The Tibetan Plateau (hereafter TP) is the highest plateau in the world with an average elevation greater than 4000 m. It has important impacts on the East Asian atmospheric circulation, climate change, as well as formation and development of disastrous weather events in China (Xu et al., 2015; Xia et al., 2016; Wan et al., 2017). Due to the special thermal and dynamic forcing of the TP, convections and thunderstorms frequently occur in summer over the TP (Zhang et al., 2003). The cloud microphysical characteristics (such as cloud droplet size, cloud droplet number concentration, cloud water formation, etc.) are important for us to understand the clouds. Meanwhile, they are crucial parameters for weather and climate modeling prediction. Therefore, it is important to study the microphysics of cloud over the TP.

At present, there are two common ways to observe clouds: by remote sensing and by flying an instrumented cloud physics aircraft into the clouds. Ground-based cloud observational instruments include laser ceilometer, millimeter wave cloud radar, laser radar, infrared imager; among which the laser ceilometer and laser radar can measure the altitudes of cloud base and cloud top with higher resolution. The millimeter wave radar can detect the microphysical structure within the cloud due to its wavelength close to the size of cloud droplet (Zhang et al., 2016; Wu et al., 2017). Although spaceborne radar and lidar (CloudSat and CALIPSO) can detect vertical distributions of cloud and aerosols along the satellite orbit, their detection range is yet not so wide (Wang et al., 2011; Zhang et al., 2015; Liu and Chen, 2017; Wang et al., 2017). The microphysical quantities such as the number concentration (N_CCN) of cloud condensation nuclei (CCN), cloud droplet size, cloud drop concentration are usually measured by an instrumented aircraft penetrating the cloud (Wang et al., 2013). These onboard and ground-based devices for cloud microphysical detection (Liu et al., 2015; Chang and Guo, 2016) are not only expensive, but also labor consuming as they require complicated maintenance and calibration. For this reason, they are seldom used in the TP where observational stations are sparsely distributed. The cloud base height is mainly observed by human vision, air ballooning, ceilometer, etc. The visual observation is subjective and discontinuous with poor data quality. The air ballooning observation is highly labor dependent. The ceilometer can yield accurate measurements, but it has not been widely applied yet.

Meteorological satellite data have the advantage of global coverage at multiple times with timely acquisition and much convenience, and have been used to retrieve various cloud microphysical properties. In the last 20 years, the retrieval methodology with satellite data in infrared channel has been widely employed to retrieve cloud parameters such as cloud particle effective radius (r_e), liquid water path, optical thickness, cloud top temperature (T_top), cloud water phase, and so on.

Rosenfeld and Lensky (1998) proposed a method to infer the microstructure of convective cloud after investigating the relationship between temperature (T) and r_e (so called T–r_e profile) based on the NOAA/AVHRR (Advanced Very High Resolution Radiometer) data, and this was later used to analyze the properties of a rainstorm and conditions for seeding topographical cloud (Dai et al., 2010; Liu et al., 2011b). The method was also applied for the Moderate Resolution Imaging Spectroradiometer (MODIS) to detect severe convective storms (Rosenfeld et al., 2008) and glaciation temperatures of convective clouds (Rosenfeld et al., 2011), as well as for Fengyun (FY)-3A/VIRR (Visible and Infrared Radiometer) to analyze the properties of convective clouds (Liu et al., 2011a).

The Suomi National Polar-orbiting Partnership satellite (SNPP) was successful launched on 28 October 2011, which provided a new opportunity for retrieval of cloud microphysical properties. The Visible Infrared Imaging Radiometer Suite (VIIRS) onboard SNPP has the resolution of 375 m in the infrared channel, which is three times higher than MODIS. The SNPP/VIIRS demonstrated unique advantages in retrieval small-scale convective cloud at the initial and development stages (Hillger et al., 2013; Rosenfeld et al., 2014b). The central wavelengths of the five high-resolution (375 m) VIIRS channels are 0.64, 0.865, 1.615, 3.745, and 11.45 μm. Unfortunately, without the high-resolution 12-μm channel, it is impossible to calculate water vapor absorption by the split-window (the brightness temperature difference between 11- and 12-μm channels) that is required to retrieve cloud particle effective radius (r_e) of the high-resolution 3.745-μm channel. Rosenfeld et al. (2014b) developed a method for retrieving the high-resolution r_e from VIIRS, which makes it possible to obtain temperature (Zhu et al., 2014) and updraft (Zheng et al., 2015) at the convective cloud base. This technique is also a breakthrough in retrieving microphysical properties of clouds on 1-km resolution, including the retrieval of N_CCN at the convective cloud base (Rosenfeld et al., 2016). The VIIRS data have also been applied to study the glaciation of aerosols in convective clouds (Zhu et al., 2015b).

Zhu et al. (2014) developed a new method to retrieve the base temperature (T_b) of convective clouds in the boundary layer by using the T–r_e profile of VIIRS. It was found that the standard deviation of the retrieved T_b is only 1.1°C compared with observations of ceilometer and soundings in two years at the southern Great Plains of the United States. Zheng et al. (2015) developed a method to retrieve the updraft of the cloud base (W_b) based on T_b retrieved from VIIRS, the cloud base height above the ground (H_b), and surface temperature. The standard deviation of retrieved W_b is 0.41 m s^–1, as validated by the ground-based lidar observation. Zheng and Rosenfeld (2015) further developed a method to calculate W_b by formulating a linear relationship between W_b and H_b, which can be universally applied to the boundary layer convective cloud. The mean absolute error of the calculated W_b is 20% by validation with the observations from the ground and shipborne equipment. Rosenfeld et al. (2014a) retrieved the maximum supersaturation (S_max) and the number concentration of adiabatic cloud droplet (N_d,a) at convective cloud base, based on the T–r_e profile from high-resolution VIIRS data. The bias of retrieved N_CCN from VIIRS is 20%–30%, as verified by ground-based and shipborne CCN counter (Rosenfeld et al., 2016). However, in-situ observations show that in the troposphere, thermal processes that deviate from being adiabatic commonly occur in convective clouds (Brenguier et al., 2011; Min et al., 2012), which may give rise to uncertainties in N_CCN and N_d,a retrievals (Merk et al., 2016; Miller et al., 2016).

On the basis of the previous studies (Rosenfeld and Lensky, 1998; Rosenfeld et al. 2014a, b; Zhu et al., 2014, 2015a; Zheng and Rosenfeld, 2015), the Automatic Mapping of Convective Clouds (AMCC) software package has been developed (Yue et al., 2019) and implemented in this study. The microphysical properties of convective clouds are derived by using AMCC via the VIIRS data from June to August of 2013–17 whenSNPP overpassed the TP. The values are gridded and domain averaged to obtain the microphysical properties on the whole. These results can partly make up for the scarce of instrumented cloud observation and further deepen our understanding on microphysical properties of convective clouds in summer over the TP.

The VIIRS data of each granule scanning the TP for 5 minutes around 0700 UTC are provided by the NOAA’s Comprehensive Large Array-data Stewardship System (CLASS) at www.bou.class.noaa.gov/saa/products/. The temperature, pressure, humidity, and potential height of Final Operational Global Analysis (FNL) data from the NCEP are used as auxiliary data.

The AMCC software package (Yue et al., 2019) includes an improved cloud mask algorithm and has the capability of automatic detection of convective clouds. It consists of major modules of cloud mask, r_e retrieval, T_b retrieval, convective cloud detection, T–r_e analysis, N_CCN retrieval, and quality control. The methodologies of AMCC are briefly introduced as follows.

It is well known that VIIRS receives visible light and infrared radiation from cloud particles near the cloud top. In order to obtain information inside the cloud, T and r_e at tops of convective cloud clusters of various altitudes in a certain spatial area are used to represent T and r_e of different heights within the single cloud, and to infer the microphysical properties of the convective cloud in this area. This is essentially a transform of space and time, assuming that the convective clouds of different states at the same time are similar to a convective cloud at different development stages. The basis of the above hypothe-sis is that the atmospheric environmental variables such as temperature, humidity, pressure, N_CCN, and vertical motion are similar within a certain spatial region. Under the similar H_b, T_b, water vapor, N_CCN, and dynamic conditions, the cloud particles formed by different air parcels that follow the adiabatic ascending are similar in size at the same height. Because the developing convective cloud can be constrained by adiabatic process, T and r_e at each cloud top within the cloud cluster can be used to approximate T and r_e at different heights within the single large convective cloud. Of course, the region of interest (ROI) for analysis cannot be too large or too small. If ROI is too large, a convective cloud of different pro-perties may be included, which is against the adiabatic constraint. If ROI is too small, the number of r_e pixels would be too small to reflect the general condition of the convective cloud in this region (Rosenfeld and Lensky, 1998; Dai et al., 2010). The testing results show that 96 × 96 pixels of 375-m VIIRS (36 km × 36 km at nadir) can be an appropriate ROI for analyzing the microphysical properties of the convective cloud. By following the same thought above, target regions of similar size are examined one by one for the entire VIIRS data. This approach has been validated against a series of observations by aircrafts penetrating clouds (Rosenfeld and Lensky, 1998; Rosenfeld and Woodley, 2000; Andreae et al., 2004; Prabha et al., 2011; Braga et al., 2017).

Since the upward radiation from the ground will pass through thin cloud and distort retrievals, the cloud mask scheme will remove thin clouds and keep thick clouds as the study targets so that the radiation from the ground cannot pass through. The cloud mask scheme is based on the look-up table of minimum reflectance at 0.6-μm high-resolution channel (ρ_0.6h) of thick cloud, whose cloud optical depth is greater than 10. By the radiative transfer model, ρ_0.6h of a thick cloud on land or ocean (with different reflectivities) are calculated under various solar zenith angle (Z_sol), satellite zenith angle (Z_sat), and relative azimuth of the solar and satellite (A_rel); and then the look-up table of the minimum ρ_0.6h of thick cloud is established [please see Section 3c(4) in Yue et al., 2019]. This scheme is a significant improvement to our previous one, in which the pixel of thick cloud is simply determined if its ρ_0.6h is larger than 0.4, consequently leading to excessively eliminated pixels on the cloud boundary and underestimated T_b. The new scheme is more physically rational than the previous one owing to consideration of solar and satellite geometry.

A multiple-layer cloud under the thin cloud may be identified as thick cloud by the cloud mask scheme. These pixels must be excluded by the dynamic threshold defined by brightness temperature difference between 11- and 12-μm channels (Rosenfeld et al., 2014b), because their brightness temperature atop comes from the multiple-layer cloud and the thin cloud. For pixels of the thick cloud, their reflectance at 3.745-μm high-resolution channel (ρ_3.7h) can be calculated from their corresponding brightness temperature, and r_e can then be obtained from the established look-up table based on ρ_3.7h, Z_sol, Z_sat, and A_rel (Nakajima and King, 1990; Rosenfeld and Lensky, 1998; Dai et al., 2010; Rosenfeld et al., 2014b).

As we know, for convective clouds, the cloud droplets are nucleated near the cloud base, and grow by the processes of condensation and collision with ascending airflow. Therefore, the temperature is highest and r_e is smallest at the cloud base. Moreover, r_e increases with decreasing temperature for a developing convective cloud. This relation can be used to identify convective clouds from non-convective clouds in a 96 × 96 pixel region (Lensky and Rosenfeld, 1997; Yue et al., 2019). The T–r_e profile and glaciation temperature (T_g) of convective cloud are obtained by T–r_e analysis (Rosenfeld and Lensky, 1998; Dai et al., 2010; Zhu et al., 2015b). After that, the warmest temperature is taken as T_b of T–r_e profile (Zhu et al., 2014), and H_b above the ground is obtained by linear interpolation using T_b and the vertical profile of FNL temperature and geopotential height. The precipitation initiation depth (D₁₄), a reflection of the cloud’s ability to initiate precipitation, is calculated to be the difference between cloud base and the height when r_e reaches 14 μm (suggesting raindrop formation in warm cloud) (Rosenfeld and Lensky, 1998; Freud et al., 2008; Zhu et al., 2015a). The cloud-top temperature (T_top) and altitude (H_top) can be obtained from the coldest temperature and its height in the T–r_e profile. The cloud depth (D_cld) is H_top minus H_b and surface height.

To avoid the effect of cloud droplet collision, the T–r_e profile near the cloud base with r_e smaller than 20 μm, which mostly likely observes the adiabatic ascending and nucleation, is selected to retrieve the N_d,a (Rosenfeld et al., 2014a) by the following expression:

where LWC_a is the cloud adiabatic liquid water content calculated by an adiabatic elevated parcel from cloud base to isotherm T; Mr_v,a = (4/3)ρπr_v³ is the mass of a adiabatic cloud drop, ρ is the water density, and r_v is mean volume radius. It is known that r_e is equal to 1.08r_v based on observations of Brenguier et al. (2011) and Freud et al. (2011). Considering the air mixing at cloud base (Freud et al., 2011), we take N_CCN = N_d,a/1.3. In addition, the maximum supersaturation at cloud base is computed by the following algorithm:

where C is the coefficient related to temperature and pressure of the cloud base (Pinsky et al., 2012), W_b is the updraft in unit of m s^–1, and equals 0.0009 × H_b (H_b is in unit of m) (Zheng and Rosenfeld, 2015).

The grid resolution of AMCC for convective clouds is about 36 km × 36 km (0.33° × 0.33°) at nadir. The output macro- and microphysical variables for convective clouds at each grid include T_b, H_b, W_b, S_max, N_CCN, D₁₄, T_top, H_top, D_cld, and T_g. The retrieval accuracy is relatively low at the granule edge having larger Z_sat, and greatly affected by cloud shadow especially when the cloud is against the sun and Z_sat is large. To ensure accuracy, only those retrieved variables whose Z_sat is between –20° and 50° are used in the following analyses.

The retrieval of T_b is critical because the accuracy of retrieved T_b directly influences the accuracy of W_b, S_max, and N_CCN. The lifting condensation level (LCL) calculated from sounding is usually used as the height of cloud base. The Third Tibetan Plateau Atmospheric Science Experiment (TIPEX-III) launched additional sounding on 0600 UTC during the summers of 2013–16 in the TP, which supplied data to validate the accuracy of retrieved T_b. The sounding data at Naqu meteorological station (31.48°N, 92.06°E), are used to calculate temperature at the LCL (T_LCL). The T_b values are retrieved by satellite data from SNPP/VIIRS that overpassed the sounding site around the sounding time. Comparison between T_b and T_LCL (Fig. 1) shows that T_b and T_LCL are consistent (with a correlation coefficient of 0.87 and standard deviation of 3.0°C for 37 samples matching well in time). This suggests that T_b from VIIRS is reliable over the TP.

Figure 1 Comparison between the temperature of lifting condensation level (TLCL) at Naqu meteorological station and the cloud base temperature (Tb) from VIIRS.

Figure options

The red–green–blue microphysical composite image (Rosenfeld and Lensky, 1998) of SNPP/VIIRS at 0702 UTC 19 June 2016 (with satellite zenith angle between –20° and 50°) is presented in Fig. 2, showing abundance of convective cloud clusters over the TP, eastern India, and most areas of Burma.

Figure 2 VIIRS RGB composite image of 0.6-μm reflectance (red), 2.2-μm reflectance (green), and 11-μm brightness temperature (blue) on medium resolution at 0702 UTC 19 June 2016. The satellite zenith angle is between –20° and 50°. The meaning of RGB composite is after Rosenfeld and Lensky (1998).

Figure options

Comparison of retrieved T_b from VIIR (Fig. 3a) and T_LCL calculated from FNL 2-m temperature (Fig. 3b) shows that the retrieved T_b is stable and reliable over the ocean, land, and large topography. As shown in Fig. 3a, the convective clouds over India, Bhutan, Bangladesh, Burma, and Thailand are clouds of warm base whose T_b varies from 10 to 25°C. Over the TP, the convective clouds have cold base with T_b of approximately –10°C over southern Qinghai and –5°C over the central Tibet. The value of T_b is relatively high and is about 5°C over Lhasa to Shigatse. Based on the intensive sounding observations at Lhasa and Naqu at 0600 UTC 19 June 2016, T_LCL values at the two sites are 5.3 and 2.1°C, while T_b values from VIIRS at 0702 UTC at these two sites are 5.0 and 2.0°C, suggesting that T_b retrieved from satellite is consistent with T_LCL derived from sounding data.

Figure 3 Distributions of (a) retrieved cloud base temperature (Tb) from VIIRS at 0702 UTC 19 June 2016 and (b) the temperature of lifting condensation level (TLCL) calculated by FNL 2-m temperature over the surface. Both Tb and TLCL are marked by the colored squares in unit of °C.

Figure options

Figure 4 shows the distribution of N_CCN. It is the highest in central Burma, exceeding 2000 mg^–1 in some parts. High N_CCN of about 1000 mg^–1 is also found over Bangladesh and the Brahmaputra River basin, India. The N_CCN is about 500 mg^–1 over Bhutan and the western TP, and is very low in the eastern TP (50–200 mg^–1), equivalent to the in-situ observed N_CCN of 10–110 mg^–1 in He-nan County and Maqu region at the upstream of the Yellow River (Huang et al., 2002). One of the important reasons for this low N_CCN is that less human activity and less pollution occur in these areas. The N_CCN is about 1000 mg^–1 near Lhasa, which reflects the connection of high N_CCN to dense population and urban industry. High N_CCN of 800–1100 mg^–1 appears in southern part of Qinghai, which is attributed to the influence of frequent sand dust there. Although no direct in-situ observation of N_CCN on 19 June 2016 over the TP is available, the distribution of VIIRS retrieved N_CCN agrees well with peo-ple’s up-to-date knowledge, indicating that AMCC has the capability to retrieve microphysical properties of convective cloud over the TP.

Figure 4 Distribution of cloud condensation nuclei concentration (NCCN; mg–1) retrieved from VIIRS at 0702 UTC 19 June 2016. Others are the same as in Fig. 3

Figure options

In order to be representative, all 569 granules of SNPP/VIIRS satellite that overpassed the TP in summer of 2013–17 are selected. For each granule, retrieval is carried out by AMCC for each ROI where only convective clouds are detected (Yue et al., 2019). The sample number is the total number of executable retrievals within each grid of 0.33° × 0.33°, and it must be ≥ 10 when deriving a grid value for a retrieved variable. The distribution of the sample number over all the grids is presented in Fig. 5. For most grids, the sample number is larger than 20 in the western TP and 80 (even 100) in the eastern TP. This suggests that convection is active in the eastern TP and relatively weak in the northern and western TP, which agrees with many previous studies (Zhang et al., 2015; Chang and Guo, 2016; Fu et al., 2016).

Figure 5 Distribution of sample number at 0.33° × 0.33° grids. The colored solid squares represent the sample number

Figure options

The distributions of mean T_b and H_b retrieved from VIIRS are displayed in Fig. 6. Over the TP, T_b is about –5°C with a standard deviation (STD) of 4°C ( Fig. 6a). It is about –2°C in Qaidam basin and 0°C at the edge of the TP with an STD of 4°C. In India, Nepal, and Bangladesh, T_b is about 20–25°C with an STD of 2°C. In western Sichuan basin, T_b is 18°C with an STD of 6°C. As shown in Fig. 6b, H_b is high: about 1800–2200 m over the central and southern TP, 1100 m over the Yarlung Zangbo Grand Canyon, and around 1500 m over the Sichuan basin. The highest (lowest) H_b is 2500 m (700 m) over the Qaidam basin (over Nepal and Bhutan). The above results are consistent with the observations in the TP (Xu and Chen, 2006), where mean H_b over the TP is higher than 1500 m and is hundreds of meters higher than H_b over plains and basins to its east in summer. Also, the results agree with those of previous studies that the cloud base height is about 6–7 km above sea level over 29.5°–35°N, 90°–100°E based on theCloudSat data (Zhang et al., 2015), and the average H_b by ceilometer is 2.38 km over the Naqu area (Chang and Guo, 2016).

Figure 6 Distributions of (a) average cloud base temperature (Tb; °C) and (b) cloud base height above the ground (Hb; m) on 0.33° × 0.33° grids

Figure options

In summer over the TP, T_b is about –5°C, corresponding to cloud base at about 6000 m (550 hPa) above sea level, and the water vapor mixing ratio is about 4.6 g kg ^–1. On the contrast, in the Sichuan basin T_b is 18°C corresponding to 2000 m (850 hPa) above sea level and the water vapor mixing ratio is 15 g kg^–1. Apparently, the water vapor content over the TP is only 1/3 that over the plains/basins to its east. Because the clouds over the TP are located at higher levels, with their main part above the level of 0°C, they are dominated by cold cloud microphysical process. Compared to convective clouds of warm base, the condensation hydrometers are fewer in convective clouds of cold base, and consequently less latent heat is released with smaller convective available potential energy (CAPE), and total precipitation and intensity are also rather small, which is consistent with the conclusion of Fu et al. (2016) that the mean convective precipitation intensity is about 1.6 mm h^–1.

Overall, due to less anthropogenic influence, the mean N_CCN at the cloud base over the TP is 200–400 mg^–1 with an STD of about 200 mg^–1 (Fig. 7) and the mean S_max is about 0.7% (Fig. 8). Around Lhasa, N_CCN is about 500 mg^–1 with an STD of 260 mg^–1 and S_max is about 0.6%. Over the Yarlung Zangbo River basin and southern Tibet, N_CCN is smaller, 100–200 mg^–1, with an STD of 100–200 mg^–1, and S_max is about 0.8%. Over the Qaidam basin, N_CCN is greater, 500–800 mg^–1, with STD of 200–800 mg^–1, which is possibly related to regional dust and dust transport. From south to north along India–Nepal–TP,N_CCN gradually decreases from 1000 to 200 mg^–1 with the increasing terrain elevation.

Figure 7 Distribution of average cloud condensation nuclei concentration (NCCN; mg–1) at cloud base

Figure options

Figure 8 Distribution of average maximum supersaturation (Smax; %) at cloud base

Figure options

Overall, S_max is obviously higher over the TP than over the surrounding plains/basins (Fig. 8). It is 0.8% in the central and western TP, and slightly lower (0.6%) in the eastern and northern TP. However, S_max is very small (about 0.3%) in western Sichuan basin. The condensation growing of warm cloud droplets can be expressed as $r\dfrac{{{\rm d}r}}{{{\rm d}t}} = {G_1}S$ , where r is the cloud droplet radius, t is time, G₁ is a constant in specified environment, S is the supersaturation (Wallace and Hobbs, 2006). The above equation shows that r² is proportional to S. Compared with that over the plains, larger S over the TP leads to faster droplet growth rate and the droplets can easily become large by condensation, which further triggers the collision process making it easy to form precipitation.

The average updraft at cloud base (W_b) is equal to or larger than 1.5 m s^–1 over most of the TP (Fig. 9) and is about 1 m s^–1 over the Yarlung Zangbo River basin and southern Tibet, slightly smaller than the other areas of the TP. W_b is relatively larger (2–2.3 m s^–1) over the Qaidam basin of Qinghai. W_b is about 1 m s^–1over India (the south side of the plateau), obviously lower than that over the TP. The large W_b over the TP is possibly attributed to the release of latent heat from deposition process at the beginning of cloud formation. Because of the cold cloud base and low environmental temperature, latent heat of condensation is small but latent heat of deposition is greater at the beginning of cloud formation (Wang et al., 2002). Xu and Chen (2006) observed an abnormally ascending hot bubble with updraft of 1 m s^–1 by use of a Sodar over the TP. Fu et al. (2016) found that in the TP, there are larger portions of precipitation from “warm” cloud tops whose height of radar echo exceeds 6 km, despite of the environmental temperature being lower than 0°C. It is speculated that updraft is strong in shallow convective clouds over the TP, and brings environmental liquid particles and warm air mass to the high level, forming a warm region in the environment of lower than 0°C. The distribution of W_b shown in Fig. 9 partly confirms this hypothesis.

Figure 9 Distribution of average updraft at cloud base (Wb; m s–1)

Figure options

The distribution of precipitation initiation depth (D₁₄) indicates that D₁₄ is about 1500–2000 m over the TP (Fig. 10). It is not easier to form precipitation if D₁₄ is higher than 4000–5000 m (Zhu et al., 2015a) on the plains. D₁₄ is about 500–1000 m over the Yarlung Zangbo River basin and southern Tibet, which is obviously lower than in other areas and equivalent to that over the ocean, indicating that precipitation can easily form in this region. D₁₄ is 1500–2000 m over the Qaidam basin of Qinghai Province. In India, northern Yunnan Province, and Sichuan basin, D₁₄ is higher than 2000 m and distinctly higher than that over the TP. Higher D₁₄ corresponds to greater N_CCN (Fig. 7), which reflects the influence of air pollution on cloud and precipitation, and it is more difficult to form precipitation for greater D₁₄.

Figure 10 Distribution of average precipitation initiation depth (D14; m).

Figure options

The distribution of mean cloud top altitude (H_top) in Fig. 11 shows that H_top is between 10 and 13 km over the TP. It is 10 km over the Yarlung Zangbo River basin and southern Tibet, slightly lower than that over other areas in the TP. H_top is about 8 km over the Qaidam basin and 6–8 km over India. The 5-yr mean H_top of all retrieval cases is 10.58 km over the site of Naqu. The observed average H_top are respectively 11.41 and 11.62 km by the C-band continuous wave radar and the Ka-band millimeter-wave cloud radar at Naqu in July and August 2014. For the same radar observation period, the H_top calculated by the FY-2E satellite TBB data is only 9.18 km (Chang and Guo, 2016), while the average H_top of 26 cases from VIIRS at Naqu is 11.8 km. Apparently, the H_top of VIIRS is closer to radar observations.

Figure 11 Distribution of average cloud top altitude (Htop; km)

Figure options

The convective cloud depth (D_cld) gradually decreases from 5000 m in the southern TP to 2500 m in the northern TP (Fig. 12), indicating that convective clouds deve-lop vigorously in the southern TP and weakly in the northern TP. The probability for heavy precipitation and disastrous weather is low due to limited cloud depth and small cloud water content in convective clouds over the TP. On the other hand, the lower precipitation initiation depth makes it easy to form precipitation. Thereby, rainfalls over the TP occur frequently with small amount and intensity, and most of them are showery. Chang and Guo (2016) proposed that the rainfall over the TP was mainly showery precipitation with duration less than one hour, which is much shorter than that over the central and eastern plains of China in the same period. Such kinds of precipitation over the TP are associated with the unique characteristics of cloud base, cloud depth, and precipitation initiation depth.

Figure 12 Average depth between cloud base and cloud top (Dcld; m)

Figure options

Glaciation temperature (T_g) is the temperature at which almost all cloud particles are glaciated (Rosenfeld and Lensky, 1998; Rosenfeld et al., 2011; Zhu et al., 2015b), and glaciation depth (D_Tg) refers to the depth between T_b and T_g. T_g is lower than –30°C in central and southern TP, and around –25°C in northern TP and Qinghai Province ( Fig. 13). The latter is possibly due to the ice crystal effect of dust aerosols. The distribution of D_Tg (Fig. 14) is basically consistent with that of D_cld (Fig. 12). D_Tg is about 4000 and 2000 m in the southern and northern TP and is larger than 6000 m surrounding the TP, which indicate that convective clouds over the TP are more prone to be glaciated than those over the surrounding areas. Since T_b of TP is about –5°C and ice particles are dominant in convective clouds, as precipitation occurs, the precipitation particles are still in solid state when they fall out of the cloud base. These particles do not completely melt until they reach the ground, and have no time to break. This is the reason why precipitation of TP is characterized by large raindrop size and small number drop concentration. Chen et al. (2013) showed that the raindrop spectral width over the TP is greater than that on the plain area of the same latitude. Chang and Guo (2016) proposed that large raindrops and ice particles frequently occurred over the TP, and the characteristics obtained from VIIRS may explain the physical mechanism behind the above phenomena.

Figure 13 Average glaciation temperature (Tg; °C)

Figure options

Figure 14 Average depth between cloud base temperature and glaciation temperature (DTg; m)

Figure options

The microphysical properties of convective clouds over the TP in five summers of 2013–17 are retrieved from the high-resolutionSNPP/VIIRS satellite data by using the Automatic Mapping of Convective Clouds (AMCC) Software Package. The following observatio-nal facts and conclusions are concluded.

(1) The cloud base temperatures retrieved from VIIRS are compared with temperatures of the lifting condensation level (LCL) at Naqu meteorological station. They are linearly correlated with correlation coefficient of 0.87 and an STD of 3.0°C, which indicates that AMCC is capable of retrieving T_b and the results are reliable.

(2) The macro- and microphysical properties of convective clouds over the TP are revealed. First, T_b is about –5°C, H_b is 1800–2200 m, and liquid water content is small in the TP clouds. Second, N_CCN is 200–400 mg^–1 with S_max of 0.7%. With small N_CCN and large S_max, the condensation growth of cloud droplet is fast over the TP. Third, D₁₄ is lower, about 1500–2000 and 500–1000 m in the Yarlung Zangbo River basin and southern Tibet, respectively, and the clouds in these areas are prone to precipitating easily. Fourth, H_top is 10–13 km, while D_cld is rather small decreasing from 5000 m in the south to 2500 m in the north. Fifth, T_g is –30°C in the central and southern TP and –25°C in the northern TP, which, in combination with the T_b below 0°C, leads to the domination of ice process in the clouds.

(3) The cloud microphysical properties over the TP determine that the TP precipitation is of high frequency in occurrence, short duration, and small amount, but with large raindrops. In addition, the limited cloud depth and small liquid water content in the cloud make a low probability for heavy precipitation and disastrous weather.

In this paper, the microphysical properties of convective clouds over the TP are investigated based on the macro- and microphysical variables retrieved from the SNPP/VIIRS satellite data. These retrievals not only effectively supplement the insufficient observations over the TP, but also take advantage of multi-channel satellite observations. For this reason, this study is worth of many follow-ups. It is noted that there still exist some deficiencies in the retrievals by the AMCC in the present. First, H_b of satellite retrieval is not evaluated based on ceilometers; instead, T_LCL calculated from soundings is approximated as the cloud base height and used for evaluation. According to the parcel theory, the lifting condensation level can be approximated as the cloud base when the convective clouds are formed by sufficient mixing and ascending from the ground. Comparison of T_b from VIIRS and T_LCL at Naqu well reflects the above theory. Second, some cloud parameters such as N_CCN, D₁₄, and T_g have not been evaluated by airborne or ground-based observations. Third, quality control of the AMCC is still imperfect. In the future, we shall streng-then the field observation, verify the retrieval results with more airborne and/or ground-based observations, and improve the retrieval algorithms and retrieval accuracy step by step, so as to make better use of the advantages of satellite detection.

Acknowledgment. We acknowledge usage of the NOAA CLASS VIIRS L1B data, the NCEP FNL analy-sis data, and the intensive observational data from the TIPEX-III organized by the China Meteorological Administration.