Comparisons and analyses of aircraft and satellite observations for wintertime mixed phase clouds

Transcription

1 JOURNAL OF GEOPHYSICAL RESEARCH, VOL. 116,, doi: /2010jd015420, 2011 Comparisons and analyses of aircraft and satellite observations for wintertime mixed phase clouds Yoo Jeong Noh, 1 Curtis J. Seaman, 1,2 Thomas H. Vonder Haar, 1,2 David R. Hudak, 3 and Peter Rodriguez 3 Received 1 December 2010; revised 6 June 2011; accepted 11 July 2011; published 24 September [1] This work presents a study of midlevel, mixed phase clouds using satellite (remote sensing) and aircraft (in situ) observations. In this study, we analyze coincident multisatellite and in situ aircraft measurements of three mixed phase cloud cases during an intensive field experiment (C3VP/CLEX 10) to better understand the microphysics and radiative properties and provide a foundation for the improvement of the satellite retrieval algorithms for these clouds. For the selected cases, various aspects observed from different instruments are presented and compared for these clouds. It is found that many areas in the Moderate Resolution Imaging Spectroradiometer (MODIS) cloud phase product classified as unknown are more appropriately classified as mixed phase based on CloudSat and CALIPSO data as well as C3VP/CLEX 10 aircraft measurements. The aircraft measurements show that a significant amount of supercooled liquid water exists at or near cloud top at very low temperatures for these midlevel, mixed phase clouds, contrary to the assumptions used in the CloudSat retrieval algorithms. The spatial distribution of liquid water content and ice water content and other cloud properties are examined for both the satellite remote sensing and in situ probe measurements. CloudSat and airborne radar reflectivity data are also compared through a structure function analysis. Radiative transfer simulations based on the aircraft and satellite observations indicate the importance of proper assignment of cloud phase within retrieval algorithms and numerical models, which use similar assumptions. Citation: Noh, Y. J., C. J. Seaman, T. H. Vonder Haar, D. R. Hudak, and P. Rodriguez (2011), Comparisons and analyses of aircraft and satellite observations for wintertime mixed phase clouds, J. Geophys. Res., 116,, doi: /2010jd Introduction [2] Clouds are a crucial factor in the study of weather forecasting and global climate. In the Earth s atmosphere, mixed phase clouds in which supercooled liquid water coexists with ice are relatively common [Deeter and Vivekanandan, 2005] and have been observed to contain liquid droplets at temperatures down to 40 C [Cober et al., 2001]. Although understanding of mixed phase clouds with both liquid and ice phase hydrometeors is important for improved satellite retrievals, numerical weather prediction and climate modeling, as well as aviation safety, many details of the microphysical and dynamic processes that determine their formation and dissipation are not fully understood 1 Department of Defense Center for Geosciences/Atmospheric Research, Cooperative Institute for Research in the Atmosphere, Colorado State University, Fort Collins, Colorado, USA. 2 Department of Atmospheric Science, Colorado State University, Fort Collins, Colorado, USA. 3 Cloud Physics and Severe Weather Research Section, Environment Canada, Toronto, Ontario, Canada. Copyright 2011 by the American Geophysical Union /11/2010JD [Shupe et al., 2008; Ou et al., 2009]. In particular, studies of cloud phase composition for mixed phase clouds have been significantly limited by a lack of intensive in situ measurements that can first discriminate between the ice and liquid phases and make direct measurements of these cloud properties. Furthermore, cloud liquid and ice water content (IWC) retrievals from remote sensing instruments are often validated in primarily single phase clouds [e.g., Wang and Sassen, 2002; Hogan et al., 2005; Mace et al., 2005; Comstock et al., 2007; Brunke et al., 2010], leading to questions about the use of these data sources to study mixedphase clouds globally. [3] Midlevel clouds (clouds with bases and tops between 2 and 7 km altitude, respectively) in general, and altocumulus and altostratus specifically, cover approximately 20 25% of the Earth [Warren et al., 1986, 1988; Sassen and Wang, 2008]. Previous field experiments have shown these clouds to exist at temperatures between 0 and 30 C (or below) [Hobbs and Rangno, 1985, 1998; Paltridge et al., 1986; Heymsfield et al., 1991; Fleishauer et al., 2002; Ansmann et al., 2008; Carey et al., 2008], with the majority of observed clouds to be either primarily liquid or mixed phase. Longer term and climatological studies have determined 1of23

2 the fraction of mixed phase clouds to all clouds in this temperature range to be 40 60%, with the fraction of liquid clouds between 30 and 60%, the latter fraction having varying degrees of dependence on temperature depending on the observations included [Korolev et al., 2003;Mazin, 2006; Shupe et al., 2006; Zhang et al., 2010]. [4] The field experiments referenced above indicate the prevalence of midlevel, mixed phase clouds to exist as a layer of supercooled liquid droplets at cloud top out of which ice particles precipitate. Zhang et al. [2010] determined 8% of clouds globally and 34% of midlevel clouds were supercooled liquid topped stratiform clouds above the boundary layer. A similar microphysical structure has been found for Arctic boundary layer clouds [Pinto, 1998; Shupe et al., 2006; de Boer et al., 2008] and other mixed phase cloud systems [Rauber and Tokay, 1991, and references therein], which increases the global average amount of supercooled liquid topped clouds above that determined by Zhang et al. [2010]. However, it is common among numerical weather prediction (NWP) models and global climate models (GCMs) to delineate cloud phase using temperature thresholds that would lead to ice above liquid in clouds that straddle the specified threshold [e.g., Smith, 1990; Bower et al., 1996; Rasch and Kristjánsson, 1998; Zhang et al., 2003; Khairoutdinov and Randall, 2003]. A recent study [Hu et al., 2010] showed that significantly more supercooled water clouds are observed by CALIPSO compared with existing parameterizations and also indicated liquid water content (LWC) values in these clouds are independent of temperature based on combined CALIPSO and Moderate Resolution Imaging Spectroradiometer (MODIS) measurements. [5] In this study, we attempt to better understand the microphysical characteristics and radiative properties of mixed phase altocumulus and altostratus clouds by analyzing detailed aircraft measurements taken during an intensive field campaign that took place during the winter season over the Southern Ontario region of Canada [Barker et al., 2008]. These aircraft observations were collected directly underneath the A Train constellation of satellites [Stephens et al., 2002, 2008] during the joint field campaigns of the Canadian CloudSat/CALIPSO Validation Project (C3VP) and the Tenth Cloud Layer Experiment (CLEX 10). Three cases of midlevel, mixed phase clouds are studied and the microphysical structure and characteristics of the clouds as determined by in situ aircraft measurements are compared with remote sensing retrievals from A Train satellites. The goals of this study are to: (1) characterize the microphysical structure and properties of midlevel, mixed phase clouds observed in a new and highly detailed data set, (2) compare these detailed aircraft data with collocated CloudSat and other A Train satellite observations, and (3) use the results from this work to guide future improvements in model parameterizations as well as satellite retrievals. [6] Validation of CloudSat and/or CALIPSO retrievals of cloud properties is currently an active area of research. Barker et al. [2008] previously examined one case of mixedphase boundary layer clouds with overrunning cirrus from C3VP/CLEX 10 using a similar methodology. Numerous other validation studies of CloudSat and/or CALIPSO data products have recently been performed for various cloud types and various locations around the globe [e.g., Bouniol et al., 2008; de Boer et al., 2008; Gayet et al., 2009; Brunke et al., 2010; Protat et al., 2010]. To the authors knowledge, the present study is the first to compare CloudSat retrievals and aircraft observations for midlevel, mixed phase clouds. [7] This paper is presented as follows: section 2 describes the C3VP/CLEX 10 field campaign and the data collected; section 3 compares the aircraft observations with the satellite retrievals; section 4 examines the implications of the assumptions used in CloudSat retrievals on radiative heating and cooling rates and fluxes; section 5 presents a discussion of the results and conclusions as well as suggested future work. 2. Data 2.1. C3VP/CLEX 10 Overview [8] The Cloud Layer Experiment (CLEX) is an ongoing effort for the study of nonorographic, midlevel, mixedphase clouds [Fleishauer et al., 2002; Carey et al., 2008] sponsored by the United States Department of Defense (DOD) Center for Geosciences/Atmospheric Research (CG/ AR). These clouds are important for a variety of aviation and military applications [Fleishauer et al., 2002] and are difficult to forecast [Zhang et al., 2005; Illingworth et al., 2007; Seaman et al., 2010]. The primary goal of CLEX is to improve forecasting of these clouds through the use and analysis of in situ and remote sensing observations. [9] The Canadian CloudSat/CALIPSO Validation Project (C3VP) was a joint field experiment sponsored by the DOD, NASA (National Aeronautics and Space Administration), CSA (Canadian Space Agency) and MSC (Meteorological Service of Canada) that took place during the autumn and winter of The primary goals of C3VP are to independently verify CloudSat and CALIPSO data products for winter season and mixed phase cloud and precipitation systems and contribute to the development of midlatitude and high latitude remote sensing products (e.g., snowfall retrievals). [10] These two field experiments (C3VP/CLEX 10) worked together during four day intensive observation periods (IOPs) between 31 October 2006 and 1 March On prescheduled flight days, the Canadian National Research Council s (NRC) Convair 580 research aircraft was flown directly underneath the A Train satellite constellation typically during the afternoon (ascending) overpasses within range of Ottawa, Ontario, Canada. Several nighttime (descending) overpasses were flown as well as weatheronly flights targeting interesting weather conditions away from A Train overpasses. An instrumented ground site was maintained near Egbert, Ontario, Canada at the Center for Atmospheric Research Experiments (CARE) [Hudak et al., 2006]. Twenty eight flights were conducted totaling 107 h of flight time. However, only flights targeting midlevel, mixedphase clouds with simultaneous A Train overpasses are presented here. [11] The Convair 580 aircraft flew along a targeted 60 nautical mile segment of the CloudSat ground track as the A Train satellites passed overhead. The aircraft may be above, below or in cloud at the time of the overpass (or in clear sky conditions for flights selected for CALIPSO aerosol retrieval validation), but as close to that line as possible. 2of23

3 Before and after each overpass, the aircraft would fly in and around relevant clouds within the target region, typically remaining along the CloudSat ground track. In the specific cases included here, the subaircraft point was m away from the subsatellite point of CloudSat at the time of the overpass, making the aircraft well within the field of view of CloudSat [Stephens et al., 2002]. In fact, there were several cases during C3VP/CLEX 10 where the presence of the aircraft and its instrumentation affected CloudSat s observed reflectivity values. Given the airspeed of the Convair 580 of 100 m s 1, the aircraft required 18 min to fly 60 nautical miles, a distance which CloudSat would travel in 18 s Aircraft Observations [12] The Convair 580 aircraft was instrumented to be a flying CloudSat/CALIPSO simulator. The Aerosol Imaging Airborne Lidar (AERIAL) [Strawbridge and Snyder, 2004] flown was modified (adding the 532 nm beam for this experiment) to provide observations similar to CALIPSO with upward and downward beams at 532 nm and 1064 nm. Upward and downward looking Ka and W band radars were included [Barker et al., 2008], although the recently acquired W band radar was not able to be installed until after the first two IOPs. The upward looking W band radar was turned off within several minutes of each overpass to avoid signal contamination with CloudSat. However, reflection of the downward looking W band radar aboard the aircraft was still captured by the satellite in several cases. Unfortunately, the W band radar was not aboard the aircraft for two of the three case studies presented here. In the analysis presented in section 2.3, only the Ka band radar was used because of this, as well as delays in the processing of the W band data, and for consistency. The Ka band radar (34.66 GHz) with fixed up pointing and down pointing 30.6 cm diameter antennae was mounted in the Convair 580 aircraft wing roots. The radar system has a vertical resolution of 25 m, a beam width of 2, and a sensitivity of approximately 30 dbz at 1 km. The radar is estimated to be accurate to within ± 2 db. Further calibration information and specifications are given by Jordan et al. [2000] and Barker et al. [2008]. [13] In addition to the remote sensing instruments, the Convair 580 was equipped with an extensive array of microphysical probes. The relevant instruments used here are briefly described below. [14] Liquid water content (LWC) was observed using a King hot wire probe [King et al., 1978], which has been estimated to have errors less than 10% [Barker et al., 2008]. The Nevzorov Total Condensed Water probe [Korolev et al., 1998] was used to measure the total water content (TWC) and ice water content (IWC). The Nevzorov probe has an estimated accuracy of 10 20% and a sensitivity of g m 3 [Korolev et al., 1998; Korolev and Strapp, 2002]. The Nevzorov probe was modified to reduce the likelihood of ice particles bouncing and/or shattering out of the cone, which was observed in wind tunnel tests and which leads to an underestimation of IWC [Korolev et al., 2008]. Values of IWC were derived from the Nevzorov probe and King probe following the procedure described in Korolev et al. [2003], which was obtained by subtracting the King liquid water content from the Nevzorov probe total water content. The King and Nevzorov probes showed good agreement in clouds where little to no ice was detected (Figure 1). A Rosemount Icing Detection probe was also used to detect the presence of supercooled liquid water [Cober et al., 2001]. [15] Two Particle Measurement System (PMS) Forward Scattering Spectrometer Probes (FSSP 100) [Knollenberg, 1981] were used to measure droplet size distributions and estimate the corresponding effective radius values. These probes were configured to count particles in 15 different size bins in the nominal range between 2.5 and 47.5 mm. Regular maintenance and calibration were performed to minimize droplet sizing errors. Dead time losses and coincidence errors were reduced during processing following the procedure of Baumgardner et al. [1985]. [16] Independent measurements of IWC were derived from analysis of PMS optical array probes including 2D C and 2D P[Knollenberg, 1981] and 2D G probes [Reuter and Bakan, 1998]. These probes collectively image the shadow cast by particles in the size range mm (2D C), mm (2D P) and mm (2D G), with 25, 200, and 10 mm resolution, respectively. Three different procedures for deriving IWC from the two dimensional images are used [Locatelli and Hobbs, 1974; Cunningham, 1978; Heymsfield et al., 2002]. These procedures will hereafter be referred to as the Locatelli, Cunningham, and Heymsfield methods, respectively. In each procedure, the collection of particle images from all three probes were summed together and converted to particle size distributions following the method of Korolev and Sussman [2000]. Incomplete images of particles (particles straddling the boundaries of an instrument s field of view) were reconstructed and particle habits were classified also following the algorithm of Korolev and Sussman [2000]. The primary particle habit (sphere, needle/column, dendrite or irregular) was then used to calculate the IWC from the size distributions using the three different sets of ice massdimensional relationships in the Locatelli, Cunningham and Heymsfield methods. [17] Ice particle shattering is known to impact the size distributions produced from the optical array probe imagery [Korolev and Isaac, 2005; Field et al., 2006] and the FSSP [Field et al., 2003]. However, recent studies have shown no significant impact by ice particle shattering on FSSP data in mixed phase clouds [Korolev et al., 2011; McFarquhar et al., 2011]. Estimates of the IWC error introduced by ice particle shattering in the 2 D probes have ranged from 20 to 30% [Field et al., 2006] to as much as a factor of 2 [Korolev et al., 2011]. Our study corrects for ice particle shattering using the methodology of Boudala and Isaac [2009]. Images of particles with maximum diameters between 100 mm and 775 mm from the 2 D C probe were included with particles larger than 775 mm from the 2D P probe. For the cases presented here, no 2D G particle images were available. During processing of the 2 D images, ice particle shattering was accounted for using the interarrival time correction of Field et al. [2006], as well as excluding images of particles smaller than 100 mm in diameter. These corrections were found to change the IWC by an average of 7% when both the 2D C and 2D P probes were used together, when tested on other C3VP cases [Boudala and Isaac, 2009]. [18] In addition to the three methods for converting particle images to IWC values (i.e., Heymsfield, Cunningham 3of23

4 Figure 1. Comparison between water contents observed by the King LWC and Nevzorov TWC probes during the flight of 19 February During this flight, no significant amounts of ice were found by any of the other aircraft instruments. and Locatelli), two different analysis modes have been applied to the optical array probe imagery prior to the use of each method. The Center in (CIN) analysis mode includes all particles where the assumed center falls within the field of view of the instrument. The Double Edge Element (DEE) analysis mode only includes images where the edges of the particle are fully within the field of view of the instrument. The CIN mode is expected to be more accurate in clouds with larger ice particles, as fewer particles (particularly larger particles) are rejected, while the DEE mode is expected to be more accurate for small ice particles as there are fewer assumptions made about particle size and shape. For greater accuracy, particle images from entire flight segments ( min. of flight data) are collected from the three probes and analyzed to compile robust size distributions, from which flight segment averaged IWC values are calculated. [19] An analysis of 203 flight segments from C3VP/ CLEX 10 revealed that IWC values derived using the DEE mode were typically 30 40% smaller than IWC values derived using the CIN mode, independent of the method. In comparison, using the CIN mode for each method, Heymsfield IWC values were typically 50 70% smaller than the Cunningham and 60 80% smaller than the Locatelli IWC values. Thus, the choice of method for calculating IWC from 2 D images has a larger impact on the resultant IWC values than the choice of analysis mode. DEE IWC values will not be shown or discussed further in this work. For the cases presented here, differences in IWC between the Heymsfield and Locatelli methods are typically a factor of 2 3, while previous studies indicate errors of up to a factor of two using any one analysis method [Jensen et al., 2009; Korolev et al., 2011]. Wind tunnel tests have also shown that various hotwire probe measurements of IWC vary by a factor of 2 or more, depending on probe geometry and the degree of ice particle shattering or bouncing [Strapp et al., 2005]. The various IWC values are presented in this work to provide an estimate of the errors inherent in measuring IWC from aircraft instruments. It is an active, ongoing area of future research to better quantify the amount and primary sources of these errors and to identify the best solutions to minimize them Satellite Observations [20] The in situ and remote sensing observations from the aircraft are compared with remote sensing observations (i.e., operational data products) from several satellite instruments in the A Train constellation, primarily the CloudSat cloud 4of23

5 profiling radar (CPR) and the MODIS on the Aqua satellite. A brief overview of these instruments and products follows. [21] The CloudSat CPR is a near nadir looking (0.16 forward) W band radar (94 GHz) with a field of view of 1.3 km in the across track direction and 1.7 km in the along track direction. During the time it takes for CloudSat to observe one profile of the atmosphere, the subsatellite point has traveled 1.1 km along the Earth s surface. This leads to overlap between adjacent profiles. Vertical resolution of CPR range gates is 480 m, which oversampling increases to 240 m. Additional details of the CloudSat CPR are given by Stephens et al. [2002, 2008] and Tanelli et al. [2008]. [22] In this work, the following CloudSat data products (release version R04) are used: 2B GEOPROF (radar reflectivity factor and cloud mask), 2B CWC RO (radar only cloud water content and effective radius retrievals), 2B FLXHR (fluxes and heating rates), 2B CLDCLASS (cloud classification) and ECMWF AUX (auxiliary temperature analysis data from the European Centre for Medium Range Weather Forecasting). The cloud classification product identifies clouds as cirrus (Ci), altostratus (As), altocumulus (Ac), stratus (St), stratocumulus (Sc), cumulus (Cu), nimbostratus (Ns) or deep convective clouds (DeepC) (Figures 2 4). Documentation for each of these data products is available online at A second cloud water content retrieval (2B CWC RVOD) is available, which uses a visible optical depth retrieval to further constrain cloud water content values. However, visible optical depth retrieval is difficult or impossible to perform in many cases due to the complexity of the targets and the necessary simplifying assumptions [Wood, 2008]. When no optical depth information is available, the 2B CWC RVOD retrieval simply reverts to the 2B CWC RO retrieval. Thus, for consistency, only the 2B CWC RO retrieval is used. For the cases presented here, the 2B CWC RVOD retrieval differed from the 2B CWC RO retrieval by less than 5%. More comparisons of these retrievals are made by Wu et al. [2009] and Protat et al. [2010]. [23] CALIOP (Cloud Aerosol Lidar with Orthogonal Polarization) Level 1 data from the CALIPSO (Cloud Aerosol Lidar and Infrared Pathfinder Satellite Observation) satellite are also used to provide direct observations of thin clouds and aerosols from space with two channels of 532 nm and 1064 nm, which are sensitive to thin liquid cloud layers [Winker et al., 2009]. While CALIPSO is ideally suited to the detection of supercooled liquid layers at cloud top [Zhang et al., 2010], strong signal attenuation prevents the retrieval of LWC or IWC values in the cases presented here. Thus, only the detection of cloud top liquid layers by CALIPSO is validated. [24] MODIS is a 36 channel scanning spectroradiometer with visible, near infrared and infrared (IR) channels [King et al., 2003]. Each scan covers a swath width of 2330 km with horizontal resolutions of 250, 500 and 1000 m, depending on the channel. In this work, the MODIS IR cloud phase retrieval product of Collection 5 MYD06 Level 2 data are used [Menzel et al., 2010]. Brightness temperature differences between the 8.5 mm and 11 mm channels are compared with brightness temperatures from 11 mm to determine dominant cloud phase (liquid, ice or mixed phase), exploiting differences in absorption by liquid water and ice at these wavelengths. This retrieval is performed after first averaging the observed radiances to a 5 km 5 km grid. MODIS true color images from Level 1B and geolocation data are shown as well, which were created by using the method of Gumley et al. [2010]. 3. Comparison Between Aircraft and Satellite Observations [25] Simultaneous observations of the mixed phase clouds were made by the A Train satellites together with the Canadian National Research Council s (NRC) Convair 580 research aircraft. Three cases presented here include 31 October 2006, 5 November 2006, and 25 February 2007 from C3VP/CLEX 10. [26] Figures 2 4 represent the multisatellite observations from CloudSat, CALIPSO, and Aqua MODIS with the CloudSat ground track and aircraft flight path for each case. Included are MODIS IR (12mm) and true color (using bands 1, 4, and 3) images, the CloudSat 2B GEOPROF CPR reflectivity and 2B CLDCLASS cloud classification, and CALIPSO total attenuated backscatter and depolarization ratio at 532 nm. MODIS images provide useful information of the horizontal cloudy scenes for each case, and two active remote sensors (CloudSat CPR and CALIPSO CALIOP) show the characteristics of the vertical structure of these clouds that have supercooled liquid water near cloud top (more details in sections ). The Convair 580 research aircraft was flown along the CloudSat ground track at the time of the A Train overpass within each target region. All times given will be UTC Case 1: 31 October 2006 [27] On 31 October 2006, the Convair 580 was flown through a cloud that the CloudSat algorithm correctly identified as altostratus (Figure 2). This cloudiness was associated with a cold frontal passage over the region. During the flight, altostratus cloud was mainly observed with some cirrus above and stratocumulus below, which were also correctly identified by CloudSat. The MODIS cloud phase algorithm, however, was largely unable to classify the cloud within the target region, marking it as uncertain, with some areas correctly labeled as mixed phase (Figure 5a). It is suggested that the presence of patchy cirrus clouds above the altostratus layer prevented the algorithm from correctly identifying the cloud system as mixed phase. [28] Between 18:00:51 and 18:15:40 UTC the aircraft flew a level flight leg at 6.3 km (near cloud top) along the CloudSat ground track in a southward direction (Figure 6a). CloudSat passed over the aircraft at 18:10:30 UTC when the aircraft was 214 m away from the subsatellite point. Figure 7a shows the radar reflectivity factor (dbz e ) from CloudSat and the Ka band radar aboard the Convair 580 for the 8 min period surrounding the overpass (4 min before and after the overpass) at each instruments native resolution. The primary differences are the extents of the cirrus clouds above and stratocumulus clouds below the midlevel cloud layer. Given the resolution and wavelength differences between the two radars, the CloudSat view of the altostratus layer appears quite similar to the Ka band radar. [29] To quantify the differences between the radars, we utilize the second order structure function (STR) to statisti- 5of23

6 Figure 2. Multisatellite observations of the 31 October 2006 case. (top) MODIS IR (12 mm) and true color (bands 1, 4, and 3) images with the CloudSat ground track (blue) and aircraft flight path (red). (middle) CloudSat CPR reflectivity and cloud classification along the long segment are shown with the target region (red vertical lines) at the time of the A Train overpass. (bottom) CALIPSO 532 nm total attenuated backscatters and depolarization ratios. cally characterize the radar reflectivity variability observed from CloudSat and the aircraft. For a variable A, STR is defined in equation (1) with the mean squared difference over all the combinations of the paired measurements as a function of their separation distance y [Hillger and Vonder Haar, 1979, 1988; Barker et al., 2008]. That is, 6 of 23 STRð Þ ¼ 1 X ½ AðiÞ Að jþ 2 ; N ð Þ i<j ð1þ

7 Figure 3. Same as Figure 2 but for the 5 November 2006 case. where N is the number of pairs for a given lag y between a position, i and j. It is assumed that the statistical relationship between paired measurements should be independent of translation of the measurement locations, which allows a large statistical base to be accumulated from a relatively small number of measurements [Hillger and Vonder Haar, 1988]. The measurements are grouped into each bin by discrete time or space sampling, in which STR is given by the average over all the pairs separated by the respective lag [de Vries et al., 2005]. For this structure function analysis, the airborne radar (Ka band) reflectivities were averaged to the CloudSat ground foot print resolution. Although the determination of the structure function values is based on separation distances and satellite measurement locations on the Earth are not on an equally spaced grid, the CloudSat and aircraft tracks are almost identical and are compared 7of23

8 Figure 4. Same as Figure 2 but for the 25 February 2007 case. over linear latitude increments of less than 1 latitude, which should be sufficient to use 1 D latitude separation (degree) as a spatial lag. [30] Figure 8 shows radar reflectivities vertically averaged in the 500 m layer from 5.5 km to 6.0 km and corresponding STRs in terms of separation latitude degree for the radar data shown in Figure 7a. The 500 m mean reflectivity values from both instruments are in good agreement in terms of trend, and the STRs also similarly tend to increase with separation, although the Convair s reflectivity is slightly higher (the mean bias is about 3.7 dbz e ) and its STR shows greater fluctuations in a short distance (within 0.12 ) than CPR s values. The smaller fluctuation of the CloudSat measurements is consistent with the finding of Barker et al. [2008] who performed a similar analysis for the 20 February 2007 case. As they indicated, the smoothness of CloudSat s 8of23

9 Figure 5. MODIS Level 2 IR Cloud Phase products (MYD06) for (a) 31 October 2006, (b) 5 November 2006, and (c) 25 February 2007, with the target regions indicated by the green circles. signals may arise from CloudSat s overlapping of pixels along the track. It should be noted that the airborne W band radar data were not available for this case. It is expected to have the quantitative difference between W band and Ka band radars due to their different characteristics such as non Rayleigh effects in the W band, which would cause smaller reflectivity values than the Ka band (depending on the distribution and particle sizes of hydrometeors within the radar bin volume) and also due to where the cloud layer is positioned in the volume and its actual reflectivity (greater than each minimum detectable signal). [31] The 2B CWC RO retrieval for the 31 October 2006 case is shown in Figure 9. The aircraft flight path for the 8 min flight segment during the overpass is also shown along with the boundaries of the CloudSat cloud mask and the temperature from the ECMWF AUX product. Figure 9 shows where the 2B GEOPROF cloud mask identifies cloudy pixels based on the observed reflectivity but the 2B CWC RO retrievals produce LWC and/or IWC < = 0 (due to a failure to converge or to LWC or IWC values below the limit of CloudSat s sensitivity). According to the CloudSat water content retrieval, the altostratus layer contains no liquid water above 6 km (Figure 9a). This is directly contradicted by the microphysical probes aboard the aircraft. Figure 10 shows the time series of LWC from the King hot wire probe and IWC derived from the Nevzorov probe during this flight segment, indicating the cloud contained significant amounts of liquid water at cloud top. This is confirmed by the signal from the Rosemount Icing Detector (Figure 10). Note also that the temperature at flight level was approximately 23 C. [32] As demonstrated in Figure 9, the 2B CWC RO algorithm uses temperature to distinguish cloud phase using simple temperature thresholds similar to those found in many operational weather and climate models [Austin, 2007]. Cloud particles at temperatures less than 20 C are assumed to be ice and cloud particles at temperatures above 0 C are assumed to be liquid. In between these thresholds, the LWC and IWC are scaled linearly based on the temperature. In fact, the cloud water content retrieval is the combination of two retrievals: one assuming the entire cloud is liquid; the other assuming the entire cloud is ice. The linear scaling is such that, at 5 C, 25% of the retrieved IWC and 75% of the retrieved LWC make up 100% of the total cloud water content, and at 10 C, 50% of the retrieved IWC and 50% of the retrieved LWC make up the total cloud water content, etc. [33] Between 18:22:31 and 18:42:30 UTC, the Convair 580 was flown in a spiral descent through the midlevel and lower level cloud layers (Figure 6b). The comparison between the aircraft and CloudSat vertical profiles of LWC and IWC are shown in Figure 11. Here only CloudSat water content profiles are used if they fall within the horizontal extent of the Convair flight path shown in Figure 6b. It should be noted that the Convair observed a different cloud at 3 km than the cloud that was observed by CloudSat approximately 30 min earlier (see Figure 9), which had advected away from the aircraft. As the altostratus layer between 5 and 6 km did not change significantly during this period, observations of this cloud layer may be more directly compared. This comparison reveals that the magnitudes of the LWC and IWC values fall within the range of values from the various measurements in the altostratus layer. The maximum LWC value of 0.18 g m 3 from CloudSat compares with a maximum of 0.17 g m 3 from the King LWC probe, and the CloudSat IWC values fall within the range of values from the Nevzorov probe and the different procedures used to derive IWC from the 2 D optical array probes. The 2B CWC RO algorithm includes an uncertainty estimate [Austin, 2007], which is 45 55% for LWC and 65 75% for IWC in the middle cloud layer. Note also that each CloudSat point in Figure 11 represents a volume of approximately 1.1 km 1.1 km 240 m, while the King and Nevzorov probe data points represent 1 s ( 100 m) of flight data. The Heymsfield, Locatelli and Cunningham IWC values cover the entire flight segment, including both (middle and lower) cloud layers. [34] There are, however, significant qualitative differences due to CloudSat s vertical resolution and its use of temperature to distinguish cloud phase. The height of maximum LWC from CloudSat ( 5.1 km) is approximately 1 km lower than was observed by the aircraft. CloudSat also distributed the liquid water over a 2 km thick layer, whereas the aircraft observed liquid water only in a 0.5 km thick layer centered near 6 km. 9 of 23

10 Figure 6. Aircraft flight legs (red) along the CloudSat ground tracks (black solid line) over the target regions (cyan circles) for (a) 18:06:30 18:14:30 UTC on 31 October 2006, (b) 18:22:31 18:42:30 UTC on 31 October 2006, (c) 18:24:33 18:32:33 UTC on 5 November 2006, (d) 18:37:14 18:48:33 UTC on 5 November 2006, (e) 18:25:05 18:33:05 UTC on 25 February 2007, and (f) 18:32:51 18:45:10 UTC on 25 February Case 2: 5 November 2006 [35] On 5 November 2006, the Convair 580 was flown into a broad area of altocumulus clouds near the instrumented ground site at CARE. Altocumulus clouds were observed at CARE continuously for more than 10 h. At the time of the CloudSat overpass (18:28:33 UTC) the aircraft was flying a level flight leg in a southward direction along the CloudSat ground track at 6.2 km, above the altocumulus layer and below a few cirrus clouds (Figure 6c). CloudSat correctly identified the altocumulus and cirrus (Figure 3), although the MODIS cloud phase algorithm was, once again, uncertain (Figure 5b) due, most likely, to the presence of the cirrus. CALIPSO clearly detected a supercooled liquid water layer at cloud top and cirrus above the cloud layer (Figure 3). 10 of 23

11 Figure 7. Radar reflectivity cross sections measured from (top) CloudSat (dotted line: aircraft flight track) and (bottom) airborne Ka band radars for (a) 31 October 2006, (b) 5 November 2006, and (c) 25 February 2007 within 4 min of the CloudSat overpass. [36] Figure 7b shows the CloudSat and Convair 580 radar reflectivity values observed during the eight minutes surrounding the overpass (18:24:33 18:32:33 UTC). In this case, the aircraft itself was observed by CloudSat as the relatively high reflectivity value at 6 km and latitude. In this case, the aircraft itself was observed by CloudSat as the relatively high reflectivity value at 6 km and latitude. It is noted that the Ka band is missing the virga below cloud base (below 3.5 km) which was weakly detected from CPR. It may be because the CPR beam width is larger than the Ka band radar so it may have sampled particles the Ka band did not see. A slight mismatch (even 100 m) in the aircraft relative to the CloudSat track might also explain the discrepancy if the virga was widely scattered. The reflectivity (column averaged between 3.5 and 4.0 km) and structure function values for both radars have also been compared (not shown here). The mean reflectivity values and the STRs are again similar between the two radars. The magnitudes of both STRs are almost constant within a short distance (<0.11 degree), which is indicative that this midlevel, mixed phase cloud layer did not significantly change in time and space as observed during the flight. [37] Between 18:37:14 and 18:48:33 UTC, the aircraft flew a slantwise descent through the altocumulus layer from which the vertical profiles of LWC and IWC are compared. The 2B CWC RO retrieval for the 5 November 2006 case is shown in Figure 12, along with the cloud mask from 2B GEOPROF, temperature from ECMWF AUX, and aircraft flight path during this descent. Once again, CloudSat water content profiles are compared with the aircraft probes if they fall within the latitude range of the Convair flight path shown in Figure 6d. This comparison with the aircraft probes is shown in Figure 13. Unfortunately, the 2 D optical array probes were not operating for approximately 100 min during this flight, including the overpass and subsequent descent through the cloud. However, the magnitudes of the IWC values from CloudSat fall within the range observed by the Nevzorov probe. The retrieved uncertainties in the Figure 8. Comparisons of CloudSat CPR and airborne radar measurements for the 31 October 2006 case: (left) the column averaged reflectivity values (dbz e ) between 5.5 km and 6.0 km and (right) their second order structure functions (lag: latitude degree). 11 of 23

12 Figure 9. CloudSat (a) LWC retrieval and (b) IWC retrieval from 31 October Gray shaded areas indicate cloud boundaries from the CloudSat cloud mask. Red dotted lines indicate ECMWF temperatures in Celsius. The black and white dashed line indicates the location of the aircraft between 18:06:30 and 18:14:30 UTC. Note the differences in the color scales. CloudSat values are 45% for LWC and 75% for IWC. The maximum LWC from the King probe of 0.29 g m 3 compares with a maximum LWC from CloudSat of 0.18 g m 3. Using a 10 s average LWC from the aircraft to more closely represent CloudSat s spatial resolution yields a maximum LWC of 0.23 g m 3 from the King probe. [38] As with the 31 October case, the aircraft observed a 500 m thick supercooled liquid layer based on the microphysical probes (Figure 13), while the retrievals of CloudSat have a cloud thickness of km (Figures 12 and 13). It is likely that the CloudSat radar observed precipitating ice virga, which would dominate the cloud property retrievals. W band radars are more sensitive to the larger ice particles comprising virga than the smaller liquid droplets found at cloud top, which would explain the much lower cloud base in the CloudSat retrieval. Virga was observed as much as 2 km below the supercooled liquid layer by the groundbased lidar at CARE ( 50 km to the southwest) during the flight. If this is true, CloudSat incorrectly identified precipitating ice virga as liquid (Figure 13). The Nevzorov probe detected no IWC below 4.2 km. It is inferred that a few very large ice particles in virga (as the smaller particles would have sublimated) would give a significant radar signal but still have very low IWC, perhaps below the sensi- 12 of 23

13 Figure 10. Time series of aircraft probe data within 4 min of the CloudSat overpass on 31 October (top) One second LWC values are plotted as dots; the solid gray line represents 10 s averaged LWC values; the solid black line represents 10 s averaged IWC values; and the dashed line represents altitude with scale given on the right ordinate. Given the flight speed of the Convair 580 of 100 m s 1, 10 s data represent 1 km in the horizontal, approximating the resolution of CloudSat. (bottom) Rosemount Icing Detector signal (solid line, left ordinate) and temperature (dashed line, right ordinate). tivity of the Nevzorov probe, although the 2 D optical array probes were inoperative during this portion of the flight so this cannot be fully proven. The King probe and Rosemount Icing Detector detected no liquid water below 4.2 km, implying any virga present was ice Case 3: 25 February 2007 [39] On 25 February 2007 a large low pressure system over the central U.S. continued to move slowly toward the northeast, near southern Ontario. Ahead of the system, a band of cirrus and a large area of midlevel cloud cover were observed over the target area. During the flight, an altostratus layer was sampled, which was once again consistent with the CloudSat cloud classification over the C3VP/CLEX10 target area (Figure 4). CALIPSO also observed supercooled liquid water layers near cloud top as shown Figure 4. For all three cases, the aircraft observations showed that CloudSat cloud classification products are in quite good agreement with the in situ observations. In this case, however, the MODIS algorithm incorrectly identified clouds in the target region as ice (Figure 5c) when no cirrus was detected by CloudSat, CALIPSO, or the aircraft. [40] The aircraft during this overpass was clearly detected by CloudSat. Reflected radiation from the downward looking W band radar aboard the aircraft contaminated five CloudSat profiles, which have been removed during this analysis (Figure 7c). The overpass occurred at 18:29:05 UTC with the aircraft flying northward along the CloudSat ground track (Figure 4) within 150 m of the subsatellite point (Figure 6e). The signal from the Ka band radar aboard the Convair 580 is qualitatively similar to CloudSat for the remaining profiles (Figure 7c) and the STRs are, on the whole, in great agreement 13 of 23

14 Figure 11. Comparison between the aircraft and CloudSat vertical profiles of (left) LWC and (right) IWC (log scale) for 31 October Aircraft data shown include 1 s average LWC from the King probe, 1 s average IWC from the Nevzorov probe, and three methods for calculating IWC from the 2 D optical array probes (flight segment average; see text for details). CloudSat data include all profiles that fall within the latitude range of the aircraft during the flight segment. (not shown here), although the aircraft radar is more sensitive to smaller spatial structures ( 0.06 or less separation). [41] Of the three midlevel clouds examined here, the 25 February 2007 case had the coldest cloud top at 28 C. It is no surprise then that the 2B CWC RO retrieval assumed this cloud to be almost entirely ice, except near cloud base (Figure 14). The CloudSat water contents are compared with the aircraft observations taken during a spiral descent between 18:32:51 and 18:45:10 UTC (Figure 15), which reveal significant amounts of liquid water at cloud top. In this case, the King probe observed liquid water over a 600 m thick layer, peaking at cloud top, with the Nevzorov probe observing the highest IWCs at the base of the liquid layer and some ice throughout. It is noted that the CloudSat IWC values fall within the range of IWC determined by 2 D optical array probes using the Heymsfield, Locatelli and Cunningham methods. The uncertainties in the CloudSat IWC values in this case were 80% Summary and Additional Comparisons [42] For each of the three cases examined above, we have compared aircraft microphysical and remote sensing observations with CloudSat and other A Train satellite observations and retrieval products. The radar data from the aircraft were compared with CloudSat within 4 min of the CloudSat overpass, in each case with the aircraft within CloudSat s field of view. Vertical profiles of LWC and IWC from hotwire probes, along with flight segment average IWC analyses from the 2 D optical array probes were compared with spatially matched CloudSat profiles from the aircraft descent closest in time to the CloudSat overpass, typically min after the overpass. In each case, CALIPSO detected a supercooled liquid layer at cloud top in agreement with the hot wire probes on the aircraft. [43] Despite differences in resolution and radiative transfer between the airborne Ka band radar and the CloudSat CPR, the radar data generally showed similar trends. The primary differences are due to the fact that the Ka band better resolves small scale cloud structures. The cloud droplet size and distribution in the radar bin volume also affects radar detectability. It is well known that both radars are much more sensitive to the larger ice particles than the smaller supercooled liquid droplets typically found in these clouds. [44] It was found, however, that the 2B CWC RO algorithm produced LWC and IWC values that were generally within the range of values presented in the aircraft measurements. The primary difference came from the assumption within the algorithm that temperature is useful for determining cloud phase. The aircraft observations of this study and many previous studies [Hobbs and Rangno, 1985, 1998; Paltridge et al., 1986; Heymsfield et al., 1991; Fleishauer et al., 2002; Ansmann et al., 2008; Carey et al., 2008] show significant amounts of liquid water at or near cloud top at temperatures down to about 30 C in altocumulus and altostratus clouds and liquid water has been observed at temperatures as low as 40 C [Cober et al., 2001]. Two of the three cases presented here show CloudSat to be in error in assuming the clouds are composed entirely of ice at cloud top, simply because cloud top temperatures are below 20 C. The third case had liquid water up to cloud top only because 14 of 23

15 Figure 12. Same as Figure 9 except for the 5 November 2006 case. the cloud top temperature was warmer than 20 C. The temperature information (ECMWF AUX) used by CloudSat is generally within 1 2 C of the aircraft observations, although the ECMWF product has some difficulty diagnosing cloud top temperature inversions (Figure 16). [45] The 2B CWC RO algorithm also retrieves cloud particle effective radius, which was observed on the aircraft with the FSSP probes. Figure 17 shows the comparison between the CloudSat and FSSP values for the three cases discussed previously. Since CloudSat often incorrectly assumed the clouds were ice when supercooled liquid water was present, we use the CloudSat liquid only effective radius retrieval, which assumes the cloud is entirely liquid. We also assume that most ice particles will quickly grow out of the FSSP s upper size limit (47.5 mm) in these mixed phase environments, making the effective radius measurements of the instrument dominated by liquid droplets. The effective radius measurements agree to a great extent when CloudSat correctly assumed liquid was present. This may be due to the a priori assumption in the CloudSat retrieval of a liquid effective radius of 10 mm, which is close to the FSSP values. 4. Radiative Fluxes and Heating Rates [46] As discussed in section 3, CloudSat uses temperature to distinguish cloud phase. Clouds colder than 20 C are assumed to be composed only of ice particles. In contrast, this study and many field experiment studies have shown 15 of 23

16 Figure 13. Same as Figure 11, except for the 5 November 2006 case. Analysis of IWC values from the 2 D optical array probes was unable to be performed due to inoperative instruments. that a significant percentage of midlevel clouds contain significant amounts of supercooled liquid droplets at or near cloud top, even at temperatures as low as 31 C (or below). Many previous studies have also indicated the importance of accurate microphysical information in radiative transfer calculations [e.g., Sun and Shine, 1995; McFarquhar and Cober, 2004; Niu et al., 2008]. In section 4, we explore how radiative fluxes and heating rates are affected by the phase assumptions used by CloudSat. [47] Given the cloud phase assumptions used in the water content retrievals, there is a possibility that the 2B FLXHR products would have significant errors in these cases. To test this, radiative transfer simulations were performed with the stand alone version of the radiative transfer model used by CloudSat, described by Stephens et al. [2001] and Gabriel et al. [2001], known locally as BUGSrad. In these experiments, the radiative transfer model is initialized with the same input profiles of temperature, pressure, ozone, etc., from the ECMWF AUX product that are used in 2B FLXHR [L Ecuyer, 2007], except that the LWC and IWC profiles from the aircraft data are substituted for the CloudSat profiles. For comparison, the CloudSat/ECMWF profiles have been averaged over the horizontal extent of the aircraft data (Figures 6d and 6f). The 2B FLXHR algorithm uses default values of 10 mm and 30 mm for liquid and ice effective radius, respectively, when the effective radius retrieval fails to converge, and those default values are used here. The aircraft profiles of LWC (from the King probe) and IWC (from the Nevzorov probe) are averaged over 240 m vertical layers to match CloudSat s vertical resolution. The Nevzorov derived IWC was typically the smallest IWC of those presented in section 3. Therefore, additional simulations were performed using the IWC from the 2 D probes (where available) using the Locatelli method, which produced the largest IWC values, to estimate the range in uncertainties in the fluxes and heating rates given the range of values in the IWC measurements. [48] Radiative transfer simulations were performed on the 5 November 2006 and 25 February 2007 cases. The results are shown in Figure 18. The 31 October 2006 case was excluded due to the aircraft sampling a different low level cloud, and not sampling the above cirrus cloud that CloudSat observed. [49] For the 5 November 2006 case, recall from section 3 that the CloudSat reflectivity values were dominated by precipitation sized ice virga that was assumed to be liquid in the 2B CWC RO retrieval products based on temperature. Thus, a much thicker layer of liquid water is found from the CloudSat retrieval. Just above the layer of maximum liquid water, we can see the longwave radiational cooling of 23 K d and shortwave radiational heating of +5 K d (Figure 18a). The differences in the LWC profiles (Figure 13) lead to increased shortwave (SW) heating and longwave (LW) cooling at cloud top using the aircraft data (Figure 18a). As mentioned about this case, CloudSat observed a deep layer of virga, likely misidentified as liquid. As a result, the CloudSat flux profile shows a greatly reduced downwelling SW flux at the surface and an increased upwelling SW flux toward space (Figure 18b). The lower cloud or precipitation base in CloudSat (not seen by the aircraft) also results in an increase in downwelling LW flux compared with the aircraft (Figure 18b) as the primary emitting layer was lower, where temperatures were warmer. The cloud base differences also account for the significant differences in the location and amount of cloud base LW heating (Figure 18a). The outgoing longwave radiation (LW up) is not significantly 16 of 23