Detection and Retrieval of Supercooled Water in Stratocumulus Clouds over Northeastern China Using Millimeter-Wave Radar and Microwave Radiometer

Abstract

Supercooled water in mixed-phase clouds plays a significant role in precipitation formation, atmospheric radiation, weather modification, and aircraft flight safety. Identifying supercooled water in mixed-phase clouds is a crucial-frontier scientific issue in atmospheric detection research. In this study, we propose a new algorithm for identifying supercooled water based on the multi-spectral peak characteristics in cloud radar power spectra, combined with radar reflectivity factor and mean Doppler velocity. Using microwave radiometer data, we conducted retrieval analyses on two stratocumulus cases in the spring over the northeastern Daxing’anling region, China. The retrieval results show that the supercooled water in the spring stratocumulus clouds over the region is widespread, with liquid water content (LWC) ranging around 0.1 ± 0.05 g/m³, and particle sizes not exceeding 10 μm. The influence of updrafts on supercooled water is evident, with both showing good consistency in spatiotemporal variation trends. Comparing the liquid water path (LWP) variations retrieved from cloud radar and microwave radiometer, both showed good consistency in variation trends and high LWC areas, indicating the reliability of the identification algorithm developed in this study.

1. Introduction

When the temperature in liquid-phase clouds drops below 0 ℃, cloud droplets do not immediately freeze but form supercooled liquid clouds or mixed-phase clouds consisting of a mixture of supercooled water and ice crystals [1,2]. Choi et al. [3] found that at the −20 ℃ isothermal surface, clouds containing supercooled water can account for 50% of the global cloud cover, and a 20% change in the amount of supercooled clouds can significantly affect the radiative effects of clouds. In high-latitude regions, the frequency of supercooled clouds is relatively high. For example, in the Arctic, approximately 30–40% of clouds are mixed-phase clouds containing a large amount of supercooled water [4,5,6]. Over the Southern Ocean, supercooled water is commonly found in long-lasting cold stratiform clouds, contributing to about 30% of shortwave reflective radiation [7]. Over the Antarctic Plateau, the occurrence rate of low-level supercooled clouds is much higher than predicted by model simulations [8].

The formation, development, persistence, and dissipation of supercooled clouds are influenced by both large-scale macro processes and microphysical processes. On a micro scale, the Wegen–Bergeron–Findeisen (WBF) process is one of the primary mechanisms affecting the content of supercooled water. This process leads to the growth of ice crystals at the expense of supercooled water droplets surrounding them [9,10,11]. This mechanism significantly reduces the lifespan of mixed-phase clouds. On the other hand, dynamic processes such as turbulence or entrainment can also affect the cloud lifecycle, where the replenishment of water vapor from the surface or above the cloud layer may contribute to the formation of supercooled water within the cloud. The interaction of these various processes influences the formation and maintenance of ice crystals and supercooled water within supercooled clouds, which is why supercooled clouds can sometimes persist for several days or even weeks [12,13]. Therefore, the detection of supercooled water within supercooled clouds is of great significance.

Aircraft instrument detection is limited by spatial and temporal constraints and the risk of aircraft icing, whereas remote-sensing detection has the advantage of wide spatial coverage and continuous observation. Shupe et al. [14] proposed an algorithm to determine the phase of hydrometeors within clouds using a combination of lidar, 8 mm cloud radar, and microwave radiometer. Their algorithm provides thresholds for millimeter-wave radar reflectivity, mean Doppler velocity, and spectral width corresponding to different types of particles. Based on Shupe et al. [14], Peng Liang et al. [15] established fuzzy logic recognition membership functions using an asymmetric T-shaped function and retrieved the vertical distribution of hydrometeor phases within clouds over Shouxian, China. By analyzing the power spectrum of millimeter-wave radar, it was found that when multiple hydrometeor phases coexist within a cloud, differences in their terminal fall velocities often result in a multi-peak structure in the power spectrum, which can be an important basis for identifying supercooled water. Rambukkange et al. [16] also identified and analyzed supercooled water layers within clouds by recognizing the bimodal characteristics of Ka-band cloud radar power spectra, finding that differences in velocity and reflectivity might be related to the growth modes of ice crystals. In the Tibetan Plateau region of China, Ren Tao et al. [17], based on Ka-band millimeter-wave radar power spectrum data and combined with radiosonde data, proposed an algorithm for identifying and retrieving supercooled water within convective clouds over the plateau. They analyzed the distribution of supercooled water within cumulus clouds over Naqu and found that the reflectivity and particle size of supercooled water in the region were mainly distributed between −25 and 20 dBZ, and 8 and 250 μm, respectively, with supercooled water content ranging from 0.01 to 1.0 g/m³. The formation and development of supercooled water within clouds were mainly influenced by updrafts.

Although the spectral peak characteristics in cloud radar power spectra can serve as an important basis for identifying supercooled water, the supercooled water identification algorithm based solely on this characteristics has certain limitations. When ice crystals begin to form or when the ice crystal volume is relatively small compared to the supercooled water, the multi-peak feature may not be apparent. Additionally, when only supercooled water is present, the power spectrum will also show a single peak only. Therefore, this paper incorporates the relatively smaller reflectivity and mean Doppler velocity of supercooled water compared to larger particles like ice crystals into the supercooled water identification algorithm as supplementary conditions, thereby improving the accuracy of supercooled water identification.

The Greater Khingan Mountains region in northeastern China is a key area for the construction and protection of national ecological barriers. Due to the influence of seasonal atmospheric circulation, precipitation in this region in spring is generally low, and forest and grassland fires occur from time to time. Artificial precipitation enhancement is an important means of forest fire suppression. This region belongs to the cold temperate continental climate zone, with an annual average temperature of −4 ℃. Due to frequent cold-air activities, the occurrence frequency of supercooled clouds is high in this region, and cold cloud precipitation is the main form of precipitation. The supercooled water content in clouds is a key factor in determining the amount of precipitation and the effectiveness of weather modification. However, there is currently little research on supercooled water within clouds in this region.

This study utilizes Ka-band millimeter-wave cloud radar data from the Tulihe Meteorological Station (50°483′E, 121°683′N) in the Greater Khingan Mountains from May to August 2023, combined with microwave radiometer and radiosonde data, to retrieve and analyze supercooled water in two typical stratocumulus processes. This paper aims to provide observational support for the detection and numerical simulation of supercooled water in clouds in the northeastern China region.

2. Data and Methods

2.1. Observation Location

The observation location is at the Tulihe Meteorological Station in Hulunbuir, Inner Mongolia Autonomous Region, with geographic coordinates of 121.683°E and 50.483°N. This meteorological station is situated on the northwest slope of the northern section of the main ridge of the Greater Khingan Mountains, surrounded by mountains on both the north and south sides. The terrain on the east and west sides shows a high-to-low east–west trend, as shown in Figure 1, where the red triangle marks the station’s location.

2.2. Observational Data

This study uses ground-based Ka-band cloud radar and microwave radiometer remote-sensing observational data, as well as conventional meteorological element observations, obtained from the Tulihe Meteorological Station in the Greater Khingan Mountains from May to August 2023. The Ka-band millimeter-wave cloud radar (from Xi’an, China, YLU1 model, Xi’an Huateng Microwave Co., Ltd.) employs an all-solid-state, fully coherent, pulse compression, pulse Doppler, single polarization, vertical pointing method for continuous 24-h observation. The cloud radar data include power spectra, equivalent reflectivity factor, mean Doppler velocity, and spectral width. The main performance parameters are listed in

Table 1. Main parameters of the millimeter-wave cloud radar.

Parameter Name	Parameter Value
Operating Frequency	35 GHz ± 200 MHz
Antenna Scanning Mode	Vertical fixed pointing
Antenna Gain	53 dB
Antenna Aperture	1.8 m
Detection Range	0.15–20 km
Temporal Resolution	5 s
Spatial Resolution	30 m

.

The microwave radiometer (SMR−100 model, Hangzhou Qianhai Technology Co., Ltd., Hangzhou, China. The operating frequency is 22–31 GHz in K-band and 51–58 GHz in V-band.) can detect data such as near-surface temperature, relative humidity, wind speed, and wind direction, as well as the cloud base height and cloud base temperature at higher altitudes. Additionally, through retrieval, it can provide data on temperature, water vapor density, humidity, and liquid water content from the surface to 10 km.

2.3. Cloud Phase Identification and Parameter Retrieval Algorithm Based on Ka-Band Cloud Radar

The presence of supercooled water in clouds is often accompanied by particles such as ice crystals, whose fall speed differences can result in multiple peaks in the radar power spectrum [18]. Therefore, this feature can serve as an effective basis for identifying supercooled water. Based on this characteristics, combined with the radar reflectivity factor and mean Doppler velocity, this paper proposes a new algorithm for identifying supercooled water. The flowchart and specific algorithm description are shown in Figure 2.

2.3.1. Data Processing and Quality Control

The radar-detected signals include not only cloud and rain signals but also noise signals. Therefore, we need to remove the noise signals and isolate the effective cloud and rain signals. The main process is as follows:

• Calculate Noise Level: We use the segment method proposed by Petitdidier et al. [19] to calculate the noise level (P_N). This method involves dividing a long-duration signal into multiple shorter segments and then analyzing each segment to estimate the overall noise level.
• Extract Effective Cloud and Rain Signals: Retrieve all continuous signal segments in the power spectrum that are above P_N. Since the signal-to-noise ratio (SNR) of cloud and rain signals is usually stronger than noise and has a broader distribution range, we follow the method provided by Zheng et al. [20] and set an SNR threshold (TS_snr = −12 dB) and a spectrum point number threshold (TS_num = 5) to identify effective cloud and rain signal segments. All spectrum points that do not meet these conditions are considered noise. The maximum noise power is calculated as the boundary between noise and signal (P_B). The intersections of P_B with the cloud and rain signals on both ends are taken as the start and end points, and the peak value between them is considered as the spectral peak.
• Correct Velocity Ambiguity: When cloud and rain are strong, velocity ambiguity occurs, shown as some cloud and rain signals exceeding the maximum velocity values on both sides of the power spectrum and folding back (as shown in Figure 3a). To avoid the impact of this issue on the accuracy of subsequent microphysical parameter retrieval, we use the de-aliasing method proposed by Zheng et al. [20]. This method corrects for velocity ambiguity by judging the type of ambiguity and its extent based on the principle of continuous change in wind speed with height, from the cloud top downward by layer comparison. The corrected power spectrum is shown in Figure 3b.
• Calculate Radar Base Data through Local Integration: This includes echo power P_r (dBm), reflectivity factor Z_e (dBZ), mean Doppler velocity V_D (m/s), and spectral width σ_v (m/s). The calculation formulae are shown in Equations (1)–(5). Here, v_s and v_e are the Doppler velocities (m/s) at the start and end points of the cloud and rain signals, P_i is the power of the cloud and rain signals (dBm), P_N is the noise level (dBm), and C is the radar constant.

$$P_r={\displaystyle \sum _{i=v_s}^{v_e}(P_i-P_N)}$$

(1)

$$Z_e=10\times {\log }_{10}({\displaystyle \frac{P_r\times R^2}{C}})$$

(2)

$$C=10\lg ({\displaystyle \frac{2.69\times {\lambda }^2}{P_t\tau \theta \varphi }})-2G+160+L_{\sum }$$

(3)

$$V_D={\displaystyle \frac{{\displaystyle \sum _{i=v_s}^{v_e}i\times (P_i-P_N)}}{{\displaystyle \sum _{i=v_s}^{v_e}(P_i-P_N)}}}$$

(4)

$${\sigma }_v=\sqrt{{\displaystyle \frac{{\displaystyle \sum _{i=v_s}^{v_e}{(i-\overline{V})}^2\times (P_i-P_N)}}{{\displaystyle \sum _{i=v_s}^{v_e}(P_i-P_N)}}}}$$

(5)

5.

In addition to radar base data, this study employs a small-particle-tracing method [21,22] to calculate the average terminal fall velocity (V_t) of the particles and subsequently retrieve the vertical air velocity (V_air). The basic principle is that for the smallest particles that can be measured by a highly sensitive millimeter wave radar, their own velocity is essentially negligible compared to the convective wind speed, so their signal can be used as a tracer to reverse the atmospheric vertical velocity [20,23].

To reduce errors caused by radar sensitivity and the minimum particle size within the cloud, the terminal fall velocity of the tracer particles (V_trace) is first estimated. The radar reflectivity factor and number concentration of the tracer particles are considered as Z_trace (dBZ) and N_trace, respectively. Given the significant variation in particle number concentration during different cloud and rain processes, this study follows Ren Tao et al. [17] in classifying the tracer particles: for Z_e less than −15 dBZ, the particles are considered as cloud droplets, with N_trace set to 10⁸ m⁻³; for Z_e equal to −5 dBZ, the particles are classified as drizzle, with N_trace set to 10⁶ m⁻³; for Z_e greater than 10 dBZ, the particles are classified as small raindrops, with N_trace set to 10⁴ m⁻³. For particles with Z_e values between these ranges, the inverse distance weighting method is used to calculate N_trace. Finally, the diameter of the tracer particles (D_trace) is determined using Z_trace through Equation (6).

$$D_{trace}={({\displaystyle \frac{Z_{trace}}{N_{trace}}})}^{1/6}$$

(6)

Next, based on the relationship between the terminal fall velocity and the diameter of the particles (Equation (7)), the terminal fall velocity of the tracer particles (V_trace) is obtained. Finally, as shown in Equations (8) and (9), the vertical air velocity (V_air) can be determined by subtracting the terminal fall velocity of the tracer particles (V_trace) from the Doppler velocity (ω₀) of the tracer particles in the power spectrum. Subsequently, V_t is obtained by subtracting V_air from V_D. Here, g is the gravitational acceleration, μ is the atmospheric viscosity coefficient, taken as 0.01615 g/m/s, and H is the altitude (m).

$$D=\left\{\begin{array}{c}{({\displaystyle \frac{18\mu }{g}}\times V_t)}^{1/2}D<0.1\&A0;\mathrm{mm}\\ {\displaystyle \frac{1}{0.6}}\times \ln {\displaystyle \frac{10.3}{9.65-\frac{V_t}{\delta (H)}}},D\ge 0.1\&A0;\mathrm{mm}\\ \delta (H)=1+3.68\times {10}^{-5}H+1.71\times {10}^{-9}{H}^2\end{array}\right.$$

(7)

$$V_{air}={\omega }_0-V_{trace}$$

(8)

$$\overline{V_t}=V_D-V_{air}$$

(9)

2.3.2. Power Spectrum Analysis

Typically, the temperature range for the presence of supercooled water in clouds is between −40 °C and 0 °C. In most cases, the area where supercooled water exists also contains ice particles. For cloud radar, if only one phase of particles is present within its detection volume, the power spectrum usually exhibits a single-peaked Gaussian distribution. However, when two or more phases of particles are present (such as supercooled water and ice crystals, graupel, etc.), the difference in fall speeds between different particles often results in a bimodal or multimodal power spectrum distribution. Numerous studies have shown that the spectral peak characteristics in the power spectrum can serve as an effective basis for identifying supercooled water in clouds [16,18,24,25,26]. When the fall speed difference between different phases of particles is insufficient to completely separate the two independent spectral peaks in the power spectrum, the saddle point between the two peaks will be on the noise–signal boundary line. This state is referred to as a multimodal peak (as shown in Figure 4a). To separate out a complete single-peaked spectrum, adjustments are needed to symmetrically complete the missing parts from the saddle point downwards according to a Gaussian distribution (indicated by the black dashed line), forming a complete single-peaked spectrum (within the red solid line box). When the fall speed difference is large enough, the two independent spectral peaks are completely separated, with the saddle point below the noise–signal boundary line, referred to as a multimodal peak (as shown in Figure 4b), allowing for the direct separation of a complete single-peaked spectrum (within the red solid line box).

2.3.3. Cloud-Phase Identification

Although the spectral peak characteristics in the power spectrum can be used to determine the presence of supercooled water, when ice crystals begin to form or their volume is relatively small compared to supercooled water, the multimodal peak feature may not be apparent, leading to errors in identifying supercooled water. To prevent this situation from causing errors in supercooled water identification, we utilize the characteristics that the radar reflectivity factor Z_e and mean Doppler velocity V_D of supercooled water are smaller compared to larger particles such as ice crystals. These two variables are used as supplementary conditions for identifying supercooled water.

First, we select data points with multimodal peak characteristics. To prevent the multimodal peak or multiple modes from being caused by the fall speed differences of particles like ice crystals of different sizes at very low temperatures, the selected height range is within 2 km above the freezing layer. Additionally, multimodal peaks or multiple modes in the power spectrum do not appear suddenly but undergo a transition from single peaks to double peaks and back to single peaks [18]. To reduce errors, for the selected data points, we check whether there are more than 4 valid points among the eight neighboring grid points. If there are more than 4 valid points, the current data point is retained; otherwise, it is excluded. At the end, we retained 7075 data points. The distributions of Z_e and V_D are shown in Figure 5, where over 98% of the data points have a Z_e below 0 dBZ, and 81.49% of the data points have a V_D greater than −1 m/s. Based on the threshold values given in previous studies by Shupe et al. [18], particles above the freezing layer with a Z_e less than 0 dBZ and a V_D greater than −1 m/s are ultimately identified as supercooled water.

2.3.4. Retrieval of Microphysics in Supercooled Liquid Region

Assuming the particle size distribution follows a log–normal distribution [27], the relationship between the radar reflectivity factor Z_e, the effective radius of particles R_e (μm), and the liquid water content LWC (g/m³) is as follows:

$$R_e=50\exp (-0.5{\sigma }^2)N^{-1/6}{Z_e}^{1/6}$$

(10)

$$LWC=\rho \left({\displaystyle \frac{\pi }{6}}\right)\exp (-4.5{\sigma }^2)N^{1/2}{Z_e}^{1/2}$$

(11)

where Z_e is the radar reflectivity factor (mm⁶/mm³) of supercooled water separated from the multimodal spectra, ρ is the density of ice or water (g/cm³), σ_v is the logarithmic spectrum width, and N is the particle concentration (1/cm³). The liquid water path (LWP) refers to the total mass of liquid water per unit area in the atmosphere (usually in the vertical direction), and its calculation formula is as follows, where n represents the number of liquid water bins in the vertical direction, corresponding to the height of the bin:

$$LWP={\displaystyle {\sum }_{i=1}^{n}LWC(i)\times \Delta z}$$

(12)

3. Retrieval Results Analysis and Validation

3.1. Vertical Structure of Stratocumulus and Supercooled Water Retrieval Effect

Figure 6a–h show the radar detection and retrieval results for a stratocumulus system case at 13:00 on 18 May 2023. From the overall Z_e echo, the cloud top height is about 6 km, the cloud thickness is about 3 km, and there is a significant echo intensity gradient within the cloud, with the strongest reaching above 10 dBZ at around 4 km. Compared to the second cloud body, the first cloud body has a wider region of large σ_v values, with most areas inside the cloud having a σ_v greater than 0.4 m/s, indicating the possible coexistence of ice crystals and supercooled droplets of different sizes within the cloud, which leads to an overall increase in the σ_v due to their different fall speeds [14]. The mean Doppler velocity V_D (Figure 6c), retrieved atmospheric vertical velocity V_air (Figure 6d), and particle mean terminal fall velocity $\overline{V_t}$ (Figure 6e) show that the first cloud is mainly accompanied with updrafts, with the maximum exceeding 5 m/s, corresponding to the regions of high σ_v; meanwhile, the second cloud has relatively smaller updrafts, mainly concentrated in the central region, and weaker near the cloud top and cloud base weak echo areas, with most of the cloud experiencing downdrafts at around 1.0 m/s. Additionally, the $\overline{V_t}$ is mostly below −1.0 m/s, with the maximum reaching −6.0 m/s, corresponding to the strong Z_e regions and matching the high V_air and σ_v areas.

Figure 6f–h show the identified regions of supercooled water, supercooled water content, and the effective radius R_e (μm) of supercooled cloud droplets. In Figure 6f, the red, blue, and gray regions represent identified supercooled water, mixed-phase, and non-supercooled water regions, respectively. We find that most of the supercooled water is concentrated in the first cloud, with only a small portion scattered in the second cloud, which is somewhat similar to the distribution of V_air, indicating that updrafts play a certain role in the forming and sustaining of supercooled water.

Additionally, from the perspectives of LWC and R_e of the supercooled water, the range of LWC is 0.1 ± 0.05 g/m³, and the maximum R_e does not exceed 10 μm. The distribution of LWC is generally consistent with R_e. In regions with stronger updrafts, the maximum LWC can approach 0.15 g/m³; whereas, in weak echo areas near the cloud top and cloud base where the updrafts are weaker or even exhibit downdrafts, the LWC does not exceed 0.1 g/m³. This further indicates that under the influence of strong updrafts, cloud condensates form and grow to larger sizes through condensation and coalescence, even forming drizzle and small raindrops that precipitate from the air [28]. This is evident from the Z_e echo, which shows a significant trailing feature around 13:20.

Figure 7a–h show radar observations and retrieval results for a case of cumulus (Case 2) on 30 May 2023, from 17:00 to 18:00. As seen in Figure 7a, the cloud top height ranges between 4 and 5 km, with a cloud thickness of approximately 3 km. The overall radar reflectivity (Z_e) is relatively low, with most areas exhibiting Z_e values below 0 dBZ. The high reflectivity region is concentrated within the cloud (between 3 and 4 km). The retrieval results of σ_v indicate that, during the period from 17:10 to 17:35, the internal parts of the cumulus cloud, especially the upper and middle parts, exhibit σ_v values exceeding 0.4 m/s, suggesting the possible coexistence of particles with multiple phases within the cloud. From the retrievals of V_D, V_air, and V_t in Figure 7c–e, it is observed that during the entire period, most of the cloud is filled with updrafts, with the updraft velocity reaching up to 5 m/s near the cloud top. The maximum V_t approaches −4 m/s, corresponding to the high σ_v region.

Furthermore, radar observations and retrievals of supercooled water (Figure 7f–h) show that supercooled water is mainly concentrated within the cumulus cloud during the period from 17:10 to 17:35. The liquid water content (LWC) is relatively low, around 0.1 ± 0.05 g/m³, and the maximum effective radius (R_e) does not exceed 10 μm. A comparison of Figure 7d,f reveals that the distribution of supercooled water corresponds with the updrafts. Under the influence of the updrafts, LWC and R_e are relatively higher compared to other regions. Although there is some distribution of supercooled water near the cloud base, the LWC and R_e values are lower.

Figure 8a–h show the radar detection and retrieval results of a stratocumulus cloud (Case 2) on 30 May 2023. From the overall Z_e echo, it can be seen that the stratocumulus appeared at 20:15 and developed higher, with the cloud top reaching over 6 km and the cloud thickness around 4 km. The high Z_e values are mainly distributed above 3 km, with the maximum reaching 15 dBZ. Additionally, from the values of σ_v, it can be observed that most areas inside the stratocumulus have a σ_v greater than 0.4 m/s, with the maximum exceeding 2.0 m/s at the cloud top, indicating the presence of two or more particle phases in most areas inside the cloud. The retrieved vertical air velocity V_air and particle terminal fall velocity $\overline{V_t}$ show that updrafts filled the entire cumulus, with high-value areas mainly in the upper–middle parts of the cloud, especially around 4.5 km, where the velocity reach a maximum of 7.0 m/s. Similarly, the high-value areas of $\overline{V_t}$ correspond to the high-value areas of V_air, with the maximum particle fall velocity also exceeding 7.0 m/s, indicating strong convection inside the cloud, with hydrometeors continuously growing due to collision and coalescence under the influence of the updraft.

From the distribution of supercooled water in the cloud (Figure 8f–h), it can be seen that under the influence of updrafts, there exists a continuous layer with relatively evenly distributed supercooled water within the cloud, Near the cloud top, due to the lower temperature and the presence of downdrafts, which are not conducive to the formation of supercooled water, it is mostly ice phase or mixed-phase (ice and water). From the LWC and R_e of supercooled water, it can be seen that the content and size of supercooled water did not significantly differ from the previous stratocumulus case, with LWC still around 0.1 g/m³ but exceeding 0.2 g/m³ in some regions. The R_e also did not exceed 10 μm. However, in this case, the distribution of supercooled water was more continuous, occupying a relatively larger proportion of the cloud, the cause of which requires further analysis to determine.

3.2. Distribution of Supercooled Water Regions

To explore the relationship between the distribution of supercooled water and water vapor density across the three cases, we compared the water vapor density distributions during each event (Figure 9a–c). It can be observed that, during the first case, the water vapor density reached its peak before 13:20, even exceeding 5.0 g/m³ near the surface (highlighted by the red solid box). Due to the updrafts, the water vapor density during the first 20 min was higher than that at the same altitudes in the subsequent 40 min. This explains why the supercooled water content in the first stratocumulus cloud was much higher than in the second.

In the second case, the water vapor density peaked between 17:20 and 17:30 (highlighted by the purple solid box), with higher values at the same altitude compared to other time periods. This corresponds to the high-value region of supercooled water distribution in the second process.

Compared to the previous two processes, the water vapor density in the third case was relatively higher, peaking at over 8.0 g/m³ (highlighted by the black solid box). As seen from the figure, between 20:30 and 20:40, the water vapor density at high altitudes (5–6 km) was significantly higher than at other times, around 4.0 g/m³, providing sufficient water vapor for the formation of supercooled water at the cloud top.

Therefore, the differences in supercooled water distribution between the three processes are mainly influenced by the availability of water vapor [5,29].

Furthermore, for single-layer mixed-phase clouds, many processes can affect the water vapor supply, such as turbulence [30] and updrafts [31]. We observed that in regions with little or no supercooled water (such as that in the first case from 13:30 to 13:40 at 3–5 km altitude, in the second case from 17:20–17:25 at 2.5–3 km altitude, and in the third case from 20:45 to 20:55 at 3–4 km altitude), the updrafts were weak or even had downdrafts. Conversely, in areas with strong updrafts, supercooled water was more concentrated, and the liquid water content (LWC) and effective radius (R_e) showed an increasing trend. Many studies have shown that updrafts play a positive role in the formation and development of supercooled water. For example, Stith et al. [32] found that in tropical regions, liquid water content and cloud base concentration significantly increased in areas with updrafts. Lohmann et al. [33] found that, in orographic clouds, when the updraft speed is high enough to exceed the supersaturation threshold for liquid water, both supercooled droplets and ice crystals can grow simultaneously, preventing the consumption of supercooled droplets and extending the lifetime of mixed-phase clouds.

To this end, we analyzed the relationship between V_air and supercooled water. As shown in Figure 10, the variation of LWC and R_e with V_air indicates that as updrafts strengthen, both LWC and R_e increase. This suggests that updrafts promote the condensation of water vapor, increasing the supercooled water content. The supercooled droplets then grow by colliding with each other as they move with the updrafts, forming larger droplets. This finding aligns with Shupe et al. [18], who concluded that there is a correlation between droplet growth and updrafts in mixed-phase clouds.

3.3. Preliminary Validation of Radar Retrieval Results

Studies have shown that microwave radiometers (MWR) provide reliable measurements of cloud water path (LWP) for non-precipitating clouds [34,35]. Therefore, to validate the supercooled water content retrieved from radar data, we utilized LWP data measured by an MWR at the same location.

A comparison of the LWP between the three case studies is presented in Figure 11. In the figure, the blue solid line represents the LWP results from the MWR, while the red solid line represents the LWP results retrieved from radar spectral data. Figure 11a–c show that the retrieval results from both instruments exhibit a similar trend over time. Although the LWP retrieved from radar data is generally lower than the MWR results, both instruments can accurately identify areas with high liquid water content (indicated by the black dashed box). There are several possible reasons for the lower overall LWP retrieved from radar data: (1) The selection range of Z_e and V_D for the single-peak spectra separated from the dual-peak spectra needs further refinement; (2) the radar’s sensitivity limitations may prevent the detection of weak cloud layers or cloud edges, leading to a lower retrieved LWP; (3) since this study only accounted for the LWC in pure supercooled water regions, the LWC in mixed-phase regions was not included, resulting in an overall lower LWP.

Additionally, at certain times (indicated by the black dashed box in Figure 11c), the radar-retrieved LWP may be overestimated. We attribute this to the following two main reasons: (1) When secondary ice production processes occur (e.g., rime splintering), a multi-peaked spectrum may appear, due to the velocity difference between large and small ice crystals. The spectrum peak with a lower fall velocity might be misidentified as supercooled liquid water, leading to an increase in LWP; (2) when convection within the cloud is strong, the spectral width increases, causing some data points to be mistakenly identified as supercooled liquid water, thereby increasing the LWP.

4. Conclusions

Currently, the accuracy of identifying supercooled water in clouds is limited, and there is little research on supercooled water in stratocumulus clouds in northeast China. This paper proposes a supercooled water identification algorithm using Ka-band millimeter-wave radar data and sounding data collected in the Daxing’anling region. The algorithm is applied to analyze supercooled water in two spring stratocumulus cases, and to study the impact of updrafts on supercooled water variations, the radar-retrieved LWP is compared with that from a microwave radiometer. The main conclusions are as follows:

• Under the background of no significant weather processes, supercooled water is widely distributed within stratocumulus clouds in the Daxing’anling region in spring. The liquid water content is generally below 0.2 g/m³, with the effective radius of particles not exceeding 10 μm. Supercooled water is mainly distributed on both sides of the cloud top and bottom, with reflectivity below 0 dBZ.
• The distribution of supercooled water varies between the two cases due to differences in water vapor density. Updrafts play a positive role in the formation of supercooled water within clouds. In areas with stronger updrafts, both the content and size of supercooled water increase, showing a good correlation between them, which aligns with the relationship between updrafts and supercooled water found by other researchers.
• When comparing the LWP retrieved by the supercooled water identification algorithm with the LWP provided by the microwave radiometer, it was found that both showed consistent trends in identifying LWP variations. The algorithm also performed well in retrieving areas with a high supercooled water content. However, when considering the overall LWP, the algorithm’s results show some discrepancies compared to those from the microwave radiometer. This discrepancy can be attributed to the selection of empirical thresholds and radar sensitivity. Additionally, in mixed-phase regions, the power spectrum may not always display clear multi-peak characteristics, and in regions of secondary ice formation, the presence of multi-peak spectra caused by different sizes of ice crystals can affect the algorithm’s accuracy. Overall, while the algorithm demonstrated a certain level of reliability, further improvements are still needed.

The supercooled water identification algorithm, which combines spectral peak characteristics from cloud radar power spectra with radar reflectivity, mean Doppler velocity, and spectral width, has demonstrated a certain level of feasibility for retrieving supercooled water within clouds. Compared to methods that rely solely on empirical formulas or spectral peak characteristics, this combined approach addresses the limitations, such as where supercooled water may exceed empirical thresholds or where spectral peak features are insufficient for identification when only supercooled water is present. If conditions permit, future improvements will involve integrating data from lidar and aircraft observations to better combine these methods and enhance the accuracy of supercooled water identification.