products for wind implicitly incorporates the entirety of the lower boundary condition-specular reflection, whitecaps, bubble rafts, and submerged bubbles, such methods are neither applicable to joint retrievals of ocean and atmospheric products together nor acceptable for higher resolution retrievals such as those performed by aircraft-mounted sensors.
While water-leaving reflectance is a subcomponent of the ocean surface
reflectance (e.g., Koepke, 1984; Vermote et al., 1997), the contribution of
subsurface bubbles to water-leaving radiance in relation to other ocean
features has yet to be explored in the context of aerosol retrievals. Within
a pixel of satellite observation, whitecaps are sporadic and scattered
whereas bubbles in water form a more or less uniform layer that could exist
over regions that are free from whitecap contamination (e.g., Monahan and Lu,
1990). While whitecaps serve as a diffuse reflector, reflecting solar
radiation directly at the surface (Frouin et al., 1996; Whitlock et al.,
1982), bubbles interact with light below the surface, enhancing
water-leaving radiance (Stramski and Tegowski, 2001; Terrill et al.,
2001; Zhang et al., 1998). Studies have shown that the contributions of
bubbles to water-leaving radiance are rather significant. For example, Zhang
et al. (1998) found that organic coatings on bubbles will significantly
enhance the backscattering and proposed that bubbles could be the strongest
contributor to the light coming out of ocean. Stramski and Tegowski (2001)
illustrated the temporal variation of the light field caused by bubble
injection under a wave breaking event (wind speed
In this study, through a theoretical approach, the impacts of subsurface
oceanic bubbles on satellite aerosol-retrieved AODs are studied, especially
over the ESOA region. The effects of oceanic bubbles on satellite AODs are
examined theoretically, using a linked oceanic and atmospheric radiative
transfer model (RTM). The HydroLight oceanic RTM (Mobley et al., 2012) is
used to estimate the bubble-induced perturbations in surface reflectance as
a function of near-surface wind speed. The HydroLight simulated bubble
concentration and surface reflectance relationship is further incorporated
into the Second Simulation of a Satellite Signal in the Solar Spectrum (6S)
atmospheric RTM (Vermote et al., 2006) for estimating the impact of bubbles
to the TOA radiation. Note that in the blue and green
parts of the visible spectrum, it is difficult to separate the contributions
of bubbles from the total background reflectance due to multiple scattering
(e.g., Zhang, 2001). In the red/infrared spectral ranges with strong
absorption due to water molecules multiple scattering is less significant,
thus the bubble contributions can be identified (Zhang et al., 2002).
Accordingly, in this study, the effects of oceanic bubbles on atmospheric
aerosol retrievals are studied at the Moderate Resolution
Imaging Spectroradiometer (MODIS) 0.66
In this study, winds derived from Advanced Microwave Scanning Radiometer EOS (AMSR-E), ship-based AOD data from Maritime Aerosol Network (MAN), and MODIS radiances and AOD retrievals were collected and collocated. A 6-year 2004–2009 study period is used. Seven years (2002–2008) of collocated AErosol RObotic NETwork (AERONET) and Aqua MODIS DT AOD data are used to aid the analysis. Two radiative transfer models, the 6S atmospheric RTM and HydroLight oceanic RTM, are also applied for studying the effects of oceanic bubbles on aerosol retrievals. The ground-based and satellite observations, as well as both RTMs, are discussed in detail in this section.
The MAN data, derived from ship-borne measurements of direct solar
attenuation by aerosol scattering and absorption over oceans, include
retrieved AOD at five wavelengths ranging from 0.34 to 1.02
MODIS is on board both the Terra (passes over the equator at 10.30 a.m.
local standard time) and Aqua (equator overpass at 1.30 p.m. local standard
time) platforms. The MODIS instrument measures TOA radiation at 36 spectral
channels, with spatial resolutions ranging between 250 and 1000 m at
nadir and a wide swath of 2330 km. For this study, the Collection 5 (C5)
Aqua MODIS dark target (DT) aerosol products are used (Remer et al., 2005).
Note that the Collection 6 (C6) MODIS DT products were released recently,
and this product includes a whitecap and ocean-foam effect in the radiative
transfer lookup tables (LUTs) used to estimate aerosol properties (Levy et al.,
2013). No accounting is made for subsurface ocean bubbles in any existing
MODIS product. For this study, the C5 MODIS DT products are chosen to be
consistent with the analysis done in Toth et al. (2013). Validated against
ground-based observations, Remer et al. (2005) suggests that the uncertainty
in over-ocean AODs is on the order of
In addition to the Aqua MODIS DT aerosol products, the Collection 5 Level 1b Aqua MODIS radiance data are also collocated with wind speed data from AMSR-E and AOD data from MAN for evaluating the RTM simulations. MODIS channel 1 (0.66 micron) TOA radiance (at a 1 km resolution), along with latitude, longitude, and viewing geometry data, is extracted from the Level 1b MODIS data. The Collection 5 MODIS cloud mask data (MYD35) are also used for excluding cloud-contaminated pixels.
The near-surface wind speed values are obtained from the collocated version
7 AMSR-E data (Wentz and Meissner, 2000). On board the Aqua satellite, the
AMSR-E is a conically scanning passive microwave radiometer with sensors in
12 microwave channels. The spatial resolutions of the AMSR-E data range from
5 to 56 km depending on the frequency (e.g., Wentz and Meissner, 2000).
Retrieved surface parameters from AMSR-E include precipitation, sea surface
temperatures, water vapor, wind speed, and other ancillary data. The AMSR-E
data used in this study are formatted as hourly gridded binary files with a
spatial resolution of 0.25
As the first step, MAN data (2004–2009) are spatio-temporally collocated
with the MODIS DT aerosol products. The temporal and spatial thresholds are
set to
Lastly, 7 years (2002–2008) of Aqua MODIS DT and Level 2 quality-assured
AERONET data (Smirnov et al., 2000) are used to study the differences in Aqua
MODIS DT and AERONET AOD (
The 6S RTM is used to simulate TOA radiation measured by MODIS at the
0.66
In the 6S model, the ocean surface reflectance is computed based on Koepke (1984)
by considering whitecap, glint, and water-leaving reflectance as
shown in Eq. (1).
HydroLight is designed to simulate radiative transfer processes in oceans (Mobley et al., 2012). In order to estimate surface reflectance changes from ocean bubbles, the parameters of viewing geometry, wind speed, the ocean bubble phase function, and the ocean bubble concentration are needed for the HydroLight model runs.
The ocean bubble phase function (or the general shape of angular scattering)
is adopted from Zhang et al. (2002), in which the bubble phase functions are
computed from coated spheres based on Mie scattering theory and are
inter-compared with laboratory (using a volume scattering meter) and field
observations. The coating represents surfactant material that adheres onto
the bubble surface almost instantaneously after bubble genesis occurs in
nature. Including the coating in optical computation is critical for remote
sensing applications, because it can increase backscattered light by bubbles
by a factor of up to 4 (Zhang et al., 1998). Little is known about the
composition and thickness of the bubble coatings. However, a recent study
shows that a coating of proteinaceous type (D'Arrigo, 1983; D'Arrigo et al.,
1984) with a refractive index of 1.18 relative to water and a thickness of
0.01
Bubbles are frequently formed by breaking waves (Thorpe and Humphries, 1980; Lamarre and Melville, 1991). Because of the rapid rising of bubbles, the density distribution of ocean bubbles decreases exponentially with depth, while the overall concentrations increase with increasing wind speed following a power law. Ocean bubble concentrations in this experiment are obtained from Zhang (2001) and Zhang and Lewis (2002), in which the bubble concentrations at different layers are modeled as a function of wind speed based on field observations.
The bubble-induced perturbations in surface reflectance are estimated at the
MODIS 0.66
The reflectance difference (
Reflectance of whitecaps and ocean bubbles (for all three bubble concentrations) as a function of wind speed (estimated from 6S).
For further testing, an experiment is done using upper and lower boundaries
of the default bubble concentrations (default bubble) that are estimated
from Zhang (2001). The upper boundary is made by doubling the default bubble
concentrations (double bubble), while the lower boundary is represented by
half of the default concentrations (half bubble). The use of upper and lower
boundaries allows analysis of two extreme conditions when compared to a
normal set of conditions. Following the steps as mentioned above, the
To compare the relative contributions from whitecaps and submerged bubbles,
Fig. 3 shows the magnitude of whitecap reflectance with that of ocean
bubbles for all three bubble concentrations, based on the 6S simulations.
Clearly, the reflectance from submerged bubbles is insignificant for wind
speed of less than 10 m s
As the first step, the modeled MODIS channel 1 radiances from the linked 6S and HydroLight models are inter-compared with MODIS observations. The MAN–DT–AMSR-E data set is constructed to build an observation-based group of data that includes AOD from MAN, wind speed from AMSR-E, and satellite-measured radiance from MODIS. The MODIS cloud mask data are used to minimize cloud contamination, and only confidently clear pixels are chosen for further analysis.
Figure 4 shows a comparison between the 6S simulated radiance values at the
MODIS 0.66
As the next step, an approach based on a LUT is adopted to
estimate the impact of bubbles on AOD retrievals. Using the linked 6S and
HydroLight model, LUTs of simulated MODIS TOA reflectance values are
constructed as functions of solar zenith angle and viewing zenith angle (SZA
and VZA; each varied from 0 to 60
A comparison of 6S–HydroLight modeled radiance versus Aqua MODIS
channel 1 radiance (0.66
For any given observing condition, the simulated reflectance values (both with and without bubble cases) are used to compute the errors in the
retrieved AOD without considering bubbles. For example, for a given bubble
concentration with a given wind speed and fixed viewing geometry, a
relationship between AOD (AOD
To illustrate the concept, Fig. 5 shows the averaged
Averaged
Also, MODIS DT AOD retrievals with cloud
fraction larger than 80 % (e.g., Shi et al., 2011a) and bad retrievals as
identified by the quality assurance (QA) flags (QA
A plot of the difference in MAN and MODIS DT AOD (AOD
Based on Fig. 5, from a theoretical analysis perspective, at low wind speeds
there is no significant impact from ocean bubbles. However, as wind speeds
increase, so does
Using the MAN–DT–AMSR-E data set, the impact of bubbles on the difference
between MODIS DT and MAN AOD (
It has been shown that, theoretically, ocean bubbles can affect satellite
aerosol retrievals in red wavelengths under conditions with near-surface
wind speeds greater than 12 m s
Yearly mean MODIS Collection 6 AOD paired with the AOD correction for the default bubble concentration.
Figure 7a is a map of the average AMSR-E global wind speed for the year 2009
at a resolution of 1
Figure 7d and e are similar to Fig. 7c but show results for the double (Fig. 7d) and half (Fig. 7e) bubble cases. The double bubble concentration has the largest impact on AOD, yet as evident in Fig. 7d, the majority of areas still experience less than a 10 % change in annual mean AOD. Therefore, these results indicate that ocean bubbles do not have a major impact on the ESOA. This is likely because the average wind speed throughout the ESOA region is not high enough to sustain a large contribution from ocean bubbles, as evident from Fig. 7a.
Cumulative distribution function (CDF) of wind speed frequency over the ESOA region for 2009 using AMSR-E data from the ascending orbits.
Recently, the C6 Aqua MODIS DT data were released. For comparison purposes,
a portion of Fig. 7 is regenerated using the C6 MODIS DT data. Figure 8a
shows the yearly mean C6 MODIS DT AOD in the ESOA region and Fig. 8b shows
the
The long-term means, however, may not represent individual cases. Figure 9
shows the cumulative distribution of instantaneous wind speeds in the ESOA
region using AMSR-E data from the ascending orbits for 2009. Similar results
are found using the data from the AMSR-E descending orbits and thus are
not shown. Although the median wind speed of the ESOA region is under 10 m s
To complement previous results, variations in the seasonal wind speed over
the region as shown in Fig. 7 are studied as shown in Table 2 for the year
2009. The annual mean wind speed is 9.50 m s
Annual and seasonal mean wind speeds and standard deviations for the year 2009 in the ESOA region.
In this study, the effects of ocean bubbles on satellite
aerosol measurements are studied through a theoretical approach using a linked HydroLight oceanic and 6S atmospheric
radiative transfer model.
LUTs of the bubble-induced uncertainties in oceanic aerosol
optical depth values retrieved from passive sensors (
It is evident that at low wind speeds there is no significant impact from ocean
bubbles on AOD retrievals using passive-based remote sensing techniques. However, the
impact becomes much more significant at wind speeds above 12 m s The impacts of oceanic bubbles on the ESOA phenomenon are evaluated using
1
year of MODIS Collection 5 data,
Advanced Microwave Scanning Radiometer Earth Observing System (AMSR-E) data, and
lookup tables of To avoid multiple scattering, the effect of subsurface bubbles on AOD retrievals
is only evaluated at the 0.66 Recently, the Collection 6 Aqua MODIS DT aerosol products have been released.
New changes to the C6 MODIS DT aerosol products include a modified cirrus cloud detection
scheme as well as the dependency of ocean surface reflectance as a function of wind speed.
As a result, the ESOA feature is much reduced (Levy et al., 2013). Still, the submerged bubbles
are not considered, and thus most of the discussions in this paper are valid for the C6 MODIS DT
aerosol products. There are several limitations in this study. Only theoretical calculations are included in
the study for simulating the effects of bubbles on aerosol retrievals. The spatial and temporal
variations of submerged bubbles and their optical and physical properties are not considered;
this variation may significantly perturb the results from this study. It is likely that oceanic
bubble contributions to the ocean surface reflectance are partially accounted for in empirical
approaches based on direct estimation of overall wind speed impact (Shi et al., 2011a). Also, this is
only a theoretical analysis. For practical applications, the uncertainties in whitecap estimates
(fractional coverage and spectral reflectance, e.g., Frouin et al., 1996; Anguelova et al., 2006) need
to be fully considered and incorporated into the analysis. The theoretical derivations used to estimate
wind speed effects on surface reflectance for MODIS and MISR satellite aerosol retrievals explicitly
include whitecaps and bubble rafts but not subsurface bubbles (Levy et al., 2013; Limbacher and Kahn, 2014).
For future applications that require accurate estimations of atmospheric aerosol concentrations from satellite
observations, oceanic bubble concentration is a factor that needs to be taken into consideration for ocean
regions with strong near-surface winds.
Jianglong Zhang and Jeffrey Reid acknowledge the support from the Office of Naval Research Codes 322 (N00014-10-0816 and N0001414AF00002). Matthew Christensen and Jianglong Zhang acknowledge the support of the NASA project (NNX14AJ13G). Xiaodong Zhang acknowledges the support of a NASA EPSCoR grant NNX13AB20A as well as NSF IIA-1355466. MODIS data were obtained from the Level 1 and Atmosphere Archive and Distribution System (LAADS). We also thank individual PIs from the AERONET sites for the sunphotometer data. We also thank Andrew Sayer and another reviewer for their constructive comments/suggestions. Edited by: A. Kokhanovsky