This study describes the microphysical properties of high ice clouds (with bases above 5 km) using ground-based millimeter cloud radar cirrus-mode observations over the Naqu site of the Tibetan Plateau (TP) during a short period from 6 to 31 July 2014. Empirical regression equations are applied for the cloud retrievals in which the parameters are given on the basis of a review of existing literature. The results show a unimodal distribution for the cloud ice effective radius re and ice water content with maximum frequencies around 36 μm and 0.001 g m−3, respectively. Analysis shows that clouds with high ice re are more likely to occur at times from late afternoon until nighttime. The clouds with large (small) re mainly occur at low (high) heights and are likely orographic cumulus or stratocumulus (thin cirrus). Further analysis indicates that ice re decreases with increasing height and shows strong positive relationships between ice re (μm) and depth h (m), with a regression equation of re = 35.45 + 0.0023h + (1.7 × 10−7)h2. A good relationship between ice re and temperature T (°C) is found, re = 44.65 + 0.1438T, which could serve as a baseline for retrieval of characteristic ice re properties over the TP.
It is well known that the Tibetan Plateau (TP), also called the third pole owing to its high altitude of more than 4 km on average and large area of approximately 2.4 million m2, has profound influences on both local and global energy and water cycles (Ye 1981; Liu and Chen 2000; Gao et al. 2003; Wang et al. 2008; Molnar et al. 2010; Kang et al. 2010; Yao et al. 2012). Owing to the high altitudes and relatively low latitudes, the TP is one of the most intensive solar radiation regions in the world. However, the TP can quickly cool down as a result of the emission of longwave radiation of the surface associated with its high altitudes. At the same time, the air mass over the TP is only approximately half of that over the adjacent low-level terrain, causing the temperature and humidity over the TP to be approximately twice as sensitive to heat and water vapor as those in the lowlands (Ye 1981). A change in the thermal and moisture structure over the TP could introduce significant impacts on the atmospheric circulation and global climate change, such as changes in the Asian monsoon (Wu and Chen 1985; Zhang et al. 2004; Wu et al. 2007, 2012; Park et al. 2012; Chen et al. 2014). Particularly, in the summertime, the TP is a huge elevated heating source (Yanai et al. 1992; Ye and Wu 1998; Duan and Wu 2005) that exerts significant effects on the atmospheric motion, even continental-scale circulation changes in the upper troposphere (Krishnamurti and Kishtawal 2000; Singh and Nakamura 2009; Xu et al. 2012).
Among the various factors that can affect the solar radiation reaching the surface and longwave radiation emitted to space, clouds play crucial modification roles depending on the radiative properties and altitude (Liou 1986; Ramanathan et al. 1989; Stephens et al. 1990; Li et al. 2006). To understand the role of clouds on climate change, information on cloud properties over various locations is needed. The cloud properties could be obtained from in situ airborne measurements or ground-based or spaceborne remote sensing observations. Although aircraft observations can obtain cloud properties directly with relatively high accuracy, they are limited by high cost and small sample volumes. Therefore, remote sensing, which is cost effective and continuous, has become a popular way to monitor clouds. Whereas satellite remote sensing observations are widely used in climate model evaluation studies, ground-based remote sensing observation is essential to evaluate the satellite observations.
Knowledge of both water vapor and high cloud properties and their variations in space and time is critical to understanding the thermal and dynamic effects of the TP on the atmosphere. Actually, many studies (e.g., Sassen and Wang 2008; Zhang et al. 2015) have shown that ice clouds have a high occurrence frequency and are the dominant cloud phase over the TP region. Cloud properties over the TP have been mainly studied using satellite remote sensing observations (Liu et al. 2002; Gao et al. 2003; Chen and Liu 2005; Li et al. 2006; Sato et al. 2007; Naud and Chen 2010; Tian et al. 2011; Jin and Mullens 2012). Gao et al. (2003) showed the seasonal variation of water vapor mixing ratios and high cirrus cloud frequencies over the TP using 2-yr MODIS satellite observations, indicating that the mean high cloud reflectance over the plateau reaches its maximum in April and minimum in November. Chen and Liu (2005) also found a large number of cirrus clouds from March to April, which were believed to be generated by relatively warm and moist air being slowly lifted over a large area by an approaching cold front and topographic lifting. Li et al. (2006) showed that a great number of high clouds exist over the TP and have a clear seasonal cycle with the largest number and highest frequency in spring and summer. However, via comparison with the surface weather station observations, Li et al. (2006) found that the MODIS/Terra product largely underestimated low clouds over the TP. When using satellite data to analyze cloud properties over the TP, certain issues regarding the large potential bias must be considered. In this study, instead of using satellite observations, we will focus on ground-based active remote sensing observation of clouds in July 2014. These observations were carried out by the third TP Atmospheric Science Experiment.
The clouds over the TP are mainly orographic clouds with frequent precipitation, particularly in summer, along with the occurrence of thin cirrus at high altitudes (Jin 2006; He et al. 2013). As indicated, the TP is a huge elevated heating source in summer, exerting a significant effect on the atmospheric motion. In addition to the dominant cirrus clouds at high altitudes, low-level convective cumulus clouds often occur in the afternoon owing to the rising air motion (Chen et al. 2012; Wang et al. 2015). These convective clouds make a significant contribution to the water accumulation on the TP. The upper parts of these clouds along with high cirrus clouds with heights above 3–5 km are mostly ice in phase owing to the high altitudes of this region. The influence of ice clouds on the radiation balance of the atmosphere strongly depends on their optical and microphysical properties, height, thickness, and temperature.
In summary, The TP plays a key role in Asian climatology and atmospheric circulation. Over the TP, however, vertically and temporally resolved measurements of cloud properties are scarce. This study will mainly investigate the microphysical properties of high-altitude ice clouds. The paper is organized as follows. Section 2 describes the data and cloud retrieval method. Section 3 presents the results of cloud properties over the TP. A summary is provided in section 4.
2. Data and cloud retrieval method
a. Experiment description
Continuous ground-based cloud remote sensing measurements were carried out from 6 to 31 July 2014 over a Meteorological Bureau weather site at Naqu (31.5°N, 92.1°E; 4508 m above mean sea level) over the TP. Figure 1 shows the location of the Naqu site and the topography of the TP. As shown, Naqu is located in the central part of the TP. The instruments used in this experiment include 35-GHz millimeter cloud radar (MMCR) and a ceilometer. Some other instruments have also been used by the site for routine meteorological observations. Considering that this study focuses only on high-altitude ice clouds, only measurements from MMCR have been used in the retrieval of ice cloud properties.
The MMCR as an active remote sensor has proved to be a valuable tool to provide the vertical structure of clouds. It is usually operated in the Ka band (approximately 35 GHz) or W band (approximately 94 GHz) and is more sensitive than conventional weather radars because of its shorter wavelength and higher power; hence, it is suitable for nonprecipitating or weakly precipitating clouds. Through the relationships between equivalent radar reflectivity Ze and ice water content (IWC) or ice effective radius re, IWC and re can be derived from MMCR observations (Sassen 1984; Liao and Sassen 1994; Matrosov 1999; Liu and Illingworth 2000; Mace et al. 2002; Shupe et al. 2005; Hogan et al. 2006; Protat et al. 2007). The MMCR used here operates at a frequency of 33.44 GHz with a peak transmitter power of 50 W. It is calibrated by the factory, and unfortunately we have not done any collocated observations from other instruments to evaluate it onsite. It works in a vertically pointing manner with measurements including radar reflectivity, Doppler velocity, radar spectrum width, depolarization ratio, and power spectra. Spatial and temporal resolutions of radar observations can be up to 30 m and 8.8 s, respectively. To reduce the uncertainties in the original radar signals, these measurements are averaged into a 1-min dataset. The vertical resolution is set as 30 m, so the uncertainties associated with the cloud boundaries are 15 m, which is approximately double that (7.6 m) used by the U.S. Atmospheric Radiation Measurement (ARM) program (e.g., Dong and Mace 2003).
To retrieve cloud properties, information about the cloud phase is important. Most methods use the lidar depolarization ratio, cloud spectral radiation, or cloud temperature. Without measurements of lidar depolarization ratio and cloud spectral radiation, here we simply determine cloud phase based on temperature. Based on the radiosonde measurements, we found that the air temperature over Naqu in July is often approximately 0°C at a height of 1.2 km and below −20°C at a height of 5 km above ground level (AGL; Liu et al. 1999). Zhao et al. (2014b) showed that the ARM microphysical baseline (MICROBASE) cloud retrieval product classified clouds as ice, mixed, and liquid for temperature ranges T ≤ −16°C, −16° < T < 0°C, and T ≥ 0°C, respectively. While several studies (e.g., Garrett and Zhao 2013) have found supercooled liquid clouds at temperatures below 0°C and mixed-phase clouds at temperatures below −16°C, this study focuses on clouds with bases above 5 km (temperature below −16°C) and assumes that they are pure ice clouds.
Note that there are three modes of MMCR observations: boundary mode (BM), cirrus mode (CM), and rain mode (RM). These three modes operate sequentially. The three modes employ different radar parameters and signal-processing techniques, causing inconsistencies among them. Generally, BM is applied for detection of low-level boundary layer clouds and fogs with weak reflectivity and small Doppler velocity; CM is applied for detection of high-level clouds (cirrus, stratus, cumulus, etc.) and has a higher transmit power and detection capability; RM is mainly used for the detection of mid- and low-level precipitation and is characterized by high reflectivity and strong vertical motion. Because of the effects of pulse compression, CM has the highest sensitivity at the height between 2.04 and 15.3 km. A minimum value from −30 to −45 dBZ can be reported by CM, and the instrument uncertainty is about 0.5 dB. As indicated, we consider only high-altitude ice clouds in this study, so we use the MMCR CM observations. Noting that the radar range gate is set as 30 m in this study, we could miss some thin cirrus clouds, particularly those with depth below 30 m. Thus, the properties obtained in this study are mainly for high-altitude ice clouds with depths above 30 m.
b. Cloud retrieval methods
Various cloud retrieval algorithms have been developed for different cloud properties based on different instrument observations, such as the retrieval algorithms listed in Zhao et al. (2012). For ice cloud retrievals, while there are a few physical methods based on the forward models using both lidar and radar, most of the cloud retrieval algorithms based on MMCR only are empirical retrieval algorithms that use the empirical regression equations derived from limited aircraft observations. It should be noted that more independent information could be provided for the retrieval of ice cloud properties by combining observations from MMCR with those from micropulse lidar (e.g., Donovan 2003) or surface solar/IR radiation (Mace et al. 1998; Matrosov 1999). However, without measurements from other instruments like lidar or radiometer, we choose the MMCR-based empirical retrieval method. In the case when microwave radiometer (MWR) observed liquid water paths (LWP) are obtained, the retrieval results could be constrained and thus improved for liquid properties. Without MWR LWP constraints, the distribution of the retrieved cloud properties could be reasonable while the absolute values might be somewhat biased.
In contrast, it is difficult for ice cloud properties from the MMCR observations to be constrained, so the results of the empirical regression equations are generally baseline or characteristic retrievals of ice properties. As in many previous studies (Heymsfield 1977; Sassen 1984, 1987; Heymsfield and Palmer 1986; Liao and Sassen 1994; Atlas et al. 1995; Brown and Francis 1995; Matrosov 1999; Liu and Illingworth 2000; Mace et al. 2002; Shupe et al. 2005; Hogan et al. 2006; Protat et al. 2007), we use the MMCR-based empirical cloud retrieval algorithm by adopting parameters based on a review of the existing literature. However, we should note that almost all of these relationships are determined using limited sets of airborne measurements. When these regression equations are applied to other locations or different kinds of clouds, they could cause relatively large uncertainties. As Matrosov (1999) suggested, the mean IWC predicted from these relationships can vary by 1.5 orders of magnitude.
Before deriving the empirical regression equations, some physical bases are generally applied. By assuming Rayleigh scattering, the liquid radar reflectivity Z is related to the particle size distribution as
where n(r) is the particle number concentration at size r. In general, effective liquid radar reflectivity Ze should be considered for ice clouds because the ice refractive index is different and ice crystals have various shapes (Yang et al. 2002). Actually, the expression equation of Ze should be
where D is the maximum (unmelted) diameter, |Kw|2 is constant for a given wavelength λ, σbsc is the Mie scattering coefficient depending on the maximum diameter, and ρ is the density, which is generally a function of D. With an assumption of exponential particle size distribution, Ze has a power-law relationship with particle size (both D and r), which has been widely used by many retrieval algorithms (Mace et al. 1998; Matrosov et al. 2003). Because the radar reflectivity often spans several orders of magnitude, it is more convenient to use a logarithmic scale, which is defined as 10 logZ and is expressed in units of reflectivity decibels (dBZ). On the basis of the definitions of cloud ice particle re and IWC, they can be expressed as
where ρi is the ice density. The cloud ice particles often follow a gamma or exponential distribution. Based on above descriptions, IWC and ice re are related to the effective radar reflectivity with power relationships,
where a, b, c, and d are empirical parameters.
The general relationships obtained in Eqs. (5) and (6) are related to the ice particle size distribution, ice particle bulk density, and ice particle habit (Matrosov et al. 2003; Delanoë et al. 2007). As Matrosov et al. (2002, 2003) indicated, the parameters in these two equations varies quite modestly for the particle size distributions of the zeroth- (i.e., exponential), first-, and second-order gamma function. However, the ice particle density and ice particle habit varies a lot for different ice clouds, which further influences the relationships between Ze and particle size, and between IWC and particle size. In other words, the parameters in Eqs. (5) and (6) could vary a lot with ice clouds, introducing large uncertainties to the retrieval results.
Table 1 lists the empirical regression equations adopted by several ice cloud retrieval algorithms using radar reflectivity observations from 35-GHz MMCR. Liu and Illingworth (2000) combined several field experiment observations and proposed a widely used parameterization equation for IWC:
Hogan et al. (2006) further related IWC to both effective radar reflectivity and cloud temperature T by deriving the following regression equation:
Although Eq. (8) might be more accurate than Eq. (7), it is not well adopted owing to the limited observations of temperature profiles. Instead, Eq. (7) has been widely used by some well-known retrieval products, including two products shown in the ARM Cloud Retrieval Ensemble Dataset (ACRED; Zhao et al. 2011). For simplicity, this study also adopts Eq. (7) for IWC retrieval over TP, whereas Eq. (8) is used for intercomparison when temperatures from radiosonde profile observations are available.
Figure 2 shows the intercomparison of IWC retrievals from those algorithms listed in Table 1 for a series of MMCR Ze. By defining the “uncertainty” in the IWC retrieval as the mean ratio between standard deviation and means of different IWC retrievals, we find that the “uncertainty” in IWC retrieval is approximately 55.1% on the basis of all 12 retrievals and is 10.7% on the basis of the six retrievals obtained after 1995. We can see that the regression equations obtained by the six studies after 1995 are similar to each other and generally larger than those before 1995. Considering this fact, the fact that Liu and Illingworth (2000) have combined observations from several field experiments, and last the fact that their algorithm was used by many previous studies, the retrievals based on Liu and Illingworth’s (2000) algorithms can serve as the characteristic values and are used in our analysis. An uncertainty of 30%–50% in IWC retrieval has been assumed in this study.
For ice effective radius, the retrievals are very challenging owing to the complexity of ice crystal habits and ice crystal density. Various algorithms exist, but all have large uncertainties. The simplest method was proposed by Ivanova et al. (2001), which is
where T is cloud temperature in degrees Celsius. Equation (9) implies an upper limit of ice re of 37.65 μm, which is often lower in real clouds. Shupe et al. (2005) derived the ice median and mean volume diameter (D0 and Dm) based on effective radar reflectivity using
where e and f are empirical parameters. As shown by Matrosov et al. (2003), ice particle effective radius re is different from D0 and their ratio varies with D0. With some assumptions, Matrosov et al. (2003) gave an approximate relationship between re and D0, which is
Here, we simply use Eq. (11) for the derivation of ice re while some uncertainties exist.
As Table 1 shows, there are very limited regression equations for ice re retrieval as compared with IWC. Equation (10) has been used by one major cloud product in ACRED. Shupe et al. (2005) indicated that f = 0.63, and parameter e generally increases with decreasing temperature with mean values of approximately 0.06 in summer and 0.09 in winter in the Arctic. The Arctic ice cloud temperature in winter is often below −25°C (Garrett and Zhao 2006, 2013). Considering that our study object is high-altitude ice clouds with temperatures generally below −20°C over the TP in summer, we set e = 0.08, and Eq. (11) can be changed as
Assuming that parameter e could vary between 0.06 and 0.09, the uncertainties associated with setting e = 0.08 should be less than 5%.
Table 2 lists the regression equations and parameters used in this study for high-altitude ice cloud retrieval. Ice water path (IWP) can be obtained by integrating IWC through cloud depth. Note that both IWC and ice re are derived from the same observations of Ze without considering the temperature or ice crystal habit dependency. As shown earlier, temperature at the observation site is only measured twice daily. Moreover, radar-only derivations could avoid the problems associated with mismatch of fields of view of different instruments and provide high temporal and vertical resolutions (Matrosov et al. 2003).
Uncertainties in the retrieved cloud properties are unknown because we do not have true values with which to evaluate them. However, we expect that the uncertainties might be no less than those retrieved using the same methods at other locations, which are 10%–50% in ice re and 10%–100% in IWC, as summarized by Zhao et al. (2012). Note that the retrievals of both ice re and IWC are highly dependent on radar sensitivity and calibration. Some thin clouds could be missed and extra uncertainties in the retrievals associated with instrument errors could exist. We next describe the cloud properties obtained using these equations over the TP.
3. Cloud properties
a. Vertical structure of radar reflectivity
The cloud layers were observed based on the vertical structure of radar reflectivity in July 2014. Figure 3 shows the probability distribution function (PDF) of the observed clouds. In contrast to some midlatitude sites at which single-layer clouds dominate (Zhao et al. 2014a), the fraction of multilayer clouds is even larger than that for single-layer clouds over the Naqu site, whereas single-layer clouds are more abundant than those with any other specific number of layers. This is likely associated with the fact that orographic clouds and high-altitude cirrus clouds often coexist in summer over the TP.
For most observation times, a clear bright band was seen at a height of approximately 1.2 km, indicating a 0°C temperature location. Assuming the decreasing rate of temperature with increasing height is approximately −6.5°C km−1, the temperature at 5 km is approximately −24.5°C, which is consistent with the findings by Liu et al. (1999). This vertical temperature distribution has also been proved by the local radiosonde observations as shown in Fig. 4, which indicates monthly average temperatures of approximately between −20° and −24°C at 0800 and 2000 Beijing time (BJT). As indicated earlier, we assume that the clouds above 5 km are pure ice in phase.
b. Characteristics of ice cloud properties
Figure 5 shows the time series of observed clouds, from which we estimated that the high-altitude ice clouds have a frequency of approximately 28%. In contrast, He et al. (2013) found a cirrus cloud fraction of 15% over the TP using micropulse lidar observation at the Naqu site for the period from 19 July to 26 August 2011. Note that the observation in Fig. 5 includes partial orographic ice clouds in addition to cirrus clouds, and the micropulse lidar observation by He et al. (2013) could be underestimated because of the strong attenuation of signals by clouds below 5 km. Generally, the cloud depths of most high-altitude ice clouds observed here are less than 2 km, with a few cases higher. These clouds are mainly composed of two types: cirrus clouds with thin depths and orographic ice clouds with relatively larger depths.
Figure 6 describes the temporal variation of ice cloud observations and properties for clouds with bases above 5 km, including MMCR Ze, ice re, and IWC. As expected from the regression equations, the spatial (vertical) and temporal distributions of ice re and IWC follow the same trend as MMCR Ze. The high-altitude ice cloud re and IWC mainly lie between 30 and 60 μm and between 0 and 0.05 g m−3, respectively. As known, the ice effective radius is generally related to the environmental temperature, vertical velocity, and water supply (Rogers and Yau 1989). The properties of ice re are further investigated by examining its vertical distribution in next subsection. Figure 7 shows the PDF of high ice cloud properties, including cloud ice particle re, IWC, and IWP. A single-mode distribution is found for all three variables. The most frequent values of ice re, IWC, and IWP are approximately 36 μm, 0.001 g m−3, and 1 g m−2, respectively. The retrieved IWC is slightly less than that within thin cirrus clouds over other locations such as the ARM Southern Great Plains (SGP) and the North Slope of Alaska sites (Zhao et al. 2012). One potential reason might be the high altitudes at which both temperature and moisture are low.
With regard to the retrieval of IWC, we have also compared our retrieval results with those retrieved using both temperature and radar reflectivity based on Eq. (8) in section 2b whenever the temperature from radiosonde observations is available. The results are shown in Fig. 8. The results from Ze–T-based retrieval algorithm by Eq. (8) are approximately 10.5% larger than those retrieved in this study. As indicated earlier, the retrieval uncertainties in IWC could be as large as 30%–50%, which is in general larger than the difference between two retrieval methods as shown in Fig. 8. However, we should note that the differences between two algorithms could be much larger for some cases, consistent with those found by the comparison study (Zhao et al. 2012) among various cloud retrievals. Unfortunately, no true IWC values exist that would allow us to determine which algorithm is more reliable. Considering the scarcity of temperature measurements, we choose the Ze-based method and assume it good enough to represent the characteristic properties of high ice clouds.
We further analyzed the diurnal variation of monthly (6–31 July 2014) averaged ice cloud re and IWC with heights above 5 km, which is shown in Fig. 9. The ice clouds with relatively small particle re and IWC mainly occur at time 0600–1300 BJT. This could be related to the occurrence of high cirrus clouds, which generally have smaller ice particles and low IWC. In contrast, the ice clouds with relatively large particle re and IWC mainly occur from afternoon until midnight, when orographic or convective clouds exist (Chen et al. 2012; Wang et al. 2015). In addition to the thermal dynamic feature, the weak bimodal distribution of ice particle re is also likely related to the status of atmospheric circulation and deep convection feature, which are beyond the scope of current study. Last, note that the statistical results shown in Figs. 7 and 9 are based on limited observations for the period of 6–31 July 2014. Whether the phenomena found here are caused by the small number of events needs further investigation when more observations are available in the future.
c. Vertical structure of ice cloud properties
Considering that both ice re and IWC are derived from the MMCR Ze using the regression equations listed in Table 2, Fig. 10 shows the joint MMCR reflectivity–height distribution dBZ−1 km−1 based on the observation data during 6–31 July 2014 over the Naqu site, TP, following the same method as shown in Zuidema and Mapes (2008). We can see that with the increase of height, more clouds with low MMCR Ze occur. Combining with the definitions of ice re and IWC shown in Eqs. (3) and (4) and the approximation of radar Ze expression shown in Eq. (2), both ice re and IWC should also decrease with height (and thus increase with temperature).
Figure 11 shows time-averaged vertical structure of IWC and ice re at 0000, 0600, 1200, and 1800 BJT. Roughly, we can see that both IWC and ice re decrease at heights above 6 km. For heights below 6 km, the vertical variations of IWC and ice re are more complicated, especially at noon and in the afternoon, when convection becomes active. Consistent with the findings from Fig. 9, the ice particle re is smaller at 1200 BJT than at other times. At 1200 BJT, the vertical structure of ice re is more complicated and requires further investigation in the future. Both derived from MMCR Ze, IWC follows the same vertical distribution as ice re. We should note that we have used a fixed re–Ze and IWC–Ze relationship at different height, which has assumed that particle N does not vary with heights. If particle N actually varies with height, this could introduce additional errors to the vertical distributions of ice re and IWC shown in Fig. 11. By averaging over all observation times, the decreasing trend of ice re and IWC is even more evident as shown in Fig. 12, consistent with the vertical distribution of Ze as shown in Fig. 10. Note that, for each height, the root-mean-square errors (RMSE) of ice re and IWC are very small and not considered here. There are at least three potential contributing factors for this decreasing distribution of ice re and IWC with height. First, there is generally more water vapor supply at relatively low altitudes, which could enhance the growth of ice particles. Second, the clouds at low altitudes consist of more orographic ice clouds than those at high altitudes. These clouds generally have stronger uplifting air motion, which can cause ice particles to grow much larger. Third, at a certain height of approximately 8 km, the air temperature can reach −38°C (Fig. 4), at which homogeneous ice formation occurs and many small ice particles could form there. The difference in homogeneous or heterogeneous ice formation could be another reason for the vertical distribution and bimodal distribution of ice particle re.
If the three likely contributing factors do play important roles, the decreasing trend of ice effective radius with heights shown in Figs. 11 and 12 has been explained, and we may expect the decreasing trend of ice effective radius with increasing cloud depth h based on the second contributing factor. Figure 13 shows the scatterplot between ice effective radius and high-altitude ice cloud depth. Note that ice effective radius has been averaged for 30-m bins of cloud depths. Whereas a good correlation between ice re and cloud depth is found for all ranges of cloud depths (brown line in Fig. 13), there are more reliable linear fitting regression lines for cloud depths below and above 1600 m. The linear fitting regression lines between ice particle re and cloud depth are shown in Fig. 13 (blue and green lines) with strong positive correlation coefficients R of 0.99 and 0.93 for h ≤ 1600 m and h > 1600 m, respectively. These linear fitting regression equations are
where re and h are in units of micrometers and meters, respectively. Of the total variation in ice particle re, 98% could be explained by the linear relationship found between the cloud depth and ice particle re for h below 1600 m, and 87% for h above 1600 m. The reason that there is a change for the re–h relationship around 1600 m could be subjective to the linear fitting regression. Using a power polynomial fitting method, we found a better-fitting curve as indicated by the black dashed line in Fig. 13, which is
Based on Eq. (14), we can see that the sensitivity of ice re to h decreases with increasing cloud depth. A likely reason is that when the clouds are thin with less water content, the cloud droplet effective radius could be more susceptible to the availability of water content and the vertical distance that the droplet goes through during its growth. These regression equations could serve as baseline equations for deriving high-altitude ice cloud particle effective radius over the TP.
For the results shown in Fig. 13, note that the regression equations show the relationships between effective radius and cloud depth for only the observation period in July; these relationships might not be generally applicable. During winter, when temperatures are lower than in summer, the distribution of effective radius with cloud depth could be different. In other words, the regressions shown in Eqs. (13) and (14) might be used only for derivation of cloud droplet re in summer over the TP. To be more general, we also examine the relationship between Ze-based ice cloud re and temperature whenever the radiosonde temperature profiles are available. The results are shown in Fig. 14. Note that we have averaged the ice cloud re for each 1°C temperature bin. A good relationship between ice cloud re and T has also been found as
The correlation between ice re and T is very significant, with R = 0.97. This relationship is somehow different from that (red line) found by Ivanova et al. (2001), which is based on aircraft measurements over the ARM SGP site in Oklahoma. Relative to that indicated by Ivanova et al. (2001), Eq. (15) gives larger ice effective radius at the same temperature. Note that the empirical relationship from Ivanova et al. (2001) indicates a maximum ice re of 37.65 μm, as compared with Eq. (15), which indicates a maximum ice re of 44.65 μm. On the other hand, both this study and Ivanova et al. (2001) show a positive relationship between ice re and T, indicating also a positive correlation of ice re between these two algorithms while their characteristic values are different. Equation (15) could also be used for retrieval of ice cloud re over the TP when radar observations are not available.
By reviewing existing empirical regression equations for ice cloud properties, this study adopts suitable empirical regression algorithms to retrieve high-altitude ice cloud properties over the TP in July 2014. Only clouds with bases above 5 km AGL have been examined in this study to ensure that most of these clouds are pure ice clouds. The air temperature at 5 km AGL has been estimated to be approximately between −24° and −20°C. The MMCR observations have shown that multilayer clouds dominate over the TP in July 2014, whereas single-layer clouds occur more frequently than those with any specific cloud layer. Most high-altitude ice clouds considered here have cloud depths of less than 2 km with a frequency of occurrence of approximately 28%.
The ice cloud microphysical properties mainly demonstrate single-mode or weak bimodal distributions in ice re, IWC, and IWP, which is well consistent with the ice properties found over other locations, as shown by Zhao et al. (2012). The ice re, IWC, and IWP over the TP for the study period mainly lie between 35 and 120 μm, between 0.0002 and between 0.01 g m−3, and 0.01 and 50 g m−2, respectively. Whereas it is impossible to know the uncertainties in the high-altitude ice cloud retrievals without true values to evaluate, their uncertainties should be no less than those retrieved using the same methods at other locations, which are 10%–50% in ice re and 10%–100% in IWC, as summarized by Zhao et al. (2012).
The diurnal variation and vertical structure of ice cloud re are analyzed. It is found that high-altitude ice clouds with small re are more likely to occur during the time from morning to early afternoon (0600–1300), and high-altitude ice clouds with large re are more likely to occur from afternoon until midnight. Vertical distributions show a strong decreasing trend of ice re with increasing height. The high-altitude ice clouds with large re mainly occur at lower heights with relatively higher temperatures, which are more likely orographic cumulus or stratocumulus clouds. The high-altitude ice clouds with small re are most likely thin cirrus clouds forming at high altitudes of approximately 8–12 km, the heights at which homogeneous ice particle formation generally occurs. Further analysis about the joint MMCR reflectivity–height distribution indicates that ice re and IWC follow the same vertical distribution, as expected from their retrieval bases.
The study also shows strong positive linear fitting regression relationships between ice cloud re and ice cloud depth for both cases in which h is no more than 1600 m and no less than 1600 m. The much smaller ice re values for thin ice clouds further prove low ice re for thin cirrus clouds. The linear fitting regression equations are re = 35.28 + 0.0023h and re = 36.80 + 0.0013h for h ≤ 1600 m and h > 1600 m, respectively. These linear relationships for h ≤ 1600 m and h > 1600 m can explain 98% and 87% of the total variation in ice re, respectively. Considering that these relationships could be applicable only for high ice clouds observed in summer, a relationship between re and temperature has also been given as re = 44.65 + 0.1438T by using 1°C-bin-averaged values. This relationship can explain 95% of the total variation in ice re, and it could also serve as a baseline for the retrieval of high ice cloud properties over the TP when radar observations are unavailable.
This work was supported by the National Natural Science Foundation of China (NSFC; Grants 91337103 and 41575143), the Ministry of Science and Technology of China through Grants 2013CB955802 and 2012AA120901, the Chinese Program for New Century Excellent Talents in University (NCET), the Fundamental Research Funds for the Central Universities, and the Scientific Research Foundation for the Returned Overseas Chinese Scholars, State Education Ministry. Data acquisition is supported by China Meteorological Administration Special Public Welfare Research Fund (GYHY201406001). We thank all reviewers for their invaluable comments, which have helped us to improve the paper quality.