CalNex cloud properties retrieved from a ship-based spectrometer and comparisons with satellite and aircraft retrieved cloud properties

Authors

  • P. J. McBride,

    Corresponding author
    1. Laboratory for Atmospheric and Space Physics, University of Colorado Boulder, Boulder, Colorado, USA
    2. Department of Atmospheric and Oceanic Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    • Corresponding author: P. J. McBride, Laboratory for Atmospheric and Space Physics, University of Colorado Boulder, 392 Campus Box, Boulder, CO 80309, USA. (mcbridep@colorado.edu)

    Search for more papers by this author
  • K. S. Schmidt,

    1. Laboratory for Atmospheric and Space Physics, University of Colorado Boulder, Boulder, Colorado, USA
    Search for more papers by this author
  • P. Pilewskie,

    1. Laboratory for Atmospheric and Space Physics, University of Colorado Boulder, Boulder, Colorado, USA
    2. Department of Atmospheric and Oceanic Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    Search for more papers by this author
  • A. Walther,

    1. Office of Research and Applications, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, Madison, Wisconsin, USA
    Search for more papers by this author
  • A. K. Heidinger,

    1. Office of Research and Applications, National Environmental Satellite, Data, and Information Service, National Oceanic and Atmospheric Administration, Madison, Wisconsin, USA
    Search for more papers by this author
  • D. E. Wolfe,

    1. Physical Sciences Division, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA
    Search for more papers by this author
  • C. W. Fairall,

    1. Physical Sciences Division, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA
    Search for more papers by this author
  • S. Lance

    1. Chemical Sciences Division, National Oceanic and Atmospheric Administration, Boulder, Colorado, USA
    2. Cooperative Institute for Research in Environmental Sciences, University of Colorado Boulder, Boulder, Colorado, USA
    Search for more papers by this author

Abstract

[1] An algorithm to retrieve cloud optical thickness and effective radius (reff) from spectral transmittance was applied to radiance and irradiance observations of the Solar Spectral Flux Radiometer (SSFR) during the Research at the Nexus of Air Quality and Climate Change Campaign (CalNex). Data from an overcast day, 16 May 2010, was used to validate the algorithm. Retrievals from the SSFR, deployed on the Woods Hole Oceanic Institute R/V Atlantis, were compared to retrievals made from an airborne SSFR, the Geostationary Operations Environmental Satellite (GOES), an Atlantis-based microwave radiometer, and the Moderate Resolution Imaging Spectroradiometer. In situ observations of reffduring a flight over the Atlantis were compared to the Atlantis SSFR and GOES retrievals. The cloud statistics for the CalNex campaign were compared to previous studies. The agreement between the different retrievals, quantified by determining the number of coincident observations when retrieval uncertainty overlapped, improved as the difference between the field-of-views (FOV) of the instruments decreased. It is shown that averaging the 1 Hz SSFR observations to the 15 minute GOES interval cannot fully account for the impact of the different FOVs. The average in situ reff (7.7 μm) fell between the average reffretrieved using the Atlantis-based SSFR radiance (5.7μm) and irradiance (9.5 μm). The CalNex clouds showed a diurnal pattern observed in previous studies of marine boundary layer clouds in the region. The distribution of cloud optical thickness and liquid water path during CalNex was shown to be a gamma distribution, consistent with previous studies of high cloud fraction marine boundary layer clouds.

1. Introduction

[2] Marine boundary layer clouds play an important role in the climate system. They exert a negative top-of-atmosphere radiative forcing due to their high albedo in contrast to the low albedo of the underlying ocean.Randall et al. [1984] have suggested that a 4% increase in areal coverage of marine boundary layer clouds could offset the warming due to a doubling of CO2. Cloud property statistics of low level marine clouds have been derived from satellite observations, such as the Clouds and the Earth's Radiant Energy System (CERES) [Wielicki et al., 1996]. From the surface, statistics of cloud type and cloud coverage [Warren et al., 2007] have been compiled, but not the properties needed to quantify the radiative impact of clouds. Retrievals of cloud properties can differ depending on the viewing geometry. Platnick [2000] show that the vertical cloud layers are more uniformly sampled in cloud transmittance than in cloud reflectance which results in two distinct views of the cloud properties. Observations of cloud properties from the surface are needed to provide a more complete picture of clouds and to provide a means for validating and understanding satellite retrievals.

[3] The cloud properties that are required to describe the radiative impacts of clouds are the cloud optical thickness (τ), the column integrated extinction, and the cloud effective particle radius (reff). The effective radius is the scattering weighted average cloud particle radius, which reduces to a ratio of the third and second moments of the cloud particle size distribution for cloud particles illuminated by visible and near infrared radiation:

display math

where n(r) is the cloud particle size distribution. McBride et al. [2011]presented an algorithm to retrieve the cloud properties using hyperspectral transmittance observations made from a single spectrometer. The algorithm uses a combination of the spectral shape in the near-infrared and the transmittance in the visible, which was shown to increase the sensitivity to effective radius when compared to more traditional algorithms developed for use with reflectance [Nakajima and King, 1990; Platnick et al., 2003; Coddington et al., 2010] and applied to transmittance. This allows for the simultaneous retrieval of cloud properties from the surface using observations from the Solar Spectral Flux Radiometer (SSFR) [Pilewskie et al., 2003].

[4] An SSFR was deployed on the Woods Hole Oceanographic Institution (WHOI) R/V Atlantis deployed in the Pacific Ocean off the coast of California during the Research at the Nexus of Air Quality and Climate Change Campaign (CalNex). The McBride et al. [2011]algorithm was applied to these observations during a 25-day deployment aboard the Atlantis that began on 14 May 2010. In this paper we will use data from a single overcast day during CalNex, 16 May 2010, to validate the transmittance algorithm. This day featured overcast skies and coordinated cloud property observations made from the ship, from an aircraft, and from satellite. By comparing the retrievals from these different platforms, we validate the cloud transmittance-based retrieval developed inMcBride et al. [2011]. Using the CalNex-wide retrievals we provide the cloud property statistics for the CalNex campaign. These statistics are compared with previous distributions of cloud properties of marine boundary layer clouds in the same region.

2. Experiment and Data Collection

2.1. CalNex

[5] The CalNex campaign was an air quality and climate change study conducted over inland southern California and the California coastal regions of the Pacific Ocean during May and June of 2010 (http://www.esrl.noaa.gov/csd/calnex/). The main focus of this multiagency (National Oceanic and Atmospheric Administration, the California Air Resources Board, and the California Energy Commission) experiment was to better understand issues related to the interactions between air quality and climate change. The experiment used instrumentation deployed at several ground sites, on a ship, the WHOI R/V Atlantis, and on three research aircraft, the NOAA WP-3D, the NOAA Twin Otter, and the National Aeronautics and Space Administration (NASA) DC-8. The platforms of interest to this work are the Geostationary Operational Environmental Satellites (GOES-11) satellite, the Moderate Resolution Imaging Spectroradiometer (MODIS), the NOAA WP-3D (“P3” for the rest of the paper), and the WHOI R/V Atlantis (“Atlantis” for the rest of the paper).

2.2. Solar Spectral Flux Radiometer

[6] One SSFR was deployed on the Atlantis and a second on the P3. Each SSFR consisted of two light collectors, each connected to two spectrometers. The first spectrometer covered the spectral range from 350 nm to 1000 nm with a spectral resolution of 8 nm full-width half maximum (FWHM), and the second covered the spectral range from 950 to 1700 nm for the Atlantis and 950 nm to 2100 nm for the P3 configurations, both with spectral resolutions of 12 nm FWHM. The data were collected at a frequency of 1 Hz. The P3 SSFR consisted of one light collector configured to observe zenith and nadir irradiance, both with a field of view (FOV) of 2π steradians. The Atlantis SSFR was configured to observe zenith irradiance and zenith radiance using a collimator that limited the FOV to 2.4°. The light collectors on the Atlantis were mounted on a stabilized platform to correct for the motions of the ship. Details are summarized in Table 1. The measured variables in the leftmost column of Table 1 will be used throughout the rest of this document to refer to the different SSFR observations used in retrieving cloud properties.

Table 1. Summary Viewing Geometry and Spectral Range of the SSFR Configurations Deployed During CalNex
MeasurementsSpectral Range (nm)Platform
Ship SSFR radiance (FOV = 2.4°)350–1700R/V Atlantis
Ship SSFR irradiance350–1700R/V Atlantis
Airborne SSFR zenith irradiance350–2100NOAA P3
Airborne SSFR nadir irradiance350–2100NOAA P3

2.3. MWR

[7] The Atlantis was equipped with a microwave radiometer (MWR) that was used to retrieve column integrated liquid water and water vapor. The retrieval used the microwave emissions at 23.8 GHz and 31.4 GHz, the former being more sensitive to water vapor and the latter more sensitive to liquid water. The data was acquired in 1 s windows every 16 s, with a field of view of 5.9° for the 23.8 GHz channel and 4.5° for the 31.4 GHz channel. The observations from the two channels were used in a simultaneous retrieval of column integrated liquid water and water vapor. More details regarding the microwave retrieval can be found in Westwater et al. [2001] or Zuidema et al. [2005].

2.4. MODIS

[8] MODIS is deployed on both the Aqua and Terra satellites. It measures radiance at 36 discrete wavelength bands in the visible through the thermal infrared. One application of these observations is the retrieval of cloud properties. The cloud properties are provided at a resolution of 1 km and are included in the MOD06 and MYD06 MODIS products (Terra and Aqua, respectively). The retrieval algorithm [Nakajima and King, 1990; Platnick et al., 2003] typically uses one measurement of cloud reflectance in a non-absorbing spectral region and one in an absorbing region with respect to liquid water. The former is more sensitive to optical thickness and the latter is more sensitive to effective radius. The center wavelength of the channels used for the standard retrievals over ocean are at 860 nm and 2130 nm [Platnick et al., 2003], though retrievals with the 1.6 μm and 3.7 μm channels are also performed.

2.5. GOES-11

[9] GOES-11 is a geosynchronous satellite located over the Western United States and Eastern Pacific Ocean. The imager onboard GOES-11 consists of five channels. The center wavelengths and spatial resolutions are shown inTable 2. The reflected radiances and corresponding cloud properties were reported every 15 min. The Daytime Cloud Optical Microphysical Properties (DCOMP) retrieval algorithm [Walther and Heidinger, 2012] was developed with the goal of providing a uniform cloud retrieval algorithm across several satellite-borne sensors, of which the GOES series is included. It was applied to the GOES observations by the NOAA National Environmental Satellite, Data, and Information Service (NESDIS), and it follows the standard two channel cloud property retrieval algorithm for cloud reflectance, as described inSection 2.4 for MODIS though with a single 3.9 μm water absorption channel. The reflected radiances were calculated with a forward model for the GOES-11 channels 1 and 2 and the observed radiances were used to find the best fit optical thickness and effective radius. The DCOMP algorithm uses optimal estimation to find the best fit, which provides the ability to incorporate the uncertainties of the observations as well as inputs to the forward model. The cloud property retrievals were reported every 15 min over the 4 km FOV of channel 2.

Table 2. The Five Channels of the GOES-11 Imager Along With the Instantaneous Field-of-View and the Midpoint of the Spectral Range
 Channel
12345
Center Wavelength (μm)0.653.96.710.712
Instantaneous Geographic Field of View (km)14844

2.6. In Situ Cloud Probe Measurements

[10] Two cloud probes were deployed on the P3, the cloud droplet probe (CDP) and the cloud imaging probe (CIP), as described in Lance [2012]. One product derived from the cloud probes is the cloud droplet effective radius vertical profile. The effective radius is obtained by measuring the amount of light scattered as the cloud droplets pass through a laser beam emitted by the probes. The CDP is able to detect particles with an effective radius between 2 μm and 50 μm and the CIP observed larger particles between 50 μm and 1550 μm.

2.7. Reflectance Retrieval Algorithm

[11] On 16 May 2010, the P3 flew level legs above and below a cloud deck between 200 m and 700 m above sea level, providing the opportunity to retrieve cloud properties from P3 SSFR measurements of reflectance and transmittance. For reflectance observations made from the airborne SSFR nadir instrument, the cloud retrieval algorithm followed previous algorithms for MODIS [Nakajima and King, 1990] and for airborne spectrometers [Coddington et al., 2010; Kindel et al., 2010] as outlined in Section 2.4. Irradiance at 515 nm was used as the non-absorbing channel and irradiance at 1634 nm was used as the absorbing channel. The estimation of the uncertainty followed the same method as inMcBride et al. [2011], where the 3% radiometric uncertainty was propagated through the retrieval algorithm, resulting in a range of retrieved optical thickness and effective radius. The retrieval uncertainty of the optical thickness and effective radius were estimated by the average of the minimum and maximum of the retrieved range.

2.8. Transmittance Retrieval Algorithm

[12] The algorithm presented in McBride et al. [2011]retrieves cloud optical thickness and effective radius by fitting the observed transmittance at 515 nm and spectral slope between 1565 nm to 1634 nm to a pre-calculated lookup table. The optical thickness and effective radius used to calculate the best fit slope and transmittance in the lookup table, found using a least-squares fit, were taken as the retrieved values. The lookup table was created with effective radii ranging from 1μm to 30 μm in steps of 1 μm and interpolated to 0.1 μm. The optical thickness in the lookup table ranged from 0.1 to 0.9 in steps of 0.1 and from 1 to 100 in steps of 1 and interpolated to steps of 0.1. The cosine of the solar zenith angle (μ) was varied between 0.1 and 1. The forward model was run assuming that the cloud particle phase was liquid. The ocean surface albedo spectrum was obtained from SSFR measurements during the International Consortium for Atmospheric Research on Transport and Transformation (ICARTT) experiment [Coddington et al., 2010, 2008]. The albedo had a value of 0.036 at 515 nm and 0.019 at 1600 nm and was used in the forward model for cloud retrievals made from radiance and irradiance observations. The gaseous concentrations of the model atmosphere were taken from the standard midlatitude summer atmosphere [Anderson et al., 1986]. Similar to the reflectance retrievals from the airborne SSFR, the uncertainty was estimated by propagating the radiometric uncertainty and taking the average of the minimum and maximum retrieved values of optical thickness and effective radius.

[13] In previous applications of the algorithm, cases known to meet the assumptions of cloud uniformity and liquid water phase were selected. To apply the algorithm across the entire CalNex data set, some additional checks on the quality of the retrievals were necessary to limit the effects of deviations from these assumptions. To be considered valid, the retrievals had to be within limits applied to: (1) the observed transmittance at 515 nm and the near infrared slope, (2) the retrieved optical thickness and effective radius, and (3) the retrieved effective radius uncertainty. Retrievals that resulted in best-fit solutions where the difference between the modeled and observed transmittance at 515 nm is larger than the SSFR radiometric uncertainty of 3% were excluded. Observations with near infrared slopes outside the limits of the lookup table were also excluded. Only retrievals resulting in optical thickness greater than 5 and with effective radius greater than 4μm were considered valid. The 4 μm limit was chosen by considering the work of Miles et al. [2000], who compiled a number of in situ effective radius observations and found the lowest effective radius over ocean to be 4.2 μm. The optical thickness limit of 5 was chosen because of the lack of sensitivity in the transmittance to the effective radius for thin cloud. Using a limit established in McBride et al. [2011], retrievals with an absolute effective radius uncertainty larger than 2 μm were also excluded. In addition to these limits on the retrievals two days were excluded entirely. On 17 May 2010, GOES phase retrievals (not shown) indicated the presence of ice throughout the day so this data was excluded due to the assumption of liquid water particles. Data from 24 May 2010 was excluded due to a failure in the stabilized platform, so the geometry of the SSFR light collectors could not be accounted for.

[14] McBride et al. [2011]demonstrated the utility of applying the algorithm to transmitted radiance observations. For CalNex, the algorithm was also applied to the transmitted irradiance. The irradiance- and the radiance-based retrievals represent two limiting cases: The radiance-based retrievals are applied to observations over a narrow field-of-view (as often the case in remote sensing applications), whereas the irradiance-based retrievals are applied to observations over the entire hemisphere, and pertain more to the energy-budget radiative properties. One difference that results is the optical thickness dependence. Irradiance decreases with increasing optical thickness and is at a maximum at an optical thickness of 0. Radiance increases with increased optical thickness, due to an increase in scattering, up to an optical thickness of about 5 and then decreases with increasing optical thickness due to attenuation. The radiance is at its maximum near an optical thickness of about 5.

[15] Figures 1a and 1b show the modeled spectral slope at 1600 nm and transmittance at 515 nm calculated using radiance and irradiance, respectively, and the difference between radiance and irradiance for optical thickness less than 5 can be seen clearly in this figure. The utility of this algorithm to separating lines of constant size and lines of constant optical thickness indicates that it may be possible to retrieve cloud optical thickness and effective radius with transmitted irradiance. In order to quantify the ability to retrieve the cloud properties, cloud retrievals were simulated with modeled irradiance data. The retrievals were simulated by applying the algorithm to modeled spectra for optical thicknesses of 5, 10, 20, 40, and 80and an effective radius of 10 μm. Table 3 lists the results of these simulated cloud retrievals. The two rightmost columns show the resulting retrieval uncertainties estimated by propagating the radiometric uncertainty through the retrieval algorithm. The effective radius retrieval uncertainty is less than the 2 μm uncertainty threshold that was established in McBride et al. [2011]. For these cloud properties, the 8% optical thickness uncertainty, resulting from only the radiometric uncertainty, is the largest for an optical thickness of 5 and decreases for thicker cloud scenes. This demonstrates that the algorithm can produce reliable results when applied to transmitted irradiance and it was applied to the transmitted irradiance observations from the ship SSFR and the airborne SSFR.

Figure 1.

Modeled transmittance and spectral slope at 1600 nm computed for (a) radiance and (b) irradiance. The slope and transmittance are shown for values of constant cloud effective particle radius (solid) and constant optical thickness (dashed).

Table 3. A Modeled Transmittance Irradiance Spectrum Calculated With the Values of Optical Thickness and Effective Radiusa
Optical ThicknessEffective Radius (μm)Optical Thickness Uncertainty (%)Relative Effective Radius Uncertainty (%)Absolute Effective Radius Uncertainty (μm)
  • a

    The modeled spectrum was input to the retrieval algorithm and the uncertainties of the derived optical thickness and effective radius (third through fifth columns) were computed by propagating the 3% radiometric uncertainty.

510811.51.15
10105101.0
20103.590.9
40102.69.50.95
80102.210.51.05

3. Retrieval Comparisons

[16] To validate the retrieval algorithm beyond the data presented in McBride et al. [2011], the cloud retrieval results from the ship SSFR radiance and irradiance observations were compared to cloud property retrievals from satellite, the airborne SSFR and cloud probes, and the ship-based microwave radiometer. On 16 May 2010 the NOAA P3 flew over the Atlantis several times at different altitudes during a coordinated flight.Figure 2 maps the area off the coast of Los Angeles, CA where the coordination took place. It includes the true color MODIS image of the cloud field taken at 19:15 UTC along with the tracks of the P3 and Atlantis during the coordinated flight which occurred between 18:26 UTC and 20:54 UTC.

Figure 2.

A map off the coast of Los Angeles where the P3 (red) flew over the Atlantis (blue) on 16 May 2010. (MODIS true color image from NASA/GSFC, Rapid Response.)

3.1. Optical Thickness Comparisons

[17] Figure 3a shows a time series of the retrieved optical thickness from the ship SSFR radiance and irradiance, the airborne SSFR, and GOES for the 16 May 2010 case. The 1 Hz observations of the ship SSFR were matched to the closest GOES retrieval in space and nearest time. The ship SSFR radiance and irradiance retrievals exhibit greater variability due to the higher spatial and temporal resolution for SSFR than for GOES. For example, just after 22 UTC, the retrievals made from the ship SSFR radiance and irradiance show that the optical thickness doubles, while the corresponding GOES retrieved optical thickness remains more or less constant. The linear footprint of the ship SSFR observations was calculated as the diameter of the circle at cloud base that encompasses half of the radiant energy reaching the SSFR light collector. For a cloud base of 200 m, a value typical during CalNex, the linear footprint of the irradiance and radiance observations were 400 m and 10 m, respectively. The linear footprint of a GOES pixel was 4000 m. By quantifying the agreement between the retrieval algorithms as the percentage of observations that agreed to within their respective uncertainties (i.e. the error bars overlap), the agreement can be shown to increase as the difference in linear footprint decreases. The ship SSFR radiance, agreed with GOES 22% of the time. The ship SSFR irradiance agreed with GOES 30% of the time, while SSFR radiance and irradiance agreed 55% of the time. The mean optical thickness from the three retrieval algorithms were 30.7, 29.1, and 30.6 for the ship SSFR radiance, ship SSFR irradiance, and GOES, respectively. Four airborne retrievals over the Atlantis are included in Figure 3. Two of the four occurred during time periods that return a valid retrieval from the ship SSFR, one above cloud at 18:59 UTC and one below cloud at 21:06 UTC. The 18:59 UTC case was within the uncertainty of the optical thickness retrievals made with the ship SSFR radiance and irradiance observations and the 21:06 case was within the uncertainty of all three. One MODIS overpass is shown near 18:30 UTC. The retrieved optical thickness is lower than the SSFR retrieved values and the GOES retrieved value.

Figure 3.

A time series plot of (a) optical thickness and (b) effective radius retrievals for 16 May 2010 from the ship SSFR radiance (blue), ship SSFR irradiance (green), GOES NESDIS (red), MODIS (pink square), airborne SSFR reflectance (up-pointing triangle), and the AIRBORNE SSFR transmittance (down-pointing triangle).

[18] In the previous analysis the SSFR retrievals were matched to the closest GOES retrieval in space and closest in time. Another way to compare these data is to use the GOES retrievals at the native 15 minute interval and to average the SSFR retrievals that occur within ±7.5 min. Figure 4 shows the GOES retrieved optical thickness with the averaged optical thickness retrieved from the ship SSFR radiance and irradiance retrievals. The GOES retrievals were, on average, within 15.1% and 18.6% of the averaged optical thickness retrievals of the ship irradiance and radiance, respectively. By comparison the GOES optical thickness retrievals were within 17.9% and 21.5% of the ship irradiance and radiance retrieved values when matched in space and time, showing that averaging the SSFR observations, in part, helps in compensating for the spatial and temporal differences in the observations.

Figure 4.

GOES optical thickness plotted versus the mean ship SSFR optical thickness, averaged over the 15 minute reporting time of GOES, for the 16 May 2010 case. The line fit with a linear regression is shown as a red dashed line along with the fit slope and y-intercept in the legend. The gray line shows the one-to-one line.

3.2. Effective Radius Comparisons

[19] The effective radius retrieved from GOES and from the ship SSFR irradiance is higher than that of the ship SSFR radiance observations throughout the day. The ship SSFR irradiance effective radius retrievals were within the uncertainty of the GOES effective radius retrievals 26% of the day. The ship SSFR radiance effective radius retrievals overlap for short periods of time with the GOES, 1% of the day, and ship SSFR irradiance retrievals, 6% of the day. The airborne SSFR nadir and zenith retrievals were at the upper limit of the ship SSFR radiance retrievals and the lower limit of GOES and the ship SSFR irradiance retrievals. The 18:59 UTC P3 overpass retrieved effective radius uncertainty does not overlap with any of the others and it is higher than the ship SSFR radiance effective radius and below the ship SSFR irradiance and GOES retrievals. The 21:06 UTC case was within the uncertainty of the other three retrievals. The MODIS retrieved effective radius is within the uncertainty of the ship SSFR radiance retrieved effective radius, but below that of the ship SSFR irradiance and GOES retrievals. The next section will explore the GOES and SSFR effective radius retrievals further with a comparison to in situ data.

3.3. In Situ Effective Radius Comparison

[20] During the Atlantis-P3 coordination, the P3 spent approximately 90 min flying above, below and within cloud, and passed multiple times over the Atlantis. To describe the in-cloud data set, the average vertical profile of cloud droplet size distributions and effective radius were calculated from CDP and CIP measurements, and the result is shown inFigure 5b. The in-cloud observations were obtained for two level legs just above cloud base (205–244 m above mean sea level (ASL)) over a total of 26 min, and one level leg just below cloud top (417–472 m ASL) over 15 min. In addition to the level legs the P3 performed six passes through the cloud deck (from above to below cloud, or vice versa) in the vicinity of the Atlantis. Since the cloud microphysical data in the vicinity of the ship was somewhat limited, a statistical comparison was made between the average in situ profile and the remotely retrieved values of effective radius recorded during the coordinated sampling period. Excluded from the in situ average profile inFigure 5b are the observations of drizzle drops that were encountered up to 100 μm in size at a droplet concentration (dN/dlogDp) of 10−3 cm−3. Inclusion of these drizzle drops would increase the in situ derived effective radius by 11% in the lower 60 m of the cloud and mid-cloud at 420–510 m elevation.Figure 5ashows the normalized histograms of the effective radius retrieved from GOES and the ship SSFR radiance and irradiance observations during the P3-Atlantis coordination. The in situ observations resulted in an average effective radius of 7.7μm, averaged over the coordinated time period and over the profile. During this same time period the average effective radius retrieved from the ship SSFR radiance observations was 5.7 μm. The average effective radius observed by GOES and the ship SSFR irradiance observations were 11.5 μm and 9.5 μm, respectively. The larger effective radius retrieved by the ship SSFR irradiance retrievals and GOES retrievals could be the result of cloud heterogeneity. Figure 6 shows the horizontal heterogeneity of the optical thickness for the 90 minute period the Atlantis and P3 were coordinated. The optical thickness was derived from the ship SSFR radiance observations which has the finest spatial resolution of the remote sensing instruments used here. Discrepancies between remotely sensed effective radius from cloud reflectance and in situ observations have been observed in the past [Twomey and Cocks, 1989; Rawlins and Foot, 1990; Nakajima et al., 1991]. Feingold et al. [2006]showed the challenges of comparing cloud observations with different sampling volumes and averaging times. They compared retrievals from surface-, aircraft-, and satellite-based radiometers and in situ cloud probes by considering the sampling volumes and observation of each instrument. The appropriate weighting functions were applied to effective radii retrieved from above, below, and from within the cloud to calculate a combined, best-estimate of the retrieved effective radii. The effective radius was shown inPlatnick [2000] to be more uniformly distributed vertically for transmittance observations than for reflectance. A simpler approach to account for the different sampling volume and observation times was used here, averaging the finer temporal observations of the SSFR, which only showed a minor improvement in the comparison of the ship SSFR and GOES retrievals. Averaging over the 15 minute interval can account for the temporal differences in the observations, but the fact that this did not completely account of the differences between the GOES and SSFR radiance derived optical thickness would suggest that the SSFR observations did not sample enough of the GOES pixel spatially.

Figure 5.

(a) The normalized histogram of the cloud effective particle radius retrieved from the ship SSFR radiance (blue), the ship SSFR irradiance (green), and GOES (red) for 16 May 2010 between 18:25:54 UTC and 20:54:14 UTC. The bin width is 1.3 μm, which is twice the average ship SSFR radiance retrieved effective radius uncertainty. (b) The average effective radius vertical profile (thick black line), the average number of in situ observations per altitude (thin black line), and the droplet number concentration derived from in situ observations for the same time period. The average effective radius profile was calculated by retrieving effective radius from the 1 Hz particle size distribution data and then averaging the retrieved radiative transfer horizontally as a function of altitude.

Figure 6.

A histogram of optical thickness retrieved from the ship SSFR radiance observations on 16 May 2010 between 18:25:54 UTC and 20:54:14 UTC, a period where the instruments on the P3 and the Atlantis made coordinated observations.

3.4. Liquid Water Path Comparisons

[21] Using the effective radius and optical thickness retrieved from the ship SSFR radiance, comparisons to the retrieved microwave radiometer liquid water path (LWP) were made. The MWR and the SSFR share the same viewing geometry and have a similar FOV. The LWP is retrieved directly in the MWR and can be estimated with the SSFR retrieved properties using equation (2), assuming the effective radius is much larger than the wavelength of light and the effective radius and liquid water content do not vary within the sampling volume. Figure 7 shows the time series of the LWP for 16 May 2010 as shaded regions that include the uncertainty. The same gaps in the SSFR retrievals, described above, appear in this plot. The two retrievals agree within their respective limits of uncertainty.

display math
Figure 7.

Time series of the retrieved microwave radiometer LWP (gold) and SSFR LWP (blue) from 16 May 2010 aboard the Atlantis. The width of the line represents the uncertainty of the respective retrieval.

4. CalNex Cloud Statistics

[22] The Atlantis was deployed between 14 May 2010 and 5 June 2010 off the coast of California. One-third of these days exhibited a diurnal cloud pattern that has been documented in previous studies [Minnis and Harrison, 1984; Blaskovic et al., 1991; Duynkerke and Teixeira, 2001]. The cloud amount and liquid water path in marine boundary layer clouds was shown in these studies to be at its maximum just before sunrise, decreasing through the afternoon, and increasing again after sunset. The LWP pattern during CalNex is shown in Figure 8bwhere the CalNex-wide hourly average LWP is shown as a function of local solar time. This pattern is consistent with the previous findings. As cloud amount decreases, the cloud scene becomes more broken and more inhomogeneous. Under these conditions the irradiance observations become more variable. Under more homogeneous clouds, the irradiance is less variable. InFigure 8a, the ratio of the mean to the standard deviation (the inverse of the variation coefficient) of the ship SSFR irradiance observations is shown. The mean and standard deviation were calculated over one minute intervals and the ratio quantifies the variability of the irradiance. The ratio was averaged hourly over the course of the entire campaign. The ratio is relatively high under homogeneous conditions, decreasing with inhomogeneity. Figure 8a shows the daylight portion of the diurnal pattern in cloud amount.

Figure 8.

(a) The CalNex-wide hourly average of the ratio of the mean irradiance ( inline image) at 515 nm and standard deviation (σF) calculated of the irradiance at 515 nm. The irradiance statistics were calculated over one minute intervals, from the ship based SSFR observations. (b) The CalNex-wide hourly average of the MWR retrieved LWP. Both are shown as a function of local solar time.

[23] The distribution of the optical thickness in marine boundary layer clouds for 18 overcast scenes (a scene was defined as a 58 km2 Landsat image) was shown to be approximated by a gamma distribution [Barker et al., 1996]. We compare the shape of the cloud optical thickness distribution averaged over the entire CalNex campaign. In comparing the statistics of optical thickness it is important to reiterate the fact that for the SSFR retrievals only values of retrieved optical thickness greater than 5 were considered here. Figure 9shows the observed probability density function (PDF) and the best-fit gamma PDF. The two parameters that define the gamma PDF are the mean optical thickness (〈τ〉) and the variability parameter (ν = (〈τ〉/σ)2) which is the ratio of the mean and standard deviation of the observed optical thickness. For the CalNex campaign 〈τ〉 and ν were 23.3 and 2.6, respectively. The mean optical thickness was slightly higher, 23.3 compared to an upper value of 18.17 from Barker et al. [1996], but ν was within the range found in the Barker et al. [1996] study. These values are summarized in Table 4.

Figure 9.

The probability density function of the ship SSFR radiance retrieved cloud optical thickness (green) and the gamma PDF (blue).

Table 4. Average Mean Optical Thickness and Variability Parameter (ν = (〈τ〉/σ)2), the Ratio of the Mean Optical Thickness to the Standard Deviation, From Barker et al. [1996] for Cloud Scenes Designated as Overcast Compared to the Values Retrieved During CalNex
 Barker Overcast Scene AveragesCalNex SSFR Radiance Retrieved Optical Thickness
τ13.43 ± 4.7423.3
ν7.98 ± 6.292.6

[24] The distribution of liquid water path has also studied from satellite data. The LWP can be approximated by equation (2) using the cloud properties retrieved from the ship SSFR radiance observations. Previous studies have shown that the PDF of LWP can also be approximated by the gamma distribution [Barker et al., 1996] and that for high cloud fraction the LWP distribution is nearly Gaussian [Considine et al., 1997].

[25] Figure 10 shows the PDF of the LWP retrieved from the SSFR and MWR during CalNex and from Considine et al. [1997] using Landsat data of marine boundary layer clouds. The Considine et al. [1997] PDF shown here was constructed with cloud scenes where the cloud fraction was found to be between 99% and 100%. The LWP was retrieved using the visible channel 830 nm to retrieve the optical thickness and assuming a constant effective radius of 10 μm. The shapes of the LWP PDF from the SSFR radiance observations and the PDF from Considine et al. [1997] are similar for low LWP, less than 50 gm−2. The LWP observed during CalNex by the SSFR tends toward higher values than the previous study from Landsat data. The LWP observed by the MWR during CalNex resulted in a PDF with larger values of LWP than both the SSFR and the previous Landsat study. The Landsat and ship SSFR derived LWP were both estimated using equation (2) which assumes a cloud effective particle radius constant with cloud height. One possible explanation for the discrepancy is that this assumption does not hold over the course of the CalNex campaign, though it matched well for the 16 May 2010 case (Figure 5).

Figure 10.

Probability density functions for CalNex liquid water path observations from the microwave radiometer (red) and the ship SSFR radiance retrieved LWP (blue), compared to a previous study of LWP using Landsat data [Considine et al., 1997].

5. Conclusions

[26] A cloud retrieval algorithm [McBride et al., 2011] developed for application to transmitted radiance was presented and compared with other cloud property retrievals with the goal of improving and further validating the algorithm. The algorithm was also applied to transmitted irradiance observed aboard the WHOI R/V Atlantis and the NOAA WP-3D. The data used in the validation was collected during CalNex on 16 May 2010, a day featuring coordinated observations between the P3 and Atlantis. The coordination allowed for comparisons of cloud properties obtained from below the cloud deck from the Atlantis and the P3, within the cloud from the P3, and from above the cloud from satellite and the P3. The comparisons with the optical thickness and LWP retrievals showed an expected dependence on the instrument FOV. The percentage of the retrievals where the uncertainties overlapped increased as the difference in the FOV of the instruments decreased. The comparisons between the ship SSFR radiance derived LWP and the MWR LWP retrievals had the highest percentage of overlap at 99% with a FOV differing by a factor of 2. The comparisons to the ship SSFR radiance retrievals to the ship SSFR irradiance and GOES retrievals resulted in agreement 30% and 22% of the time with the FOV differing by a factor of about 50 and 400, respectively. Comparisons of the GOES retrieved optical thickness and the SSFR optical thickness retrievals averaged over 15 min were shown to be within 15.2% and 19.1% for the radiance and irradiance retrievals, respectively.

[27] During the 16 May 2010 case, there was a period of approximately 90 min where the P3 flew a coordinated pattern near the Atlantis. The average in situ effective radius profile was collected during this time period and resulted in a vertical profile with an average effective radius of 7.7 μm. The mean effective radius retrieved from the ship SSFR radiance during the coordinated flight leg was 5.7 μm. The retrievals from the ship SSFR irradiance and GOES resulted in mean values of 9.5 μm and 11.5 μm, respectively. The variability in effective radius obtained through remote sensing techniques with different sampling volumes has been documented in the past [Twomey and Cocks, 1989; Rawlins and Foot, 1990; Nakajima et al., 1991] and is the most likely contributor to the variability seen in these retrievals.

[28] In addition to the validation of the transmittance algorithm, cloud statistics for the CalNex campaign were provided. It was shown that the marine boundary layer clouds encountered during the campaign followed a diurnal pattern observed in marine boundary layer clouds similar to analysis provided in previous studies. The LWP retrieved from the MWR showed the entire diurnal cycle of the LWP. For the optical methods, the daylight half of the diurnal pattern for cloud amount was shown using the variability of parameters derived from SSFR irradiance measurements. Using Landsat observations over cloud scenes with high cloud fraction, the cloud optical thickness [Barker et al., 1996] and cloud LWP [Considine et al., 1997] were shown to be representative of a gamma distribution. The CalNex surface observed cloud optical thickness PDF was shown to be representative of a gamma PDF and the fit gamma parameters, 〈τ〉 and ν, similar to the averages obtained by Barker et al. [1996] for 18 high cloud fraction scenes. The average optical thickness for CalNex was 23.3 compared to the upper value of the Barker et al. averages of 18.17 and the ν for CalNex was 2.6 compared to the range of 7.98 ± 6.29. The shape of the CalNex ship SSFR radiance derived LWP PDF was similar to that found from Landsat observations taken over cloud scenes with high cloud fraction. The LWP observations from the MWR during CalNex resulted in a PDF that was different from the SSFR LWP and the previous Landsat study. The difference is most likely the result of the different sampling volumes of the instruments.

[29] Further work with the transmittance based algorithm will employ the method in Coddington et al. [2012] which uses the Shannon Information Content. This allows for a more systematic way to quantify the ability of the algorithm to retrieve cloud properties. This method has been applied to hyperspectral cloud observations by Coddington et al. [2012]. The information content can provide a more quantifiable method to select retrieval wavelengths and possibly improve the algorithm. The discrepancies in the effective radius retrievals require further study. It was suggested in Section 3.3 that the differences could be the result of horizontal heterogeneity, but other possible explanations include vertical heterogeneity, viewing geometry, and instrument calibration. Marshak et al. [2006]showed that over heterogeneous clouds, cloud reflectance retrieved effective radius can be both overestimated and underestimated. The extent to which heterogeneity plays a role in the transmittance effective radius retrievals can be explored through 3D modeling studies, including the individual contribution of horizontal and vertical heterogeneities. The CalNex data set was the first application of the transmittance retrieval algorithm outside of hand-selected cases and the impacts of cloud heterogeneity on cloud transmittance retrievals is a matter of ongoing research. The spectral observations of the SSFR can potentially be utilized to quantify cloud heterogeneity;Kokhanovsky and Rozanov [2011] is an example of how this information is exploited for deriving for vertically distributed effective radius.

Acknowledgments

[30] This work was supported through the NOAA CalNex grant for the SSFR group. The authors would like to recognize the efforts from Warren Gore and Tony Trias from NASA Ames for their continued and reliable support of SSFR during these deployments.