Journal of Geophysical Research: Atmospheres

Retrievals of vertical profiles of stratus cloud properties from combined oxygen A-band and radar observations

Authors


Abstract

[1] A synergetic algorithm of retrieving vertical distribution of cloud drop effective radius (Re), liquid water content (LWC), and optical depth has been developed by combining oxygen A-band spectral measurements with radar reflectivity. The underlying physics is to utilize the photon path length information from oxygen A-band measurements to constrain the retrievals of cloud microphysical properties from radar reflectivity. Through a radiation closure of the rotating shadowband spectrometer (RSS) diffuse-horizontal irradiances in the oxygen A-band (which is sensitive to vertical distribution of the 2nd moment of particle size distribution), vertical distributions of cloud microphysical property (the 3rd moment of particle size distribution) are optimized and retrieved from Millimeter Wave Cloud Radar (MMCR) reflectivity (the 6th moment of particle size distribution). The critical issues in the synergetic cloud retrievals of combining passive and active instruments are discussed and carefully taken into account. Evaluation and validation through case studies show that the retrieved column mean values of cloud optical depth, effective radius, and liquid water path agree well with independent measurements from the microwave radiometer (MWR) and independent retrievals from a combined algorithm of multifilter rotating shadowband radiometer (MFRSR) and MWR and the column physical characterization (CPC) of MMCR. These studies demonstrate that this active-passive synergetic retrieval algorithm is feasible and accurate for retrievals of vertical distribution of cloud Re and LWC for stratus clouds.

1 Introduction

[2] Understanding radiation transfer in the atmosphere is a crucial step to properly characterize radiative forcing and quantify the response of climate change to the forcing. Clouds play an important role in climate systems through their radiative effects and their vital link in the hydrological cycle. Detailed knowledge of the vertical distribution of cloud macrophysical/microphysical properties is important to calculate the Broadband Heating Rate Profile, a fundamental driver in climate systems [Stephens, 1978; Li & Min, 2010]. However, our current remote sensing methods for retrieving cloud properties (microphysical properties in particular), such as effective radius (Re) and liquid water content (LWC), are still uncertain. Current remote sensing measurements are not able to observe all the relevant parameters that define cloud microphysical and optical properties. All retrievals from those measurements with limited information, one way or another, have to make certain assumptions, which would introduce some uncertainties [Frisch, Feingold, Fairall, & Uttal, 1998; Mace & Sassen, 2000; Shupe, Uttal, Matrosov, & Frisch, 2001; McFarlane, Evans, & Ackerman, 2002; Dong & Mace, 2003]. An important challenge is to reduce ambiguity and uncertainty by minimizing and constraining those assumptions in retrievals. To do so requires a synergetic approach that combines various measurements to enhance information content.

[3] Advances in radar technology, such as MMCR operated at various Atmospheric Radiation Measurement (ARM) sites, enable continuous observation of cloud properties that have significant influence on radiative transfer through cloudy atmosphere. However, observed radar reflectivity is a function of the 6th moment of hydrometeor particle size distribution. To derive cloud water content (liquid or ice) for practical applications, which is a function of the 3rd moment of hydrometeor particle size distribution, requires an assumption of particle size distribution. To constrain assumed size distribution, integrated cloud optical depth from radiation measurements and/or integrated liquid water path from a microwave radiometer have been applied [Frisch, Fairall, & Snider, 1995; Frisch et al., 1998; Moran et al., 1998; Miles, Verlinde, & Clothiaux, 2000; Mace & Sassen, 2000; Shupe et al., 2001; Austin & Stephens, 2001; McFarlane et al., 2002; Dong & Mace, 2003]. Because of limited information from most of these integrated measurements, both total number concentration and width of size distribution are assumed to be vertically uniform and sometimes constant. As observed from in-situ measurements [Slingo, Nicholls, & Schmetz, 1982; Min et al., 2012], vertical distribution of cloud microphysical properties, size distribution, and liquid water content vary sub-adiabatically. Min et al. (2012) pointed out that not considering such vertical inhomogeneity would result in substantial errors in retrieved optical properties.

[4] The advantages of using oxygen A-band measurements have been recognized by the remote sensing community [Grechko, Dianov-Klokov, & Malkov, 1973; Fischer & Grassl, 1991; O'Brien & Mitchell, 1992; Harrison & Min, 1997; Pfeilsticker, Erle, Veitel, & Platt, 1998; Pfeilsticker, 1999; Veitel, Funk, Kruz, Platt, & Pfeilsticker, 1998; Min & Harrison, 1999; Stephens & Heidinger, 2000; Heidinger & Stephens, 2000; Portmann, Solomon, Sanders, & Danel, 2001; Min, Harrison, & Clothiaux, 2001; Heidinger & Stephens, 2002; Davis & Marshak, 2002; Min & Clothiaux, 2003; Min et al., 2004a; Min et al., 2004b; Min & Harrison, 2004; Scholl et al., 2006; Li & Min, 2010; and many others]. Since oxygen is a well-mixed gas in the atmosphere, pressure dependence (as a surrogate of altitude) of oxygen A-band absorption line parameters provides a vehicle for retrieving vertical profiles of atmospheric constituents from oxygen A-band spectrometry. Information content analysis by Min and Harrison (2004) showed that there are four or five independent pieces of information from high resolution oxygen A-band measurements, depending on instrument resolution, out of band rejection and signal-to-noise ratio. Even from the modest resolution measurements of rotating shadowband spectrometry, Min and Clothiaux (2003) demonstrated that two independent pieces of information can be derived. Recently, Li and Min (2010) analyzed variance and mean of the photon path length distribution inferred from RSS measurements at the ARM Southern Great Plains (SGP) site. They also showed that photon path length from RSS oxygen A-band spectrum is able to detect possible “missed” upper layer clouds by a collocated high-sensitivity MMCR. More importantly, the photon path length inferred from RSS oxygen A-band spectra is a function of the 2nd moment of particle size distribution, which can be used to better constrain assumed particle size distribution of radar retrievals.

[5] In this paper, we develop a synergetic algorithm to retrieve vertical distribution of Re, LWC, and optical depth (τ) for stratus clouds by combining oxygen A-band spectral measurements with radar reflectivity. Specifically, we use RSS measurements in the oxygen A-band and MMCR measured reflectivity at the ARM SGP site, and evaluate our retrievals with independent microwave radiometer (MWR), liquid water path (LWP), and MMCR-based Column Physical Characterization (CPC) [Mace et al., 2006].

2 Passive and Active Measurements and Retrieval Algorithm

2.1 MMCR and Radar Reflectivity

[6] The MMCR is a centerpiece among the ground-based instrument suite at the ARM SGP site. The MMCR operates at a wavelength of 8 mm (35 GHz) in a vertically pointing mode, and provides continuous profiles of radar reflectivity of hydrometeors, mean vertical velocity, and Doppler spectrum width simultaneously [Moran et al., 1998]. The beam width of MMCR is 0.2o, which yields lateral resolutions of 35 m at a height of 10 km for a vertically directed beam. To detect different kinds of clouds, the MMCR is coded with various pulse wave forms, resulting in higher and lower vertical resolutions (45 and 90 m). For each data acquisition mode, sensitivities, height coverage, folding velocities, and other features are different. The sensitivity of the coded modes is almost -50 dBZ at a height of 5 km above ground level.

[7] At the MMCR wavelength, water cloud droplets are small enough to be treated as Rayleigh scatterers [Stephens, 1994]. With this assumption we can write the expression of radar reflectivity Z in terms of droplets size distribution N(r):

display math(1)

The cloud microphysical properties, Re and LWC, can also be expressed in terms of cloud droplet size distribution:

display math(2)
display math(3)

where ρw is the density of water.

[8] In general, a lognormal size distribution of water cloud droplets is assumed in the retrieval of cloud microphysical properties as

display math(4)

where Nt is the droplet number concentration, σlog is the logarithmic width of the size distribution, and rg is geometric mean radius, which is defined by

display math

[9] With the lognormal assumption, using equation (4) to substitute the size distribution in equations (1)(3), radar reflectivity Re and LWC can be written in terms of Nt, σlog, and rg as

display math(5)
display math(6)
display math(7)

[10] We take 10log10 on both sides of equation (5), then we relate cloud microphysical properties to the radar reflectivity by combining equations (6) and (7) with equation (5) to eliminate rg. Then we can obtain:

display math
display math

where the units of Z, Nt, re, LWC, and ρw are mm6m- 3, cm- 3, µm, gm- 3and gm- 3, respectively. Then the Re and LWC can be expressed as

display math(8)
display math(9)

where

display math
display math

[11] In equations (8) and (9), a and b are two free parameters to be determined in our retrieval of Re and LWC. We assume that both of them are constant with height and vary with time.

2.2 RSS and Oxygen A-band Spectrum

[12] The RSS uses prisms to disperse the sunlight from 360 to 1100 nm and couples a 1024-pixel CCD spectrograph with a spectral resolution of 2.3 nm within the oxygen A-band. The RSS implements the same automated shadow banding technique as MFRSR [Harrison, Michalsky, & Berndt, 1994], so that it can simultaneously measure spectrally-resolved direct-normal, diffuse-horizontal, and total-horizontal irradiances. Applying Mie theory, cloud extinction coefficient inferred from solar radiation is a function of the 2nd moment of particle size distribution [Austin & Stephens, 2001]. Such information provides a better constraint on cloud microphysical retrievals (the 3rd moment of size distribution) from radar reflectivity (the 6th moment of size distribution). Furthermore, the pressure dependence of oxygen A-band absorption line parameters enables retrieval of vertical profiles of atmospheric scatterers (aerosols and/or clouds). In our passive-active synergetic retrievals, measured diffuse-horizontal irradiances are used to constrain the vertical distribution of cloud properties derived from radar reflectivity through a radiation closure. However, the accuracy of RSS spectral calibration is about 4%. At the ARM SGP site, there is a well-calibrated MFRSR, which measures the spectral irradiances of direct normal, diffuse horizontal, and total horizontal at 415, 500, 615, 673, 870, and 940 nm wavelengths, with Langley calibration accuracy of about 1%. More importantly, we have developed a well-validated family of cloud retrieval algorithms for MFRSR [Min & Harrison, 1996; Min et al., 2004; Wang & Min, 2008]. To avoid the uncertainties associated with absolute calibration of RSS, we only utilized the oxygen A-band spectral absorption features by normalizing the observed oxygen A-band spectrum with a linear fit of nonabsorption shoulders at each side of oxygen A-band, and used MFRSR measured irradiance for total cloud optical depth retrievals.

[13] The detailed retrieval algorithm will be discussed in the next section. A critical issue in the retrieval is that each spectral irradiance measurement (i.e., pixel response) represents an integral over a wavelength region corresponding to the slit function of the pixel, which can be transformed into an integral over a distribution of the oxygen absorption optical depth. To simulate the diffuse radiation measured by the RSS,we used the MODTRAN 4.0 [Anderson et al., 1999] as our forward radiative transfer model, in which oxygen absorption and atmospheric optical properties (Rayleigh, aerosol, and cloud) are combined to simulate scattering and absorption of the atmosphere.

[14] The oxygen absorption spectrum used in MODTRAN (derived from HITRAN) and the RSS slit function measured in the laboratory are not sufficiently accurate or congruent to allow the necessary accuracies in the needed diffuse irradiation calculation. To minimize these problems we utilized RSS solar direct-beam observations to empirically calibrate RSS slit function. As RSS uses an automated shadow banding technique to provide spectrally resolved solar direct-normal, diffuse-horizontal, and total-horizontal irradiances, the passbands and pixel responses are guaranteed to be identical for the separated spectral irradiance components. For our purposes, measurements of the solar direct beam under clear skies, in which the photon path length is well known, allow for accurate calibration of the RSS slit function from MODTRAN oxygen A-band spectroscopy. Once calibrated, the modified RSS slit function can be applied to cloudy sky cases.

[15] For doing so, we convolved the MODTRAN simulated direct-beam irradiance spectrum with a modified RSS slit function to simulate the RSS oxygen A-band spectrum, at corresponding solar zenith angle, surface pressure, and total aerosol optical depth for each RSS measurement. Again, to avoid the uncertainties associated with absolute calibration, we normalized both observed and modeled oxygen A-band spectrum with a linear fit of nonabsorption shoulders at each side of the oxygen A-band. By minimizing the difference between modeled and observed direct-beam spectra for a large range of solar zenith angles through a least square regression, the calibrated or modified RSS slit function is derived (Figure 1a). Compared with the originally measured RSS slit function, the calibrated slit function has a lower out of band rejection, suggesting that either the strengths of oxygen absorption lines are too strong in MODTRAN or the RSS straylight is deteriorated from the original in-lab calibration. The difference between modeled and observed direct-beam spectra for solar zenith angles varying from 35 to 65 degrees on 4 March 2000, shown in Figure 1b, is less than +/-4%. As the wavelength region corresponding to each individual pixel may drift as a result of ambient temperature variations, we have registered wavelengths using both oxygen A-band spectroscopy and Fraunhofer lines for each observed spectrum. However, the residual difference in the band center shifts from positive to negative, indicating that the accuracy of wavelength registration is still an issue. It is worth noting that the number of independent pieces of information inferred from the measurements is mainly affected by instrument resolution, out of band rejection, and the signal-to-noise ratio [Min & Harrison, 2004]. Wavelength registration is important to characterize instrument performance and to compare measurement with theoretical calculations (or our forward modeling). It does not impact information content, unless the changing air pressure (which affects refractive index) and temperature (which affects optical alignment) within the RSS also affect the three key functions, i.e., instrument resolution, out of band rejection, and signal-to-noise ratio. We have no direct indication that such will occur. Therefore, we assumed that this wavelength drift does not affect the number of independent pieces of information inferred from the measurement. We should use all pixels of oxygen A-band for the radiation closures or for the optimization of the retrievals.

Figure 1.

(a) Comparison of the calibrated RSS slit function (MOD) with the originally measured RSS slit function (ORI). (b) Difference between modeled and observed normalized direct-beam spectra for various solar zenith angles.

[16] So far, RSS is the only instrument at the ARM sites that provides a modest resolution of spectral irradiance in oxygen A-band. As demonstrated by Min and Clothiaux (2003), RSS measurements can provide two independent pieces of information: mean and variance of photon path length distribution. Information of photon path length from the RSS oxygen A-band spectrum is able to detect possible “missed” upper layer clouds by a collocated high-sensitivity MMCR [Min & Clothiaux, 2003; Li & Min, 2010]. Recently, a high resolution oxygen A-band spectrometer (HABS) has been developed and tested during the field campaign at Howard University's Beltsville Center for Climate System Observation (BCCSO) [Li & Min, 2012]. HABS has the capability to measure both zenith and direct-beam radiances with a field of view of 2.7 degrees. Direct-beam measurements can be used to calibrate the spectrometer and construct retrieval kernels for zenith measurements. HABS also measures polarizations of A-band spectra with four polarizers, which substantially enhances retrieval capability for aerosols and ice clouds. Our tests indicate that HABS achieves an out-of-band rejection of 10-5, a resolution of better than 0.3 cm-1, and a high signal-to-noise ratio. HABS is capable of providing at least four independent pieces of information [Min & Harrison, 2004]. We expect better retrievals of cloud profile from the combination of HABS and cloud radar in the near future.

2.3 Retrieval Algorithm

[17] Microphysical properties of clouds, specifically Re and LWC, are derived from radar observed reflectivity through equations (8) and (9), in which two free parameters, a and b, are functions of cloud particle size distribution. Our retrieval algorithm utilizes oxygen A-band spectral irradiance measurements to constrain cloud size distribution and determine both free parameters by radiation closure. To avoid absolute calibration of RSS and take advantage of well-calibrated MFRSR and associated cloud optical depth retrievals, we defined the retrieval cost function, y, as a summation of difference squares of cloud optical depth (from MFRSR) and normalized RSS oxygen A-band spectrum:

display math(10)

where τ, Gτ(a,b) and  στ are the inferred and modeled cloud optical depth and error associated with inferred cloud optical depth, respectively. difi and Fi(a, b) are the measured and modeled normalized irradiance for the ith pixel in oxygen A-band, respectively. [σi] is the measurement uncertainty. Through a nonlinear leastsquares regression, a and b can be determined by minimizing cost function. Consequently, the vertical profiles of Re and LWC can be derived from equations (8) and (9). For the nonlinear leastsquares regression, a set of a priori parameters a and b are derived from the statistical study of cloud droplet size distribution [Miles et al., 2000]. Our tests indicate that the final retrievals are insensitive to the initial guesses and the nonlinear leastsquares regression is converged within 10 iterations.

[18] It is worthnoting that the cloud boundaries for atmospheric pressure, which are observed by MMCR, define the basic depth of oxygen absorption at a given wavelength in the oxygen A-band. Since cloud optical properties depend on cloud droplet size distribution and cloud LWC, any change in cloud effective radius and LWC will result in changes in multiple scattering properties, i.e., phase function, single scattering albedo, and optical depth. As multiple scattering within a cloud would enhance the photon path length, the enhanced photon path length will result in additional absorption and, consequently, reduction of the diffuse irradiance. Observed oxygen A-band spectral irradiance is sensitive to change of cloud optical/microphysical properties. Figure 2 shows sensitivities of normalized diffuse irradiance measured at five different oxygen absorption depths and cloud optical depth (retrieved from the irradiance at nonabsorption wavelength or the A-band shoulders) to parameters a and b. For a fixed cloud optical depth (62), the cloud with larger parameter a, i.e., a larger Re and stronger forward scattering, has a smaller scaled optical depth. Hence, photons experience less oxygen absorption, resulting in a relative higher irradiance (Figure 2a). For fixed parameter a, the larger the parameter b, shown in Figure 2b, the larger LWC and cloud optical depth are (or the lower the nonabsorption irradiance). Also, as shown in Figure 2a, normalized diffuse irradiance at a greater oxygen absorption wavelength (larger oxygen absorption depth) is more sensitive to parameter a. The consistence within oxygen A-band spectral irradiances further constrains the set of parameters.

Figure 2.

The sensitivities of the normalized diffuse irradiance measured at five different oxygen absorption depths and the cloud optical depth (retrieved from the irradiance at the nonabsorption wavelength or the A-band shoulders) to both parameters of a and b.

3 Results

[19] Validation and evaluation of retrieved products are key to showing the effectiveness of a retrieval algorithm. To demonstrate the feasibility of this active-passive synergetic retrieval algorithm, i.e., A-band radar algorithm (ARA), we processed the measurements of RSS, MMCR, and MFRSR at the ARM SGP site for the year 2000. Based on the multiple layer cloud diagnosis algorithm developed by Li and Min (2010), we selected two single-layer stratus cloud cases (18 March and 15 April) to test our retrieval algorithm. To evaluate the retrievals of vertical distribution of cloud properties, we compared the ARA retrieved vertical profiles of Re and LWC with CPC [Mace et al., 2006]. Also, we validated the vertically integrated cloud properties (cloud optical depth, averaged effective radius (Re), and LWP) against the retrievals of MFRSR and measurements of MWR.

[20] In general, various instruments have different sampling rates and observational geometries. It is essential to understand the effects of spatial-temporal variability on radiation parameters retrieved from multiple instrument measurements, particularly when combining radiative measurements and active or passive observations that are confined to relatively narrow vertical columns through clouds. Furthermore, radiation smoothing due to multiple scattering of cloud particles results in relatively large field of view radiation measurements [Marshak, Davis, Wiscombe, & Cahalan, 1995; Von Savigny, Funk, Platt, & Pfeilsticker, 1999]. Through a scale-by-scale statistical analysis, Min et al. (2001) showed that there is a break-point scale, or “radiative smoothing scale,” at about 3 km-1 that separates two physically distinct regimes. In addition, cloud microphysical properties and radiation fields have the same power spectrum for scales larger than the break-point scale. Further analysis indicated that the optimal spatial scale is at 3 times the “radiative smoothing scale,” or the temporal scale is about 5-minutes, for the cloud advection speed of 5 ms-1 [Min et al., 2001]. Hence, in our ARA retrievals, we synchronized the RSS measurements with 5- minute averaged cloud fields from active radar and passive MWR.

3.1 Case 1 (15 April 2000)

[21] Based on the MMCR reflectivity measured at the ARM SGP site (shown in Figure 3), a continuous single-layer stratus with nearly constant cloud geometric thickness of 2 km occurred between 1500 UTC and 1600 UTC on 15 April 2000. The ARA retrieved profiles of Re and LWC were compared with results of CPC, shown in Figure 4. The CPC retrievals are based on combining information of MMCR reflectivity with LWP from MWR [Mace et al., 2006]. Although magnitudes of Re and LWC are slightly different, the vertical variations are consistent with each other: Re increases with height, reaches maximum at about 1 km altitude, and then decreases rapidly to about 5 µm. LWC also increases with height and reaches maximum at about 1 km altitude.

Figure 3.

MMCR radar reflectivity of cloud layers during the 15 April 2000 case study period: (a) whole day radar reflectivity and (b) 5 minutes averaged radar reflectivity from 1509 UTC to 1612 UTC.

Figure 4.

Vertical profiles of cloud microphysical properties derived from the ARA and the CPC during the 15 April 2000 case study period: (a) effective radius retrieved by the CPC, (b) effective radius retrieved by the ARA, (c) liquid water content retrieved by the CPC, and (d) liquid water content retrieved by the ARA.

[22] To further test an optimal averaging interval of MMCR reflectivity with RSS radiation, we simply applied the ARA algorithm to different averaging intervals of MMCR. As shown in Figure 5, at 1554 UTC, a one-minute averaging profile of MMCR reflectivity is substantially different from the profiles of longer averaging intervals, particularly in the cloud top. Even though the cloud top layer is optically thin, the higher cloud top could result in an enhanced photon path length. From the retrieval perspective, it requires a strong forward scattering to compensate for the enhanced path length due to the increase of cloud thickness for a given total cloud optical depth, resulting in a relatively large cloud effective radius and LWC. The differences in retrieved effective radius and LWC due to such an optically thin layer manifest the sensitivity of oxygen A-band to the profile of cloud microphysical structure. Furthermore, the retrievals with 5-minute averaging intervals of MMCR reflectivity exhibit consistent results, which confirm our assessment on the optimal scale of a 5-minute averaging interval for synergizing radiative measurements with narrow field of view active or passive observations. The CPC retrieved optical depth is slightly smaller than the ARA retrievals, while the Re and LWC are larger than the ARA retrievals.

Figure 5.

Retrieved vertical profiles of cloud microphysical properties based on three averaged methods (1 minute, 2 minutes, and 5 minutes) at 1554 UTC on 15 April 2000: (a) profiles of MMCR radar reflectivity, (b) profiles of optical depth, (c) profiles of effective radius, and (d) liquid water content.

[23] To further validate the ARA retrievals, we compared the averaged Re (weighted by cloud optical depth) with total LWP and total cloud optical depth, shown in Figure 6. As discussed previously, Min and Harrison (1996) developed an accurate retrieval algorithm of cloud optical depth and effective radius with the measured LWP from a MWR. This algorithm has been extensively validated against in-situ aircraft measurements [Min, Duan, & Marchand, 2003] and compared with many other retrievals [Turner et al., 2007]. It has been implemented as an operational algorithm (value-added product) in the ARM. Figure 6a shows the intercomparison of effective radius from MFRSR/MWR, ARA, and CPC. In general, the ARA retrieved Re agrees well with the MFRSR/MWR combined retrievals, and both of them are slightly smaller than the CPC retrievals. It is noteworthy that the ARA retrievals utilize the information of the cloud optical depth from MFRSR, i.e., a constraint of atmospheric transmittance. The effective radius retrieved from the MFRSR and MWR combination, however, is strongly dependent on the LWP, which is measured from a MWR and is totally independent of MMCR reflectivity and RSS oxygen A-band spectral measurements or the ARA retrievals.

Figure 6.

Comparison of averaged effective radius, liquid water path, and optical depth: (a) averaged effective radius retrieved by MFRSR, ARA, and CPC, (b) liquid water path retrieved by ARA, CPC, and MWR, and (c) optical depth retrieved by MFRSR, ARA, and CPC for Case Study 1.

[24] The ARA retrieved LWP agrees well with the MWR LWP, a fully independent retrieval. It is clear that the LWP measured from the MWR exhibits a larger variation than the LWP retrieved from ARA. As the ARA retrievals are based on a radiation closure of the oxygen A-band spectrum, radiative smoothing due to multiple scattering evens out fluctuation of the cloud microphysical field, resulting in a much smoother LWP. The detailed variation of cloud microphysical fields is important for the understanding of cloud dynamical and microphysical processes. The smoothed cloud fields, however, are key for radiation processes in the atmosphere.

[25] Cloud optical depth derived from ARA should agree with that from MFRSR (shown in Figure 6c), since it is used as a constraint. Retrievals of cloud optical depth from CPC are also consistent with the MFRSR retrievals, although the CPC retrieved LWP and Re are both relatively larger than the MWR and MFRSR retrievals. Basically, the ARA retrievals of cloud microphysical/optical properties are consistent with both MFRSR/MWR and CPC retrievals.

[26] All measurements suffer some uncertainties and biases. To understand retrieval uncertainty associated with measurement uncertainties, we varied each key measurement within its uncertainty range: ± 0.5 dBZ for radar reflectivity, 1% for RSS A-band spectral irradiance, and 1% for MFRSR retrieved cloud optical depth. The ARA retrieved Re and LWC vary within ±10%, suggesting our retrieval uncertainty is about 10%.

3.2 Case 2 (18 March 2000)

[27] Case Study 1 illustrates the capability of our ARA retrievals for a warm stratus cloud. However, some stratus clouds grow higher above the freezing level and may form some ice particles through heterogeneous nucleation. To illustrate the applicability of our retrieval algorithm for some mixed phase clouds, we selected a stratus cloud that occurred between 1806 and 1900 UTC on 18 March 2000. To get the cloud's robust reflectivity profile, particularly for the optically thin cloud layer at the cloud top, we not only used the MMCR's most accurate mode 4 reflectivity [Clothiaux et al., 1999], but also considered the other three modes. As shown in Figure 7a, there was a horizontally relatively uniform stratus cloud layer with cloud top at 4 km, well above the freezing level of 2.2 km determined from the balloon sounding at the ARM SGP site. It is plausible that this cloud layer may contain some ice particles.

Figure 7.

The measurements of MMCR radar during the 18 March 2000 case study period: (a) MMCR radar reflectivity, (b) 5 minutes averaged reflectivity from 1806 UTC to 1900 UTC, (c) MMCR radar mean Doppler velocity, and (d) MMCR radar reflectivity derived from liquid water.

[28] Due to large particle size and shape, the sensitivity of radar reflectivity to ice particles is very different from that to liquid water droplets. It is important to classify cloud phases and to partition different phases in terms of radar reflectivity. Here we adopted a cloud phase classifier proposed by Shupe (2007), which utilizes the combined information of temperature and radar mean Doppler velocity (Figure 7c) and Doppler spectrum width. The total reflectivity profiles are shown in Figure 7b, while the reflectivity classified as liquid water clouds is shown in Figure 7d. As the reflectivity from ice particles is much weaker than that from liquid droplets, we treated the ice clouds as a perturbation; the optical properties of ice clouds are simply retrieved via the empirical retrievals of Matrosov, Korolev, and Heymsfield (2002). Compared with the CPC ice cloud retrievals (shown in Figure 8), our empirical retrievals are much thicker with smaller ice water contents (IWC) and larger ice particles.

Figure 8.

Vertical profiles of ice cloud microphysical properties retrieved by ARA and CPC during the 18 March 2000 case study period: (a) ice effective radius retrieved by CPC, (b) ice effective radius retrieved by ARA, (c) ice water content retrieved by CPC, and (d) ice water content retrieved by ARA.

[29] With the retrieved ice cloud properties, we then applied our ARA retrieval algorithm to the combined ice and liquid cloud column to retrieve the liquid cloud optical/microphysical properties. Overall, the ARA retrieved vertical profiles of  Re and LWC (shown in Figure 9) are similar to the CPC retrievals. However, the ARA retrieved  Re are smaller than the CPC retrievals from 1848 to 1854 UTC, and the ARA retrieved LWC are smaller than the CPC retrieved LWC from 1848 to 1854 UTC.

Figure 9.

The same as Figure 4, but for 18 March 2000.

[30] The upper layer clouds, even though optically thin, can enhance photon path length and consequently enhance oxygen absorption, resulting in high sensitivity of oxygen A-band measurements to the upper layer clouds [Li & Min, 2010]. Therefore, it is important to test detection and phase classification of upper clouds for the ARA retrievals of cloud microphysical properties. To demonstrate the effect of our phase classification and parameterization on the retrievals, we made two extreme assumptions: (1) clouds above the freezing level are super-cool liquid water clouds and (2) clouds above the freezing level are all ice clouds. Once again, we used the integrated cloud optical/microphysical properties for our comparisons, shown in Figure 10. The retrievals of average Re indicate that the ARA retrievals with phase classification agree well with MFRSR/MWR retrieved average Re from 1806 to 1842 UTC. The CPC retrieved Re are systematically smaller than the MFRSR/MWR retrievals. With the two extreme assumptions, the retrieved average Re is either too large under the all liquid cloud assumption or too small under the all ice cloud assumption. The ARA LWP retrievals with phase classification also show consistent results with the MWR measurements, whereas the retrievals with the extreme assumptions disagree with the MWR measurements. The results suggest that the upper part of the stratus cloud contains both ice and liquid phase particles, and that phase classification and empirical retrieval produce better and consistent retrievals from 1806 to 1842 UTC. However, from 1842 to 1900 UTC, the ARA retrieved Re is smaller than that retrieved by both CPC and MFRSR/MWR, while the ARA retrieved LWP is smaller than that retrieved by both CPC and MWR. These findings indicate that the LWC or LWP are underestimated between 1842 and 1900 UTC, implying that the phase classification needs some modification.

Figure 10.

Comparison of averaged effective radius, liquid water path, and optical depth: (a) averaged effective radius retrieved by MFRSR, ARA, and CPC, (b) liquid water path retrieved by ARA, CPC, and MWR, and (c) total optical depth retrieved by MFRSR, ARA, and CPC and optical depth of ice cloud retrieved by ARA and CPC.

[31] As shown in Figure 10c, the ARA retrieved optical depths are consistent with MFRSR retrievals. However, the CPC retrieved optical depths are much larger than MFRSR retrieved optical depths. The optical depths of ice clouds from both ARA and CPC (Figure 10c) are small, compared to the total optical depth. The fundamental issue of cloud phase classification is whether the observed reflectivity at the cloud top should be treated as liquid water clouds. The retrieved optical depth from observed reflectivity may be small if the cloud layer is assumed to be in ice phase, but may be optically significant if the cloud layer is assumed to be in liquid phase, as MMCR reflectivity is less sensitive to small, liquid particles. This case illustrates the importance of phase classification for clouds cooler than 0oC.

4 Conclusion

[32] A synergetic algorithm of retrieving vertical distribution for Re, LWC, and optical depth has been developed by combining oxygen A-band spectral measurements with radar reflectivity. Through a radiation closure of RSS diffuse-horizontal irradiances in the oxygen A-band, two coefficients associated with cloud particle size distribution in the retrievals of cloud properties from the MMCR reflectivity are optimized or constrained. Furthermore, three critical issues in the synergetic cloud retrievals of combining passive and active instruments are discussed and carefully taken into account:

  1. Instrument calibration, particularly on the RSS slit function. The radiometric calibration of MFRSR has been transferred into the RSS calibration of total transmittance, and the normalized RSS oxygen A-band spectrum has been used in the radiation closure. The RSS slit function is calibrated using direct beam observations of the RSS and adjusted to minimize the difference between MODTRAN calculated spectra and observations. The calibrated RSS slit function is an optimal slit function consistent with MODTRAN oxygen absorption coefficients.
  2. Cloud spatial-temporal variability and measurement consolidation. Considering radiation smoothing and the viewing geometries of MMCR, MWR, and RSS, photons received by the RSS most likely interact with clouds observed by MMCR and MWR within 5 minutes duration. Therefore, RSS measurements are synchronized with 5-minute averaged cloud fields from active radar and passive MWR.
  3. Cloud phase classification and reflectivity partition. Currently, the ARA is a liquid phase retrieval algorithm that can also be used to retrieve stratus with small ice particle contamination, where optical depth of ice clouds is small and can be treated as perturbation. However, sensitivity of radar reflectivity to ice particles is much stronger than that to liquid water droplets. In such situations, the fundamental issue of cloud phase classification is to determine where MMCR reflectivity can be treated as liquid clouds. In the present study Shupe's cloud phase classifier, which utilizes the combined information of temperature, radar mean Doppler velocity, and Doppler spectrum width, is adopted to select the phase of cloud particles.

[33] To evaluate and validate this active-passive synergetic retrieval algorithm, ARA retrieved vertical profiles of Re and LWC are compared with independent retrievals of CPC [Mace et al., 2006]. Also, vertically integrated cloud properties, cloud optical depth, averaged effective radius, and liquid water path are validated against the retrievals of MFRSR and measurements of MWR. For the liquid water cloud case, ARA retrievals of cloud microphysical/optical properties are consistent with both MFRSR/MWR and CPC retrievals. For the ice particle contaminated cloud case, our sensitivity study indicated the importance of a good cloud phase classification. For a given MMCR reflectivity, the cloud with pure liquid particles may have much larger water content and smaller particle size than the cloud with ice particles, resulting in greater cloud optical depth for the liquid cloud. After proper phase classification, the retrieved cloud microphysical/optical properties are consistent with both MFRSR/MWR and CPC retrievals. The sensitivity study suggests that retrieval uncertainty is about 10% for cloud effective radius and cloud liquid water path.

[34] Photon path length distribution inferred from oxygen A-band provides unique information on vertical profiles of scattering and absorption in the atmosphere. As illustrated here and discussed in Min and Clothiaux (2003), the modest resolution measurements of RSS provide two independent pieces of information that constrain particle size distribution in radar retrievals. Combining photon path length information and radar reflectivity yields not only an accurate liquid water path but also vertical distribution of effective radius and liquid water content. This algorithm sought to retrieve vertical distribution of Re, LWC, and optical depth by combining oxygen A-band spectral measurements with radar reflectivity. However, validation is still ongoing to test the robustness of the RSS measurements and radar reflectivity in many different situations covering the variability of cloud properties. With high resolution oxygen A-band measurements, more independent pieces of information can be obtained. Additional independent information may further constrain active radar retrievals to allocate cloud optical properties for multilayer clouds, particularly for mix-phase clouds or ice clouds over a liquid cloud. Such synergetic retrievals could provide accurate and unique information for studying aerosol-cloud interactions and their impacts on climate.

Acknowledgments

[35] This work was supported by U.S. DOE's Atmospheric System Research program (Office of Science, OBER) under contract DE-FG02-03ER63531, by the NSF under contract AGS-1138495, and by the NOAA Educational Partnership Program with Minority Serving Institutions (EPP/MSI) under cooperative agreements NA17AE1625 and NA17AE1623.

Ancillary