1. Introduction
Midlevel clouds impact both the energy budget and vertical profile of heating in the atmosphere, yet the radiative and latent heating impacts are particularly difficult to calculate because they depend on knowledge of both frequency and phase of the clouds. Global climate models often underestimate midlevel cloud frequency (Zhang et al. 2005) and have difficulty accurately predicting hydrometeor phase in mixed-phase regimes (e.g., Klein et al. 2009; Gregory and Morris 1996; Rotstayn et al. 2000). Tropical midlevel clouds are also uniquely important because of the role they play in convective dynamics. Midlevel clouds moisten the tropical atmosphere leading to the onset of deep convection in the Madden–Julian oscillation, a process that convection parameterizations have difficulty capturing (e.g., Inness et al. 2001; Thayer-Calder and Randall 2009; Hagos et al. 2011). Improving climate model predictions requires accurate observational constraints. Observational assessment of midlevel cloud frequency and properties is challenging because they are often blocked by lower or higher clouds when observed remotely from the surface or top of atmosphere, respectively, and they fall into temperature ranges that could contain ice, liquid, or both phases of hydrometeors. The combination of active remote sensors available at the Atmospheric Radiation Measurement Program (ARM) site in Darwin, Australia, thus provides a valuable dataset to explore the characteristics of tropical midlevel clouds.
Previous studies indicate that tropical midlevel clouds are different than their counterparts in midlatitudes and polar regions in their properties, including frequency, thickness, and phase. Zhang et al. (2010) found higher frequencies of thin midlevel clouds in Cloud–Aerosol Lidar and Infrared Pathfinder Satellite Observations (CALIPSO) night overpasses than during daytime overpasses. This difference was substantially higher in the tropics than in other regions. Additionally, midlevel layer clouds are particularly thin in the tropics (Zhang et al. 2010; Seifert et al. 2010; Yasunaga et al. 2006; Ansmann et al. 2009) in comparison with midlatitudes (Seifert et al. 2010).
Though fewer studies detailing the phase of midlevel clouds in the tropics exist compared to the number of studies of midlevel clouds in the Arctic (Shupe et al. 2008, and references therein) and midlatitudes (e.g., Smith et al. 2009; Korolev et al. 2003; Fleishauer et al. 2002; Hobbs and Rangno 1985; Hogan et al. 2003), the studies that have looked at tropical midlevel clouds find that they are more likely to contain liquid water than those of higher latitudes. Satellite studies indicate that a higher percentage of midlevel clouds consist of only supercooled liquid water rather than mixed ice and water in the tropics than in midlatitude or polar regions (Zhang et al. 2010; Hu et al. 2010). Ansmann et al. (2009) found liquid layer-topped clouds at temperatures as cold as −36°C over Cape Verde, Africa, with a substantially lower fraction containing ice than in similar observations over Europe (Seifert et al. 2010). Seifert et al. suggested that the lower rate of ice formation in tropical than midlatitude midlevel clouds could be explained by differences in ice nuclei (aerosols) or meteorology or both. They also found that tropical midlevel clouds tended to be optically thin, shorter-lived altocumulus layers rather than the thicker, more stable stratiform layer clouds of the midlatitudes. The thinner clouds might have a narrower liquid drop spectrum, leading to less heterogeneous ice formation (Hobbs and Rangno 1985).
The formation of midlevel clouds is also unique in the tropics, depending strongly on convective meteorology. Observations from the Tropical Ocean Global Atmosphere Coupled Ocean–Atmosphere Response Experiment (TOGA COARE) field campaign show relationships between stability, cumulus cloud detrainment, humidity, and midlevel clouds. Stable layers are quite frequent around the 0°C level in the tropics, due to a number of factors including advection of dry layers (Mapes and Zuidema 1996), radiative effects (Johnson et al. 1996), and melting effects (Johnson et al. 1996; Yasunaga et al. 2006). These layers of increased stability are hypothesized to inhibit convective growth and cause detrainment in the midtroposphere (Mapes and Zuidema 1996; Johnson et al. 1996, 1999), which may produce midlevel clouds. Zuidema et al. (2006) found that dry air intrusions enhanced sublimation cooling of thick anvil clouds in the eastern Pacific Ocean, causing increased midlevel detrainment due to the temperature anomalies above these dry layers. A second potential mechanism for the formation of thin midlevel clouds is the increased condensation following the latent cooling of stratiform precipitation ice hydrometeors (Johnson et al. 1996; Yasunaga et al. 2006). The relative importance of these two methods in forming thin midlevel layer clouds is not known. Yasunaga et al. suggests that the high frequency of midlevel cloud during the active phase of the Madden–Julian oscillation (MJO) is likely due to the melting–cooling formation method since stratiform precipitation is predominant over convective precipitation during that period.
Midlevel cloud frequency is typically assumed to peak around the melting layer, however, Riley and Mapes (2009) noted a second peak in cloud top height around 8–9 km (−15°C) using measurements from CloudSat. The double peak existed in both cumulus congestus and altocumulus/altostratus type clouds at these levels. The reason for the peak in cloudiness near −15°C is unclear although Riley and Mapes propose two possible mechanisms. The first is that dendritic ice crystal growth is greatest at −15°C, though the peak in the growth curve is broader than that seen in the cloud top heights. The second mechanism suggested by Riley and Mapes (2009) is that a forced gravity wave reverberation above the melting layer might cause the upper-cloud peak.
Two science questions are addressed in this study: what are the properties of tropical midlevel clouds, and how do thin tropical midlevel layer clouds form? The first question is addressed in section 3 by developing a climatology of midlevel clouds at the Darwin Department of Energy (DOE) ARM site. Vertical profiles of cloud boundaries are calculated for over four years of simultaneous micropulse lidar (MPL) and millimeter cloud radar (MMCR) observations. A 1.5-yr subset of the observations is available to calculate the phase of thin midlevel clouds. Tropical observations for this length of time are rare, making this a valuable summary of seasonal and diurnal variation. In section 4 we investigate the formation of thin midlevel layer clouds in the tropics. We hypothesize that thin midlevel clouds at Darwin are formed by increased condensation following latent cooling at the melting level and by detrainment from convection with the former mechanism being predominant in active monsoon periods, and the latter in break monsoon periods. To test this hypothesis, frequencies and properties of midlevel clouds in active and break monsoon periods are analyzed in relation to the convective environment determined by scanning precipitation radar and radiosonde observations. The scanning precipitation radar observations are available for three monsoon seasons during the period of study.
2. Data and methodology
a. Detecting midlevel clouds
Midlevel clouds are detected using vertically pointing lidar (MPL) and cloud radar (MMCR) observations at the ARM measurement site in Darwin (12°25′S, 130°53′E). Periods between November 2005 and April 2010 containing measurements from both the lidar and radar are used in this study. We exclude a 45-min period of data around solar noon each day when the lidar was covered to keep solar radiation from damaging the sensor. Additionally, we visually inspect the data and remove observations when either the lidar or radar was not functioning properly. About 20% of observations are removed owing to missing radar or lidar data. The timeline when both MPL and MMCR data are simultaneously available is shown in Fig. 1 (thick black line).
Timeline of data used for midlevel cloud mask from (top) both radar and lidar, (middle) lidar backscatter and radar reflectivity for phase determination, and (bottom) C-POL precipitation radar.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
An example of one day of vertical profiles of MMCR reflectivity and MPL attenuated backscatter (Fig. 2) shows the value of having simultaneous measurements from both instruments. The lidar backscatter detects more midlevel cloud between 2000 and 0000 UTC than the radar because of its greater sensitivity to small droplets. The sensitivity differences of the radar and lidar also indicates that the thin midlevel cloud between 0000 and 0300 UTC (Fig. 2) is likely precipitating, as can be seen by the strong radar reflectivity throughout the vertical cloud layer and the weak lidar backscatter in the lower part of the cloud.
Example of (top) radar reflectivity, (middle) lidar attenuated backscatter, and (bottom) derived cloud mask for 31 Dec 2005. The colors in the cloud mask indicate whether cloud layers are detected by radar only (green), lidar only (red), or both radar and lidar (blue). The clouds with tops around 5-km heights at times around 0000–0300 and 2100–0000 UTC are identified as thin midlevel clouds.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
The observations are averaged to a common 2-min temporal resolution and 30-m vertical resolution grid. Lidar cloud layers are detected using the algorithm of Wang and Sassen (2001), which uses the change in the slope of the lidar backscattered signal as a function of height to identify the presence of cloud and aerosol in the atmosphere. The cloud mask determined by the lidar is then merged with the radar reflectivity measurements, where it is assumed that the MMCR detects cloud when the radar reflectivity is greater than −50.0 dBZ. We use the radar reflectivity values derived from the Active Remotely-Sensed Cloud Locations (ARSCL) algorithm (Clothiaux et al. 2000) produced by the ARM Program but apply a water vapor attenuation correction to the reflectivity values before using them for cloud detection.
Each vertical profile is then analyzed to identify distinct cloud layers. A cloud layer is defined as a contiguous block of vertical bins containing hydrometeors identified using either the MPL, MMCR, or both. Cloud layers less than eight vertical bins (240 m) apart are combined into a single layer. Cloud layers must be at least eight vertical bins (240 m) in depth.
A frequency distribution of cloud-top heights for all cloud layers determined using the above method shows a relative maximum of occurrence at around 5.5 km (Fig. 3). Based on this distribution, we define midlevel clouds as cloud layers with cloud-top heights between 4 and 8 km (Fig. 3, gray lines). We note that, since our definition is based on cloud top, some clouds defined here may have low-level cloud bases. Further discussion of the different types of clouds with tops between 4 and 8 km is given in section 3.
Distribution of cloud-top heights from combined MMCR and MPL cloud mask at Darwin. Midlevel clouds are those with cloud-top heights between the gray lines.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
b. Determining thin cloud phase
Midlevel clouds may be all ice, all water, or mixed phase, which has implications for their radiative effects. Depolarization ratio can be used to determine whether hydrometeors are ice or water drops, however, because the ARM MPL is uncalibrated and circularly polarized, the depolarization ratio was not used to identify phase in this study. A linear depolarization ratio has been calculated from the circular depolarization measurements, but the values are very noisy and difficult to interpret. Huang et al. (2010) compared linear depolarization ratios derived from linear and circularly polarized lidar observations and also found differences that were not well understood. More work is needed before the depolarization ratio calculated from the circularly polarized lidar observations will be sufficiently reliable for phase determination.
Instead, we use differences between lidar and radar sensitivities to ice and water clouds for estimating the phase of thin midlevel clouds. The backscattering of electromagnetic radiation is proportional to the scattering cross section, which is a function of particle concentration and the ratio of particle size to wavelength. Lidar scattering is most sensitive to the concentration of cloud particles because ice and liquid cloud particles are large compared to the wavelength of the lidar. Scattering from the MMCR, however, is more highly sensitive to the size of hydrometeors because most hydrometeors are small compared to millimeter wavelengths and are therefore in the Rayleigh scattering regime where the backscatter cross section is proportional to the sixth power of the particle diameter. Liquid clouds contain high concentrations of small water droplets, so they have high lidar backscatter and cause the lidar beam to attenuate quickly. Ice particles are generally much larger, but exist at lower concentrations than water droplets. Thus, although water particles have larger backscatter cross sections than ice particles of the same size at millimeter wavelengths, ice particles are generally much larger than cloud water droplets, leading to higher radar reflectivities. Drizzle and rain droplets are also larger than cloud droplets, but the frequency of ice clouds is only calculated for temperatures below 0°C when large particles are assumed to be ice.
Thin clouds containing supercooled liquid water are identified using lidar backscatter and attenuation criteria similar to methods described in the literature (Zhang et al. 2010; Hogan et al. 2004). The top panel of Fig. 4 shows a joint histogram of cloud-top temperature and the log of the attenuated backscatter. The strong peak around 0°C shows the backscatter values for thin clouds containing liquid water. A second peak at temperatures colder than −40°C contains clouds in which all hydrometeors are ice particles. Note that the peak at these colder temperatures corresponds to lower attenuated backscatter values. An attenuated backscatter threshold (Fig. 4, black line) is subjectively determined to separate layer clouds containing liquid water from those containing only ice. The sensitivity of the results to this choice of threshold is examined by decreasing or increasing the threshold to the values given by the dashed lines plotted in Fig. 4. In addition to the backscatter criterion, we also require the lidar to attenuate above the liquid water layer, that is, the minimum attenuated backscatter in the 210 m above the maximum backscatter must be at least a factor of 20 less than the maximum backscatter value.
(top) Attenuated backscatter threshold definition (thick black line) separating thin layer clouds (≤2 km) containing supercooled liquid water from those containing only ice. Dashed lines show lower and upper limits used for threshold sensitivity test. Note that the attenuated backscatter units on the x axis are not physically meaningful because the lidar has not been calibrated. (bottom) Joint histogram of cloud layer maximum radar reflectivity with temperature for thin liquid cloud layers containing liquid water. Black line indicates threshold used to distinguish supercooled liquid water clouds from mixed-phase clouds using the method described in the text. Color scales indicate log of number of observations in each bin.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
The MMCR radar reflectivity is used to obtain an estimate of the fraction of those supercooled liquid water clouds that also contain ice. The subset of thin clouds that have been identified as containing liquid water using the attenuated backscatter criteria are identified as mixed-phase clouds when their reflectivities are greater than a temperature-dependent threshold determined by the method of Zhang et al. (2010). The bottom panel of Fig. 4 shows that threshold (lower panel, black line) on a joint histogram of the occurrence of thin liquid water clouds versus reflectivity and temperature. The temperature dependence of the maximum liquid water content in a cloud layer is calculated assuming that a cloud layer is saturated at cloud base and grows adiabatically to a layer thickness of 400 m. The radar reflectivity is calculated as a function of liquid water content using the same formula chosen by Zhang et al.
Note that we restrict our phase determination to thin layer clouds because we require clouds that are optically thin enough for the lidar to see all or most of the cloud. For similar reasons, we also remove thin midlevel clouds with underlying clouds in order to ensure that the lidar backscatter is not attenuated by lower clouds (55% of clouds are removed using this criterion). We also only analyze data from June 2007 to December 2008 when the same MPL was installed at Darwin since the lidar backscatter is not calibrated.
c. Precipitation radar
A C-band (5-cm wavelength) scanning precipitation radar (C-POL) (Keenan et al. 1998) operated by the Australian Bureau of Meteorology during the wet season at Darwin provides horizontal context to the ARM vertically profiling measurements. Three monsoon seasons of processed C-POL data were available for this study (10 November 2005–29 March 2006; 12 October 2006–31 March 2007; and 3 December 2009–30 April 2010, as shown in Fig. 1). The C-POL does not operate during the dry season and was also inoperational during the 2007/08 and 2008/09 monsoon seasons. The instrument is located ~25 km northeast of the MPL and MMCR (Fig. 5). The method used to process the C-POL observations is described in detail in Frederick and Schumacher (2007). Here we will summarize four classes of cloud/precipitation identified from C-POL observations: shallow convection, medium convection, deep convection, and stratiform rain. Convective and stratiform rain are distinguished based on the reflectivity at 2.5 km. Types of convection are then defined as shallow, medium, and deep when their tops are less than 4 km, between 4 and 8 km, and greater than 8 km, respectively. These identified cloud types are gridded horizontally with 2.5-km resolution out to a horizontal radius of 120 km from the radar. Observations are available every 10 min.
Example of 120-km radius, C-POL radar reflectivity scan at an altitude of 2.5 km from 31 Dec 2005 at 2100 UTC. The locations of C-POL and MMCR/MPL sites are shown.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
d. Radiosondes
Static stability and change in humidity with height are calculated from radiosonde profiles of the atmosphere so as to investigate the relationship between stability and thin midlevel clouds. Radiosonde temperature and humidity profiles are available twice per day at Darwin, around noon and midnight UTC (about 0930 and 2130 local time). Vaisala RS80 radiosondes were used until 18 January 2006; then Vaisala RS92 radiosondes were used during the remainder of the analysis period (Holdridge et al. 2011).
3. Climatology of midlevel clouds
a. Cloud thickness
There are multiple cloud types that could have top heights between 4 and 8 km. The joint histogram of cloud thickness and cloud-top temperature (Fig. 6) shows two distinct groupings of midlevel clouds. The peak in frequency at the largest thicknesses corresponds to thick clouds with precipitation that reaches the surface. These clouds could be a mix of cumulus congestus and stratiform clouds. Thin clouds have the highest frequencies with a strong peak in clouds less than 1 km thick.
Cloud-top temperature vs thickness for midlevel clouds.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
Based on the clear separation of the two cloud types in Fig. 6, we distinguish between thin and thick midlevel clouds using a cloud thickness threshold of 2 km. Seventy percent of midlevel clouds are less than 2 km thick, more than 50% are less than 1 km thick, and over 25% of midlevel clouds are 0.5 km thick or less, noting that we only consider clouds at least 0.24 km thick. This high frequency of thin clouds agrees with other studies (Seifert et al. 2010; Yasunaga et al. 2006; Ansmann et al. 2009) that found layer clouds to be especially thin in the tropics.
b. Cloud temperature
Cloud properties at Darwin are strongly influenced by the Australian monsoon (Pope et al. 2009). During the wet season (defined here as November–April) there is heavy monsoon rain and a high frequency of deep convective clouds, while during the dry season (May–October) there is very little local convection or precipitation. We expect midlevel cloud properties to differ during the wet and dry season due to the large differences in humidity and convective activity. Figure 7 shows histograms of midlevel cloud-top temperature in the wet season (top panel) and the dry season (bottom panel) at Darwin. The higher frequency of deep convection also leads to a higher frequency of midlevel clouds during the wet season (note the different abscissa scales of each panel in Fig. 7).
Cloud-top temperature distributions for thick (dashed), thin (solid gray), and all (thick black) midlevel clouds by season. Wet season is November–April; dry season is May–October.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
A double peak in cloud-top temperature is visible in both the wet (top) and dry (bottom) seasons, and in both thin (solid gray line) and thick clouds (dashed line). The double peak is more pronounced in the thin clouds, particularly in the dry season. The first, stronger, peak in thin midlevel cloudiness occurs at temperatures slightly colder than 0°, or slightly above the melting layer, in both seasons. The second, smaller, thin midlevel peak exists around −12.5°C and corresponds to the second peak in midlevel cloudiness seen by Riley and Mapes (2009). The second peak is also seen in thick midlevel clouds at the same temperature although it is less distinct than the peak in thin clouds. The two peaks can also be seen in the cloud-top height distribution (Fig. 3) with one peak centered at ~5.5 km and one at ~7.5 km. The two peaks are broader in height than in temperature, suggesting that peak locations correspond more closely to temperature than height. Although Riley and Mapes found the upper peak at temperatures around −15°C, rather than at −12.5°C as found here, the peak is at the same height (between 7 and 8 km). It is not known what causes this temperature difference, though it could be due to regional differences between the studies. Some part of the difference might also be due to differing sources of atmospheric temperature profiles, European Centre for Medium-Range Weather Forecasts analysis data in the study by Riley and Mapes (2009), and radiosondes in this study.
c. Cloud frequency
The frequency of thin midlevel clouds is a minimum around noon local time for both wet and dry seasons (Fig. 8, top). The gap in the figure around noon is caused by removing times when lidar data is unavailable, as described in section 2a. The higher frequency of thin midlevel cloud at 0130 LST than at 1330 LST is consistent with the results seen by Zhang et al. (2010) who found higher frequencies of thin midlevel clouds during CALIPSO night overpasses than during daytime overpasses. However, the ground-based sensors observe the full diurnal cycle of midlevel cloud frequency and indicate an overall diurnal cycle with a minimum around 1400–1500 LST and a peak after midnight around 0200–0300 LST.
Diurnal frequency of occurrence for (top) thin and (bottom) thick midlevel clouds. Wet season (thick line) is November–April; dry season (thin line) is May–October.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
The diurnal cycle in thin clouds may be due to the effects of radiative heating. Longwave cooling at the top of clouds maintains clouds by creating instability that drives internal circulations (Heymsfield et al. 1991; Starr and Cox 1985). Shortwave heating by solar radiation will counteract some of that longwave cooling, leading to more dissipation of clouds during daytime. Radiative heating was hypothesized to explain the stronger diurnal cycle of midlatitude altocumulus clouds in summer than in other seasons (Lazarus et al. 2000). Radiative heating would also cause a stronger diurnal cycle in the tropics than at higher latitudes as seen by Zhang et al. (2010).
The additional noise from solar radiation makes it more difficult to detect thin clouds with the lidar during the day. However, we do not think lidar detection issues are the main driver of the observed diurnal cycle in midlevel clouds since it is also detected by the cloud radar. The cloud radar operates at a millimeter wavelength, so is not impacted by solar radiation, though it may miss clouds with small water drops.
The diurnal variations in thick midlevel clouds (Fig. 8, bottom) are less uniform. Because of the lack of convection and stratiform precipitation in the dry season, thick midlevel clouds are observed less than 2% of the time, insufficient for clear distinction in the diurnal cycle. During the wet season, there appear to be peaks around 0700 LST and also in the afternoon between 1200 and 1600 LST. These wet season diurnal variations correspond to peaks in medium-height cumulus clouds presented in Section 4.
d. Cloud phase
A climatology of the phase of thin midlevel clouds is developed from 11/2 years of MPL and MMCR observations (see Fig. 1) using the methods described in section 2b. To examine the full possible range of mixed-phase clouds, we consider all thin clouds (less than 2 km thick) with cloud tops between 0° and −40°C, rather than using the 4–8-km cloud-top height restriction. The number of clouds that fit these criteria in each 5°C temperature bin is shown in the top panel of Fig. 9. Note that bins with temperature less than −20°C would typically be higher than 8 km and would not be included as midlevel clouds in our classification.
(top) Number of thin clouds observed in each 5°C temperature bin. (middle) Percentage of those clouds in each temperature bin that contain liquid water layers. (bottom) Percentage of liquid water clouds that also contain ice. Dashed lines represent the sensitivity of the results to the attenuated backscatter threshold as plotted in Fig. 4.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
The percentage of thin midlevel clouds containing liquid water is consistent with the few relevant studies in the literature. Eighty percent of thin clouds warmer than 0°C are found to contain liquid water (Fig. 9, second panel), consistent with the uncertainty of the backscatter threshold method of determining phase as described by Hogan et al. (2004). Only clouds containing liquid water are analyzed in this section. The remaining clouds are assumed to contain all ice. The percentage of clouds containing liquid water decreases monotonically and approximately linearly with decreasing temperature (Fig. 9, second panel). The dashed lines in Fig. 9 show that the results are not highly sensitive to using the lower and higher attenuated backscatter thresholds plotted in Fig. 4. Approximately 50% of thin clouds at −20°C contain liquid water, agreeing well with the 50%–60% estimates calculated from CALIPSO data around Darwin (Choi et al. 2010, read from their Fig. 1). Ansmann et al. (2009) also found clouds containing liquid water at temperatures as cold as −36°C, matching our finding that a nonnegligible percentage of clouds colder than −30°C contain liquid water, though they found much higher fractions of clouds that were all liquid water at this site (nearly all altocumulus clouds contained liquid water at all temperatures).
The percentage of liquid water clouds at Darwin that also contain ice (Fig. 9, third panel) remains lower than 50% for temperatures as cold as −30°C, lower than the results of Zhang et al. (2010), who found a sharp increase in the fraction of midlevel clouds that are mixed phase, from 35% to 75% between the temperatures of −7° and −16°C. One reason for the difference between our results at Darwin and those from Zhang et al. may be due to geographical variation. Zhang et al. calculated mixed-phase fractions for zonal averages between 0° and ±30° latitude. However, there is evidence for spatial variability in the phase of midlevel clouds in the tropics. Choi et al. (2010, Figs. 1–2) find that the percentage of clouds at −20°C that contain liquid water ranges from 25% to 75% throughout the tropics and is anticorrelated with the concentrations of dust at the same altitude, suggesting that the regional variability of ice nuclei in the tropics strongly impacts the phase and radiative impact of midlevel clouds. Similarly, Kanitz et al. (2011) found higher frequencies of ice in the Northern Hemisphere than in similar observations in the Southern Hemisphere, indicating that the higher ice nuclei concentrations in the Northern Hemisphere caused more heterogeneous ice nucleation. The fraction of thin midlevel clouds containing ice and indicative of heterogeneous nucleation between −15° and −20°C ranged from less than 10% to 40% at Southern Hemisphere sites (Kanitz et al. 2011), lower than the approximately 60% of clouds estimated to be either ice or mixed phase at the same temperature at Darwin. This difference could be due to seeding of midlevel clouds by ice falling from higher clouds, uncertainties in the phase determination (this method likely overestimates the fraction of clouds containing ice at −20°C), or high concentrations of continental aerosols at Darwin. Distinguishing between these causes is a subject for future research.
4. Monsoon season midlevel cloud formation
Midlevel cloud formation is investigated by examining differences in cloud frequency and properties in different stages of the Australian monsoon. The wet season at Darwin occurs approximately between November and April. It is characterized by active monsoon periods with westerly winds bringing oceanic air and convection interspersed with break monsoon periods consisting of easterly winds and continental convection (e.g., Nicholls et al. 1982; Drosdowsky 1996; Pope et al. 2009). Convective differences between these two periods are a useful tool for analyzing the formation and properties of midlevel clouds.
We identify active and break monsoon periods using the method of Pope et al. (2009). They determine five wet season regimes by clustering thermodynamic and wind soundings at Darwin. We use only the two most moist regimes for this study. Active monsoon periods correspond to the deep west regime (Pope et al. 2009, regime 2) with westerly winds at the surface changing to easterly winds at 400 hPa. The break monsoon period is described by the moist east regime (Pope et al. 2009, regime 5) with easterly winds throughout the profile. The remaining three regimes are primarily transition regimes between dry and wet seasons with less moisture and convection than the active and break monsoon periods.
a. Connection to convection
Differences between the oceanic and land convection during the active and break monsoons are examined in relationship to the frequency of thin midlevel clouds to investigate whether midlevel clouds form differently when these two types of convection are dominant. The frequency of convective and stratiform fractional area observed by the C-POL precipitation radar data (determined as described in Frederick and Schumacher 2007) and averaged over a 40 km by 40 km square grid box centered on Darwin is compared with the frequency of midlevel clouds observed by the MPL and MMCR during the active and break monsoon periods. Analysis of the C-POL data is not performed for the other three regimes.
The diurnal cycle of convection is different in the active and break monsoon periods (regimes 2 and 5 described above) due to the predominance of oceanic and land convection in those periods respectively. During the active monsoon period, convective precipitation is highest in the morning (0000–1200 LST) though there is not a clear maximum (Fig. 10, top left, dashed line). If we consider only deep convection (Fig. 10, middle left), there are several peaks between 0200 and 1000 LST. This is consistent with the diurnal cycle of oceanic convection that shows a higher frequency of convective events at night and in the morning, peaking around 0500 LST and less convection in the afternoon with a minimum around 1800 LST (Nesbitt and Zipser 2003; Yang and Slingo 2001). During the break monsoon period, however, the peak of convection occurs around 1500 LST (Fig. 10, top right) with the peak in deep convection slightly later than that of shallow and medium convection (Fig. 10, middle right). Both the distinct maximum in deep convection and its timing are characteristic of continental convection (Nesbitt and Zipser 2003). Land convection tends to a distinct diurnal cycle in intensity and precipitation caused by the solar heating of the land surface (e.g., Futyan and Del Genio 2007; Nesbitt and Zipser 2003).
Average diurnal cycle of cloud frequency from (left) active and (right) break monsoon periods. (top and middle) Stratiform precipitation and convective precipitation identified from C-POL observations over a grid box 40 km on a side centered on Darwin. (bottom) Midlevel clouds identified from MPL and MMCR. Midlevel cloud time series are smoothed with a 30-min moving average.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
The diurnal peak of stratiform precipitation generally follows the maximum convective occurrence in both periods. In the active monsoon periods, the peak in stratiform precipitation occurs around noon (Fig. 10, top left), 4–10 h after the peaks in frequency of deep convection and roughly coincident with the largest peak in medium convection. In the break monsoon period, two stratiform precipitation peaks are observed in the diurnal cycle (Fig. 10, top right). One is coincident with the peak in convection, particularly with the peak in medium convection. The second peak extends from approximately 2100 to 0300, 5–11 h after the peak in deep convection. The coincident peaks in medium convection and stratiform precipitation may indicate the uncertainty in distinguishing convective and stratiform precipitation from radar data. The peak in stratiform precipitation following deep convection (4–11 h after the convection) is likely representative of the life cycle of deep convective clouds. Following strong convection, deep convective clouds detrain large amounts of moist air and hydrometeors into stratiform clouds that continue to precipitate. The amplitude of the peak in stratiform precipitation is over twice as large during the active monsoon period than the break monsoon period (Fig. 10, top panels; note difference in scales), consistent with the greater fraction of stratiform precipitation to convective precipitation seen in convective systems over the ocean than land in the tropics (e.g., Futyan and Del Genio 2007; Schumacher and Houze 2003).
The diurnal cycle of midlevel clouds is more distinct in the active than the break monsoon period (Fig. 10, bottom panels), though both show higher frequencies at night than during the afternoon. In the active monsoon period, midlevel clouds have the highest frequency at night or in the morning and the lowest frequency in the afternoon. This is particularly true for thin midlevel clouds. There is a smaller variation in the midlevel clouds of the break monsoon period but, to the extent that there is a diurnal variation, the values at night and in the morning are also larger than those of the afternoon. As discussed in section 3c, this may be due to solar heating, and it is not clear from this analysis whether there is a connection between the diurnal cycle of convection and the frequency of midlevel clouds. The comparison is also difficult because the limited spatial sampling of the vertically pointing MPL and MMCR leads to more noise in the midlevel cloud time series than the C-POL observations, even with the smoothing applied in Fig. 10.
To better examine whether a relationship exists between the timing of stratiform and midlevel clouds that may indicate specific formation mechanisms, thin midlevel cloud events are identified in the time series. Consecutive hours containing midlevel cloud frequencies greater than zero are grouped into midlevel cloud events and the hours preceding those events are examined for stratiform cloud. A greater fraction of midlevel clouds are preceded by stratiform cloud during active than break monsoon periods. Figure 11 shows 65% of active, but only 45% of break, midlevel clouds follow stratiform cloud, for which the existence of stratiform cloud is defined as a frequency of at least 5% of C-POL pixels in the 40-km box surrounding Darwin at some point in the previous 5 h. The overall fraction of midlevel clouds preceded by stratiform cloud is sensitive to this definition, but the percentage of active midlevel clouds preceded by stratiform clouds is consistently about 20% greater than the percentage of break midlevel clouds preceded by stratiform clouds for a wide range of definitions (not shown). This higher occurrence of stratiform cloud before midlevel clouds suggests that the melting–cooling formation mechanism described by Yasunaga et al. (2006) is a more predominant cause of midlevel cloud formation during active than break monsoon periods.
Bar graph shows the percentage of midlevel clouds in active and break monsoon periods with greater than 5% stratiform cloud frequency sometime in the preceding 5 h.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
We also examined the crosscorrelation between the hourly frequency of stratiform and thin midlevel clouds. During active monsoon periods, midlevel cloud frequency is most strongly correlated with the frequency of stratiform cloud 13–17 h prior (Fig. 12, black line). During break periods, the peak in the correlation occurs between stratiform cloud 5 h earlier (Fig. 12, gray line), though this is not as strong as the correlation in the active periods. Visual inspection of cloud observations shows that active monsoon periods are dominated by long-lasting organized convective systems with periods of convection and stratiform cloud lasting for long periods of time, followed by long stretches containing midlevel clouds (see the example in Fig. 13). These long stretches of cloudiness are likely the cause of the correlation at long lag time seen in Fig. 12. The peak in correlation between midlevel cloud frequency 10 h prior to stratiform cloud frequency is also likely caused by the recurrence of new convection and stratiform precipitation shortly following previous convective events (also seen in Fig. 13). The character of convective systems appears to vary more during break monsoon periods (not shown), including both long-lasting convective systems and more isolated deep and medium convection lasting a few hours or less. Midlevel clouds are also occasionally seen in the break monsoon period at times without substantial convective or stratiform cloud, indicating that advection of clouds or humidity into the region might also be important. An initial quantitative assessment of this difference was made by defining cloud events that contained consecutive hours of convective, stratiform, or midlevel cloud. During active monsoon periods, 40% of these events last longer than 10 h, but only 20% of events during break monsoon periods are sustained longer than 10 h.
Crosscorrelation between hourly time series of stratiform thin midlevel cloud frequency are calculated for ±20 h for the active and break monsoon periods. A positive lag indicates midlevel clouds follow stratiform precipitation.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
Time series of hourly frequency of convection (red line), stratiform cloud (green line), and midlevel cloud (blue line) from (top) 9 Mar 2006, in regime 2, along with (second row) MMCR reflectivity, (third row) MPL backscatter, and (bottom) a composite cloud mask. Panels 2–4 use the same scales and color schemes as in Fig. 2.
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
b. Relationship to stability
The heights of midlevel clouds have been shown to correspond to stable layers in the atmosphere (e.g., Johnson et al. 1996). Johnson et al. found moist stable layers just below the 0°C level within or very near precipitating systems suggesting the melting–cooling mechanism was responsible for thin clouds formed at this level. Soundings taken outside of precipitating clouds contained dry stable layers slightly above the 0°C level. Johnson et al. suggest that it is these stable layers that are likely to cause detrainment from convection.
We find that thin midlevel clouds in the break monsoon period frequently occur at temperatures (heights) corresponding to stable layers using three different stability thresholds. During the break monsoon there is a single strong peak in the stability around 0°C and a strong peak in the occurrence of dry layers at colder temperatures than for the stability peak (Fig. 14, bottom center and right respectively) that is matched by a strong peak in midlevel cloudiness just above 0°C (Fig. 14, bottom left). While we cannot say from these observations that the increased stability and humidity drops precede or cause the increase in cloudiness, the stability peak is consistent with the tropical profiles typically seen by Johnson et al. (1996, Fig. 5) during TOGA COARE that are suggested to be associated with convective detrainment.
(left) Cloud-top temperature histograms for thick (dashed line), thin (thin gray line), and all (thick black line) midlevel clouds for (top) active and (bottom) break monsoon periods. (middle) Frequency of stable layers defined at three different stability thresholds. (right) Frequency of decreases in relative humidity with height less than three different thresholds [% (100 m)−1].
Citation: Journal of Climate 25, 19; 10.1175/JCLI-D-11-00599.1
During the active period (Fig. 14, top) there is a clear double peak in midlevel cloud-top temperatures around −2.5° and −12.5°C, particularly in thin midlevel clouds (gray line). The occurrence of stable layers does not share a strong double peak with cloud top temperatures. Instead, there is a relatively high occurrence of stable layers stronger than −5°C km−1 over a broad range of temperatures (between about −10° and 5°C). Similarly, there is no clear relationship between the occurrence of humidity drops and the peaks in cloudiness. There are also two peaks in thin midlevel cloud base temperature, at 0° and −5°C, in the active monsoon period though only a single peak at 0°C in the break period (not shown). The lack of a strong peak in stability coupled with the high frequency of stratiform precipitation in this regime might indicate that the peak in cloudiness near 0°C is caused by the melting–cooling mechanism. There is a higher frequency of stable layers below the 0°C level, as would be expected from the increased latent heating of condensation following melting of solid hydrometeors in stratiform precipitation. The peak in thin cloud top temperature is lower than that of the break period, but is still slightly above the 0°C level, higher than the melting–cooling clouds described by Johnson et al. (1996).
Several possible explanations exist for the increased frequency in thin clouds at −12.5°C during the active monsoon. Latent heating from rapid ice growth around −15°C might cause more ice formation. This microphysical mechanism for the upper peak would be stronger when there is more cloud growth and moisture as in the case of the large oceanic convective clouds of the active monsoon period. There is a small peak in stability around −15°C that could be caused by increased ice growth at this level. However, this peak is actually slightly higher than the cloud top peak. There is also no evidence for increased ice hydrometeors in the thin cloud phase calculations of section 3d, although the increased ice growth could be occurring within thick clouds rather than thin clouds. A second possible explanation is increased detrainment or cloud spreading from wind shear. The active monsoon period is characterized by strong wind shear in the middle troposphere, where the wind switches from westerly to easterly zonal flow around 400 hPa (Pope et al. 2009), approximately −15°C. However, a conclusive assessment of the impact of wind shear on these clouds would also require investigating the dependence on the humidity and microphysics of clouds as shear also increases evaporation of clouds and reduces their lifetimes (Fan et al. 2009, 2010; Khain et al. 2005).
5. Discussion and conclusions
The results of the midlevel cloud climatology from a uniquely long tropical dataset at Darwin are consistent with the few other studies of tropical midlevel clouds. Both thin (defined here as less than 2 km) layer clouds and thick precipitating clouds exist with tops between 4 and 8 km. Seventy percent of these midlevel clouds are less than 2 km thick, and more than 50% of the clouds are less than 1 km thick. We only consider clouds at least 0.24 km thick, or these numbers would likely be even higher. The predominance of thin clouds agrees with other studies from the tropics (e.g., Seifert et al. 2010). We find a marked diurnal cycle in the frequency of thin clouds with a maximum at night and in the morning and a minimum around noon. This matches a strong diurnal difference seen in night and day CALIPSO overpasses in the tropics (Zhang et al. 2010). The diurnal cycle may be caused by cloud dissipation due to solar heating, which would explain the stronger diurnal cycle in the tropics than higher latitudes seen by Zhang et al. Thin layer clouds are likely to contain liquid water throughout the mixed-phase temperature range (from 0° to −40°C), with as many as 50% of clouds containing liquid water at −20°C, and only about 30% of those clouds also containing ice. This agrees with other findings that tropical midlevel clouds are more likely to contain supercooled liquid water than their counterparts at higher latitudes (Seifert et al. 2010; Zhang et al. 2010; Ansmann et al. 2009), either because of a lack of aerosols appropriate for nucleating ice (Seifert et al. 2010; Choi et al. 2010) or a sufficiently narrow droplet size spectrum to inhibit heterogenous nucleation (Seifert et al. 2010; Hobbs and Rangno 1985).
Two peaks were seen in cloud top temperature, around 0° and −12.5°C, as seen in satellite data (Riley and Mapes 2009). The upper peak is more prominent in the dry season (May–October) than in the wet season (November–April). When the wet season is divided into different meteorological regimes, the −12.5°C peak in cloud-top temperature is seen primarily during the active monsoon period rather than the break monsoon period. One possible cause of this peak is increased cloudiness or latent heating from ice dendritic growth owing to the higher frequency of stratiform precipitation in the active than in break monsoon periods. There is very little indication of increased ice at this temperature in the analysis of the phase of thin midlevel clouds, however no analysis was done of the phase of thick clouds because of the limitations of the data. Another cause of the second peak could be the stronger wind shear at these heights in the active monsoon period than other wet season periods. The wind shifts from westerly to easterly around 400 hPa (about −15°C) during the active monsoon (Pope et al. 2009). This could cause an increase in cloud horizontal area around this temperature.
Several differences were found in cloud properties during the active and break periods of the monsoon season.
The diurnal cycle of convection in active periods is consistent with oceanic convection as would be expected from the westerly winds. The break periods, characterized by easterly winds, show a diurnal cycle matching that of land-based convection.
The frequency of clouds associated with convection is higher in active than break monsoon periods. In particular, there is approximately twice the frequency of stratiform precipitation in active than break periods.
Analysis of the timing of stratiform and midlevel clouds shows that 65% of midlevel clouds are preceded by stratiform cloud in active periods, and there is a high correlation between the frequency of stratiform precipitation and the frequency of thin midlevel cloud about 13–17 h later. Only 45% of midlevel clouds are preceded by stratiform cloud in the break monsoon season. Stratiform precipitation is most highly correlated with thin midlevel cloud 5 h later, though the correlation is weaker than during active periods.
There is a distinct peak in the frequency of stable layers around 0°C in the break monsoon periods and a corresponding single strong peak in thin midlevel clouds slightly above this layer around −5°C.
The strongest peak in decreases in the vertical relative humidity profile occur at heights above (temperatures colder) than −5°C, just above the main peak in cloudiness during break monsoon periods.
In the active monsoon periods, a broad peak in the frequency of stable layers is seen between 5° and −10°C. This peak does not correspond clearly to the sharp peak in thin cloud frequency slightly above 0°C.
The temperatures at which decreases in the vertical relative humidity gradient occur do not show a clear correspondence to the temperatures of maximum cloud in active monsoon periods.
One possible interpretation of the differences between active and break monsoon periods is that thin midlevel clouds are formed differently during these times. We propose that midlevel clouds formed by the melting–cooling mechanism following stratiform precipitation are more likely to be found during the active monsoon period when stratiform precipitation is more frequent. This hypothesis is supported by the high frequency of stratiform cloud preceding thin midlevel clouds. During the break period the frequency of midlevel cloud corresponds well to stable layers and less well to the timing of stratiform precipitation, suggesting that other formation mechanisms like convective detrainment may be important as well.
Several recent modeling studies have examined the dynamical and microphysical processes behind the formation and maintenance of tropical midlevel clouds (Posselt et al. 2008; Yasunaga et al. 2008). Further understanding of the processes driving midlevel cloud formation and their relative importance during different dynamical regimes could be gained by using the long-term dataset at Darwin to constrain and evaluate such simulations.
Acknowledgments
We thank Courtney Schumacher for providing C-POL cloud type data, Mick Pope for providing the dates of the monsoon regimes, Christian Jakob for helpful discussions regarding the stability, Connor Flynn for technical discussions on the lidar, and Jim Mather and anonymous reviewers for helpful comments on the manuscript.
REFERENCES
Ansmann, A., and Coauthors, 2009: Evolution of the ice phase in tropical altocumulus: SAMUM lidar observations over Cape Verde. J. Geophys. Res., 114, D17208, doi:10.1029/2008JD011659.
Choi, Y.-S., R. Lindzen, C.-H. Ho, and J. Kim, 2010: Space observations of cold-cloud phase change. Proc. Natl. Acad. Sci. USA, 107, 11 211–11 216, doi:10.1073/pnas.1006241107.
Clothiaux, E., T. Ackerman, G. Mace, K. Moran, R. Marchand, M. Miller, and B. Martner, 2000: Objective determination of cloud heights and radar reflectivities using a combination of active remote sensors at the ARM CART sites. J. Appl. Meteor., 39, 645–665.
Drosdowsky, W., 1996: Variability of the Australian summer monsoon at Darwin: 1957–1992. J. Climate, 9, 85–96.
Fan, J., and Coauthors, 2009: Dominant role by vertical wind shear in regulating aerosol effects on deep convective clouds. J. Geophys. Res., 114, D22206, doi:10.1029/2009JD012352.
Fan, J., J. Comstock, and M. Ovchinnikov, 2010: The cloud condensation nuclei and ice nuclei effects on tropical anvil characteristics and water vapor of the tropical tropopause layer. Environ. Res. Lett., 5, 044005, doi:10.1088/1748-9326/5/4/044005.
Fleishauer, R., V. Larson, and T. Vonder Haar, 2002: Observed microphysical structure of midlevel, mixed-phase clouds. J. Atmos. Sci., 59, 1779–1804.
Frederick, K., and C. Schumacher, 2007: Anvil characteristics as seen by C-POL during the Tropical Warm Pool International Cloud Experiment (TWP-ICE). Mon. Wea. Rev., 136, 206–222.
Futyan, J., and A. Del Genio, 2007: Deep convective system evolution over Africa and the tropical Atlantic. J. Climate, 20, 5041–5060.
Gregory, D., and D. Morris, 1996: The sensitivity of climate simulations to the specification of mixed phase clouds. Climate Dyn., 12, 641–651.
Hagos, S., L. Leung, and J. Dudhia, 2011: Thermodynamics of Madden–Julian oscillation in a regional model with constrained moisture. J. Atmos. Sci., 68, 1974–1989.
Heymsfield, A., L. Miloshevich, A. Slingo, K. Sassen, and D. O’C. Starr, 1991: An observational and theoretical study of highly supercooled altocumulus. J. Atmos. Sci., 48, 923–945.
Hobbs, P., and A. Rangno, 1985: Ice particle concentrations in clouds. J. Atmos. Sci., 42, 2523–2549.
Hogan, R., P. Francis, H. Flentje, A. Illingworth, M. Quante, and J. Pelon, 2003: Characteristics of mixed-phase clouds I: Lidar, radar and aircraft observations from CLARE’98. Quart. J. Roy. Meteor. Soc., 129, 2089–2116.
Hogan, R., M. Behera, E. O. Connor, and A. Illingworth, 2004: Estimate of the global distribution of stratiform supercooled liquid water clouds using the LITE lidar. Geophys. Res. Lett., 31, L05106, doi:10.1029/2003GL018977.
Holdridge, D., M. Ritsche, J. Prell, and R. Coulter, 2011: Balloon-Borne Sounding System (SONDE) handbook. U.S. Department of Energy Tech. Rep. DOE/SC-ARM/TR-029, 27 pp.
Hu, Y., S. Rodier, K. Xu, W. Sun, J. Huang, B. Lin, P. Zhai, and D. Josset, 2010: Occurrence, liquid water content, and fraction of supercooled water clouds from combined CALIOP/IIR/MODIS measurements. J. Geophys. Res., 115, D00H34, doi:10.1029/2009JD012384.
Huang, Z., N. Sugimoto, J. Huang, T. Hayasaka, T. Nishizawa, J. Bi, and I. Matsui, 2010: Comparison of depolarization ratio measurements with micro-pulse lidar and a linear polarization lidar in Lanzou, China. Proc. 25th Int. Laser Radar Conf., St. Petersburg, Russia, NIES, 528–531.
Inness, P., J. Slingo, S. Woolnough, R. Neale, and V. Pope, 2001: Organization of tropical convection in a GCM with varying vertical resolution; implications for the simulation of the Madden-Julian oscillation. Climate Dyn., 17, 777–793.
Johnson, R., P. Ciesielski, and K. Hart, 1996: Tropical inversions near the 0°C level. J. Atmos. Sci., 53, 1838–1855.
Johnson, R., T. Rickenbach, S. Rutledge, P. Ciesielski, and W. Schubert, 1999: Trimodal characteristics of tropical convection. J. Climate, 12, 2397–2418.
Kanitz, T., P. Seifert, A. Ansmann, R. Engelmann, D. Althausen, C. Casiccia, and E. Rohwer, 2011: Contrasting the impact of aerosols at northern and southern midlatitudes on heterogeneous ice formation. Geophys. Res. Lett., 38, L17802, doi:10.1029/2011GL048532.
Keenan, T., K. Glasson, F. Cummings, T. Bird, J. Keeler, and J. Lutz, 1998: The BMRC/NCAR C-band polarimetric (C-Pol) radar system. J. Atmos. Oceanic Technol., 15, 871–886.
Khain, A., D. Rosenfeld, and A. Pokrovsky, 2005: Aerosol impact on the dynamics and microphysics of deep convective clouds. Quart. J. Roy. Meteor. Soc., 131, 2639–2663.
Klein, S., and Coauthors, 2009: Intercomparison of model simulations of mixed-phase clouds observed during the ARM mixed-phase arctic cloud experiment. I: Single-layer cloud. Quart. J. Roy. Meteor. Soc., 135, 979–1002.
Korolev, A., G. Isaac, S. Cober, J. Strapp, and J. Hallett, 2003: Microphysical characterization of mixed-phase clouds. Quart. J. Roy. Meteor. Soc., 129, 39–65.
Lazarus, S., S. Krueger, and G. Mace, 2000: A cloud climatology of the Southern Great Plains ARM CART. J. Climate, 13, 1762–1775.
Mapes, B., and P. Zuidema, 1996: Radiative–dynamical consequences of dry tongues in the tropical troposphere. J. Atmos. Sci., 53, 620–620.
Nesbitt, S., and E. Zipser, 2003: The diurnal cycle of rainfall and convective intensity according to three years of TRMM measurements. J. Climate, 16, 1456–1475.
Nicholls, N., J. McBride, and R. Ormerod, 1982: On predicting the onset of the Australian wet season at Darwin. Mon. Wea. Rev., 110, 14–17.
Pope, M., C. Jakob, and M. Reeder, 2009: Regimes of the North Australian wet season. J. Climate, 22, 6699–6715.
Posselt, D., S. van den Heever, and G. Stephens, 2008: Trimodal cloudiness and tropical stable layers in simulations of radiative convective equilibrium. Geophys. Res. Lett., 35, L08802, doi:10.1029/2007GL033029.
Riley, E., and B. Mapes, 2009: Unexpected peak near −15°C in CloudSat echo top climatology. Geophys. Res. Lett., 36, L09819, doi:10.1029/2009GL037558.
Rotstayn, L., B. Ryan, and J. Katzfey, 2000: A scheme for calculation of the liquid fraction in mixed-phase stratiform clouds in large-scale models. Mon. Wea. Rev., 128, 1070–1088.
Schumacher, C., and R. Houze Jr., 2003: Stratiform rain in the tropics as seen by the TRMM precipitation radar. J. Climate, 16, 1739–1756.
Seifert, P., and Coauthors, 2010: Saharan dust and heterogeneous ice formation: Eleven years of cloud observations at a central European EARLINET site. J. Geophys. Res., 115, D20201, doi:10.1029/2009JD013222.
Shupe, M., and Coauthors, 2008: A focus on mixed-phase clouds. Bull. Amer. Meteor. Soc., 89, 1549–1562.
Smith, A., V. Larson, J. Niu, J. Kankiewicz, and L. Carey, 2009: Processes that generate and deplete liquid water and snow in thin midlevel mixed-phase clouds. J. Geophys. Res., 114, D12203, doi:10.1029/2008JD011531.
Starr, D. O., and S. K. Cox, 1985: Cirrus clouds. Part II: Numerical experiments on the formation and maintenance of cirrus. J. Atmos. Sci., 42, 2682–2694.
Thayer-Calder, K., and D. Randall, 2009: The role of convective moistening in the Madden–Julian oscillation. J. Atmos. Sci., 66, 3297–3312.
Wang, Z., and K. Sassen, 2001: Cloud type and macrophysical property retrieval using multiple remote sensors. J. Appl. Meteor., 40, 1665–1682.
Yang, G.-Y., and J. Slingo, 2001: The diurnal cycle in the tropics. Mon. Wea. Rev., 129, 784–801.
Yasunaga, K., and Coauthors, 2006: Melting layer cloud observed during R/V Mirai cruise MR01-K05. J. Atmos. Sci., 63, 3020–3032.
Yasunaga, K., A. Hashimoto, and M. Yoshizaki, 2008: Numerical simulations of the formation of melting-layer cloud. Mon. Wea. Rev., 136, 223–241.
Zhang, D., Z. Wang, and D. Liu, 2010: A global view of midlevel liquid-layer topped stratiform cloud distribution and phase partition from CALIPSO and CloudSat measurements. J. Geophys. Res., 115, D00H13, doi:10.1029/2009JD012143.
Zhang, M., and Coauthors, 2005: Comparing clouds and their seasonal variations in 10 atmospheric general circulation models with satellite measurements. J. Geophys. Res., 110, D15S02, doi:10.1029/2004JD005021.
Zuidema, P., B. Mapes, J. Lin, C. Fairall, and G. Wick, 2006: The interaction of clouds and dry air in the eastern tropical Pacific. J. Climate, 19, 4531–4544.
