The polarization lidar has a configuration similar to our 532 nm system (Kong and Yi, 2015), but the transmitter employs a frequency-tripled Nd:YAG laser. It produces emissions of ∼ 150 mJ per pulse at 355 nm with a repetition rate of 30 Hz. A Brewster polarizer is added to improve the polarization purity of the transmitting laser (up to ∼ 10 000 : 1). After beam expansion, the beam with a divergence of 0.15 mrad is transmitted vertically into the atmosphere (to the zenith). The backscattered light is collected by a 20 cm Cassegrain telescope. The field of view (FOV) of the receiver is ∼1 mrad. After collimation, the elastically backscattered light passes an interference filter (with a 0.3 nm bandwidth centered at 355 nm) and is then incident on a polarization beam-splitter prism (PBS). To decrease the cross talk between the two orthogonal polarization channels, two additional polarizers are placed on the two output sides of the PBS. The light exiting from the two polarizers is focused onto two photomultiplier tubes (PMTs). The signals from the two PMTs are gathered by a PC-controlled two-channel transient digitizer (TR40–160, manufactured by Licel).

The raw lidar data are stored in both analog and photon counting modes with a range resolution of 3.75 m and a temporal resolution of 1 min. Based on a method originally proposed by Newsom et al. (2009) that was further developed by Zhang et al. (2014), the stored analog and photon-count data are glued to form a reasonable photon-count profile with a large dynamic range. For the cases in this study, the altitude range of signal gluing was ∼ 1.2–3.3 km. The range and temporal resolution of the processed photon-count profiles are 30 m and 1 min, respectively. The starting altitude of the lidar measurements is ∼ 0.3 km, determined based on the overlap of the laser and the field of view of the telescope. The altitude values referenced in this article are all relative to sea level.

The range-corrected lidar signal X is utilized to represent the backscattering intensity (returned laser power) of cloud particles and gravitationally falling hydrometeors (Ansmann et al., 2008). The volume depolarization ratio δv, defined by the ratio of the perpendicular- to parallel-polarized backscatter coefficients, can be obtained from the two-channel lidar signals along with the relative gain of the parallel and perpendicular channels. The relative gain is determined in advance using a conventional method (Freudenthaler et al., 2009). The magnitude of the δv value allows us to identify whether the dominant backscattering is attributed to ice crystals or water droplets in a given backscatter volume (Shupe, 2007). In general, liquid water droplets suspended in the atmosphere are nearly spherical and produce a very low depolarization ratio (close to zero) for single scattering at exactly 180∘, while ice crystals, which are usually nonspherical, generate a quite large depolarization ratio in the 180∘ backscattering direction. For some mid-level stratiform precipitations, gravitationally falling hydrometeors form initially at altitudes above the 0 ∘C isotherm level. They fall often as ice-phase-dominant hydrometeors at subzero temperatures during their early descent. After the falling hydrometeors pass through the 0 ∘C isotherm level, the snowflake-to-raindrop (ice-to-raindrop) transition can yield a shallow layer of relatively smaller lidar echoes (a local X minimum) that is called “lidar dark band” (Sassen and Chen, 1995; Di Girolamo et al., 2012). The lidar dark band can be used to differentiate between the altitudinal regions with ice-containing particles above the dark band and pure liquid raindrops below the dark band.

It should be mentioned that the particle depolarization ratio δp is conceptually a more suitable quantity in discriminating spherical and nonspherical particles (hydrometeors) in virga/cloud than the volume depolarization ratio δv. But, the volume depolarization ratio δv represents the more basic lidar measurement. In order to validly utilize the δv magnitude in discriminating spherical and nonspherical depolarizations, we have examined the relationship between δp and δv. The δp magnitude is a well-defined function of δv, lidar backscattering ratio R and molecular depolarization ratio δm (Cairo et al., 1999). The molecular depolarization ratio δm has a value of ∼ 0.004 in terms of our lidar receiver bandwidth (0.3 nm) (Behrendt and Nakamura, 2002). Information about the R value range is available from a combined consideration of the earlier lidar measurements and our current observations on precipitation-related cloud/virga. The typical values of R for enhanced aerosol load are around 2 and for optically thin clouds up to around 10 (Lampert et al., 2010). The R values are ∼5–8 on the upper part of typical shallow (∼ 400 m thick) evaporating ice virgae (see Fig. 4 in Cheng and Yi, 2020). In this study, the R value should certainly be larger than 7 on the precipitation-related virga layer. Based on the analysis for the δp expression (Cairo et al., 1999) for clouds and virgae, the particle depolarization ratio δp has a quasi-linear dependence on the volume depolarization ratio δv and a very weak dependence on lidar backscatter ratio R (when R≥5). This favorable feature of the functional dependences allows us to utilize δv in discriminating whether the dominant lidar backscattering is attributed to spherical or nonspherical particles in a given backscatter volume. If Rmin is the minimum of the R value range for the clouds/virgae of interest (e.g., Rmin=7 for the precipitation-related virgae in this study), the discrimination criterion of spherical particles expressed by δvz (equivalent to δp<0.1) takes the form (see Appendix A for mathematical derivation) 1δvz<0.1-0.11Rmin+0.1. The discrimination criterion of nonspherical particles expressed by δvz (equivalent to δp>0.2) is given approximately by 2δvz>0.2-0.24Rmin+0.2. As noticed from the right-hand sides of Inequalities (1) and (2), the absolute differences between the discrimination threshold values expressed by δp (0.1 and 0.2) and by δv are small for clouds/virgae with Rmin>7. The unambiguous cloud-phase discriminations based on the volume depolarization ratio δv in earlier literature (Wang and Sassen, 2001; Intrieri et al., 2002; Shupe, 2007; Ansmann et al., 2009; Lampert et al., 2010) have confirmed the functional relationship between δp and δv mentioned above. This allows us to employ δv with very little threshold-value change in discriminating whether the dominant lidar backscattering is attributed to spherical or nonspherical particles in a given backscatter volume. Specifically, at altitudes above the dark band, the δv-based discrimination criteria are δv<0.09 for spherical water drops/droplets and δv>0.17 for ice crystals (based on the above discrimination criteria when Rmin=7), while an enhanced depolarization ratio (δv>0.1) at altitudes below the dark band indicates the presence of large raindrops.

We examined the multiple-scattering-induced depolarization ratio enhancements for an opaque cloud layer composed of dense spherical water droplets by putting a motorized iris on our polarization lidar system. It is indicated that for a receiver FOV of 1 mrad, the enhanced depolarization ratio δv values due to multiple scattering increased from ∼ 0.03 at the X peak altitude to a maximum value of ∼ 0.27 at the weak-signal cutoff altitude with increasing penetration of laser light into the opaque water-droplet cloud layer. Note that for the same receiver FOV (∼ 1 mrad), the multiple-scattering-induced depolarization ratio δv values were all less than 0.04 within the laser light penetration range in a slightly dense water-droplet cloud layer (Hu et al., 2006). Combining the earlier multiple-FOV polarization lidar measurements (Hu et al., 2006) and our similar observations yields a suggestion that for the 1 mrad receiver FOV the multiple-scattering-induced depolarization ratio values larger than 0.10 should result from an opaque water-droplet cloud layer (see Figs. 2 and 4 in Yi et al., 2021). In other words, for the 1 mrad receiver FOV, the vertical structure of hydrometeors and aerosols present above a dense water-droplet cloud layer with δv values larger than 0.1 is undetectable by ground-based lidars. An artificial water-splashing experiment was performed on the lidar roof windows to examine the effects of water accumulation. A comparison of the lidar profiles with and without water accumulation on the lidar roof windows is given in Fig. 1. Enhanced lidar signal (X) and depolarization (δv) values at altitudes around 4.0 km resulted from an optically thick (opaque) water-droplet cloud layer, because there existed a high X value and near-zero δv value (∼ 0.008) on the cloud base (∼ 3.9 km) (Wang and Sassen, 2001), and also there initially existed a monotonic rapid increase in both the values of X and δv with increasing penetration of laser light into the layer. The cloud-related structures shown in both the X and δv profiles were consistent before and after water splashing (particularly, cloud base altitudes). This comparison clearly shows that water accumulation on the lidar roof windows yielded nearly height-independent lidar signal (X) attenuation, and neither the cloud-related X vertical structure nor the profile of the volume depolarization ratio δv (the magnitude and vertical structure) were altered. This result is physically reasonable.

The configuration of the water vapor Raman lidar used in this study is similar to our 45 cm aperture Raman system (Wu and Yi, 2017), but the current Raman lidar shares the same transmitter with our 355 nm polarization lidar depicted above. It detects inelastic Raman backscatter from water vapor at 407 nm and nitrogen molecules at 387 nm as well as detecting elastic backscattered light by using a 20 cm receiver telescope. The water vapor mixing ratio qv, which is defined as the mass ratio between water vapor and dry air in a given volume, can be obtained from the Raman signals representing water vapor and nitrogen molecules (Whiteman et al., 1992). The Raman lidar system was calibrated by corresponding local radiosonde measurements. A comparison analysis showed that the lidar-derived water vapor mixing ratio profiles agree well with the coincident radiosonde data (the relative deviation is less than 10 % when the water vapor field is horizontally homogeneous on a scale of ∼ 20 km). During the daytime, the water vapor Raman signal is quite noisy at high altitudes due to strong sky background light, so the water vapor mixing ratio profiles are available only at altitudes below ∼ 2 km. A similar artificial water-splashing experiment to that described above was performed on the water vapor Raman lidar roof window. Water accumulation on the roof window hardly had an impact on the obtained subcloud qv profiles.

Figure 3 presents the radiosonde profiles that are pertinent to the warm-front cloud at different stages and during precipitation, together with the 1 h mean lidar profiles obtained during the radiosonde launches. The temporally varying cloud properties (e.g., falling cloud base, increasing cloud thickness and variable cloud types) between 20:00 LT on 26 December and 20:00 LT on 27 December 2017 coincided with the classical picture of preceding upglide clouds of an advancing warm-front system. Accordingly, a downgoing moist layer was observed strengthening and broadening with time during this period (Fig. 3b and c). At the cloud base (except cirrus), the relative humidity over ice had values close to the relative humidity threshold of 84 % that is conventionally used to determine the cloud base heights (Wang and Rossow, 1995; Zhang et al., 2018). Furthermore, the radiosonde data showed that the southwesterly wind mostly prevailed at the cloud altitudes (Fig. 3d, e and f), and the air pressure at altitudes of ∼ 0–5 km dropped continuously by ∼3–5 hPa in the period (not shown here), which did belong to the typical warm-front features.

The radiosonde released at 08:00 LT on 28 December 2017 provided measurements of the meteorological conditions when precipitation reached the surface, although the lidar measurements had already terminated (at 05:38 LT) ∼ 2 h earlier. As seen from Fig. 3b (red), the relative humidity reached a maximum of 98 % with respect to water in an altitude range of ∼ 3–4 km, immediately above the tops of the liquid precipitation streaks (at ∼ 3 km; see Fig. 2a and b). Water vapor at altitudes of ∼ 3–9 km was advected from the southwest, as seen in the wind component profiles (Fig. 3f, red). The high water vapor mixing ratios observed at altitudes below ∼ 3 km came from the evaporation of falling raindrops.

The X and δv precipitation streaks were visible in the period between 23:51 LT on 27 December and 05:36 LT on 28 December 2017 (Fig. 2a and b). The streaks extended from the starting height (∼ 0.3 km) of the lidar measurements to an altitude of ∼ 2.88 km when surface precipitation occurred. A lidar dark band (X minimum) appeared persistently on the top of the precipitation streaks at an ∼ 2.88 km altitude except when the dark band was concealed by a drifting small-scale cloud (at 2.2–2.6 km altitudes during 04:18–05:36 LT on 28 December). This is consistent with earlier lidar observations of stratiform precipitation (Sassen and Chen, 1995; Demoz et al., 2000; Roy and Bissonnette, 2001; Di Girolamo et al., 2012). An inapparent local depolarization (δv) minimum was also persistently present at an altitude of ∼ 2.76 km, lying just ∼ 100 m below the dark-band minimum (Fig. 2b). The local δv minimum represented the completion of the melting process of most falling ice/snow particles. Note that the δv value decreased as a whole from the ice/snow (including partially melted large particles) values (0.10–0.34) at altitudes above the lidar dark band to the small liquid drop level (≤ 0.04, far less than the δv-based discrimination threshold value of spherical particles when the lidar backscatter ratio R≥5) at an altitude ∼ 100 m below the dark-band minimum. The lidar dark band definitely differentiates the altitude regions of precipitating ice-containing hydrometeors occurring above and liquid raindrops occurring below. Although the rainfall-induced water accumulation on the roof window of the lidar varied with time, the precipitation streaks and dark band were steadily reasonably displayed in the X and δv time–height plots (Fig. 2a and b). This is consistent with the result of our water-splashing experiment.

To further clarify the microphysical process of precipitating hydrometeors, two sets of representative lidar profiles (X, δv and qv) for the period that precipitation reached the surface (in Fig. 2) are plotted in Figs. 4 and 5. Figure 4 gives three 1 min X and δv profiles from 01:12 to 01:14 LT on 28 December 2017 and a 1 h averaged qv profile centered at 01:13 LT on the same day. The lidar dark band appeared at a 2.88 km altitude at approximately 01:13 LT, while the local δv minimum (<0.04, far less than the δv-based discrimination threshold value of spherical particles when R≥5) was located at a 2.76 km altitude. These altitudes represent a typical lidar signature of the snowflake-to-raindrop transition for a variety of stratiform precipitation events. An ice-containing bright band (ice bright band hereafter) with δv values ranging from ∼ 0.13 to ∼ 0.39 was visible at altitudes ∼ 3.0–3.45 km, just above the lidar dark band (Fig. 4); these altitudes correspond to the “relative lidar bright band” in the literature (Sassen and Chen, 1995; Di Girolamo et al., 2012). The ice bright band peaked on its bottom (∼ 3.0 km). It showed a variable vertical structure and intensity (in both X and δv) on the timescale of minutes, representing the presence of small-scale fluctuations in the precipitating ice crystals and snowflakes. A liquid-water bright band appeared as a layer of relatively large particle backscatter values, located at ∼ 1.50–2.76 km altitudes, just below the lidar dark band (Fig. 4). It is called “weak lidar bright band” in the literature (Sassen and Chen, 1995; Di Girolamo et al., 2012). The δv values in the water bright band were ∼ 0.03–0.06, indicating that the enhanced lidar backscattering therein was caused mainly by high-concentration quasi-spherical raindrops with diameters ≤ 1 mm. The water bright band actually represents a major precipitation-related lidar backscattering layer in the liquid-phase stage of the light precipitation event. The water bright band appeared to have a larger vertical extent (∼ 1.26 km) than that of the lidar ice bright band.

The lidar qv profile (Fig. 4c) shows an enhanced water vapor mixing ratio at altitudes from ∼ 1.7–3.4 km, indicating the subcloud evaporation of precipitating hydrometeors. In particular, qv was maximized (5.95 g kg-1) around the water bright band center (at ∼ 2.34 km), suggesting that this altitude was a primary subcloud evaporation region for this light warm-front precipitation event. Furthermore, the qv values in the water bright band increased as precipitation continued (Fig. 2c). Combining the vertical structures of X, δv and qv in the water bright band (Figs. 2 and 4) yields the suggestion that most falling small-sized raindrops shrunk or vanished in the water bright band due to evaporation, whereas a small portion of large-sized raindrops survived via collision–coalescence processes and fell out of the water bright band.

At altitudes below the water bright band, the precipitation-related lidar backscattering (X) apparently weakened (Fig. 4a, in which the enhanced X values at altitudes from 0.3–0.7 km resulted from the boundary layer aerosols), indicating low-density raindrops there, whereas δv first increased with decreasing height and then decreased after reaching a maximum (0.13–0.16) at an altitude of approximately 0.6 km (Fig. 4b). Here the magnitude and altitude variation of the lidar depolarization ratio δv values allow us to identify where large-sized raindrops form and break up. Falling small-sized raindrops (equivalent diameter ≤ 1.0 mm) are quasi-spherical (Pruppacher and Klett, 1997) and yield small δv values (generally less than 0.1), whereas falling large-sized raindrops (equivalent diameter >2.8 mm) become nonspherical (with flat or hollow bottom in falling direction) (Pruppacher and Klett, 1997) and lead to large δv values (larger than 0.1). In fact, prominent δv peaks (∼ 0.1–0.4) at altitudes of approximately 0.6 km are always observed in the δv profiles related to reaching-surface precipitation in the present light-rain case (Fig. 2). The δv maxima at an altitude of ∼ 0.6 km are much larger than the typical values (< ∼ 0.07) observed by our 355 nm polarization lidar at approximately the same altitude during rainless days. Here we can exclude a possibility that the δv maxima (∼ 0.1–0.4) at ∼ 0.6 km altitude resulted from multiple scattering by dense droplets around this altitude. As mentioned above, for the 1 mrad receiver FOV, a dense water-droplet cloud layer with the multiple-scattering-induced depolarization ratio δv values larger than 0.1 is optically opaque. In contrast to this situation, in our case, when the prominent δv peak (∼ 0.1–0.4) around 0.6 km altitude occurred, the vertical structure of the precipitation streaks at altitudes far above 0.6 km (e.g., ice bright band, lidar dark band and lidar water bright band) was unambiguously detected by our polarization lidar, indicating that the enhanced depolarization ratios around 0.6 km altitude cannot be caused by multiple scattering from dense spherical water droplets therein. Furthermore, since most falling raindrops evaporated and vanished in the liquid-water bright band as indicated by the enhanced water vapor mixing ratio therein and rapidly decreasing lidar signal on the bottom of the water bright band, small droplets at altitudes below the water bright band were hardly dense enough to generate a strong multiple scattering with δv≥0.1. Therefore, our observational results suggest that sparse large raindrops that fall out of the water bright band with higher fall velocities further grow in size by collecting smaller raindrops along their fall paths. They grow to sizes at which spontaneous breakup occurs at an altitude of approximately 0.6 km. In brief, our lidar observations reveal for the first time (to our knowledge) the collision–coalescence growth and subsequent spontaneous breakup of falling raindrops that actually take place in the natural atmosphere. They represent the posterior microphysical processes necessary for the reaching-surface precipitation production. Interestingly, the size maximization of falling raindrops as shown by the strongest nonspherical shapes (maximum depolarization ratio values) always appeared at an altitude of ∼ 0.6 km for a variety of mid-level stratiform precipitations (in light of our observations). Obviously, the explanation to this ubiquitous feature needs further observational and modeling efforts. As seen in Fig. 2b, the boundary layer aerosols had little impact on the δv precipitation streaks. In addition, at altitudes below 1.5 km, the qv values decreased with increasing altitude, reflecting a normal altitude distribution of the boundary layer water vapor.

Based on the radiosonde temperature data obtained at approximately 20:00 LT on 27 December 2017 (Fig. 3a, orange), the 0 ∘C isotherm level was at an altitude of ∼ 3.6 km, and a warm-front-related inversion layer appeared just below the 0 ∘C level with a local temperature maximum (2.2 ∘C) at 3.33 km and a local minimum (1.0 ∘C) at 2.84 km. The lidar dark band (at 2.88 km, with a temperature of ∼ 1.0 ∘C) was located ∼ 720 m below the 0 ∘C level. In comparison with the results reported in the literature (Sassen and Chen, 1995; Demoz et al., 2000; Sassen et al., 2005; Di Girolamo et al., 2012), the observed ∼ 720 m distance of the dark-band minimum to the 0 ∘C level and the low dark-band temperature (∼ 1.0 ∘C) are somewhat peculiar for light-precipitation cases. In the current case, the melting process might be delayed by the temperature structure (with a small lapse rate) of the inversion layer. However, it should be mentioned here that the radiosonde launching site was ∼ 23.4 km away from our lidar site.

Figure 5 presents three 1 min lidar X and δv profiles displaying the time span from 02:30 to 02:32 LT on 28 December 2017 and a 1 h averaged lidar qv profile centered at 02:31 LT on the same day, depicting the microphysical process of precipitating hydrometeors for slightly strong precipitation that reached the surface during the light rain event. Although the water bright band and aerosol backscatter layer below the dark band became evidently weak compared to those seen in Fig. 4a (due to precipitation attenuation), the altitude of the dark-band minimum (2.85 km) was very close to that (2.88 km) obtained from Fig. 4a. The magnitude (∼ 0.03) and occurrence altitude (2.76 km) of the local δv minimum were consistent with the corresponding values (less than 0.04 and 2.76 km, respectively) observed in Fig. 4b. Furthermore, the depolarization maxima (∼ 0.17) associated with reaching-surface precipitation still appeared at an altitude of approximately 0.6 km, which was also similar to that seen in Fig. 4. The observational facts confirm the result gathered from our water-splashing experiment in which the thin liquid-water layer on the roof windows of the lidars caused nearly altitude-independent attenuation on the X profiles and had no effect on the δv profiles. The profile characteristics shown in Fig. 5 are mostly similar to those mentioned above for Fig. 4, but some newly emerging features need to be illustrated. Figure 5 exhibits an ice bright band stronger than the concurrent water bright band. This result is different from our observations obtained at ∼ 01:12 LT (Fig. 4) but is consistent with earlier lidar observations (Sassen and Chen, 1995; Di Girolamo et al., 2012). The ice bright band observed at approximately 02:30 LT had the X maxima at its bottom (at an altitude of ∼ 3.0 km) and a small vertical extent (∼ 0.21 km, from 3.00 to ∼ 3.21 km due to precipitation attenuation). The X maxima corresponded to the local minima of the depolarization ratio (Fig. 5a and b). Interestingly, this inverse relationship between the backscatter and depolarization values on the bottom of the ice bright band is nearly ubiquitous in the precipitation lidar profiles obtained in the present case. Since the depolarization δv showed moderate minima (∼ 0.08–0.10) at an altitude of ∼ 3.0 km (Fig. 5b), the ice bright-band maxima observed at approximately 02:30 LT might reflect backscattering from high-concentration partially melted large particles therein. On the band's altitudinal extension (from 3.06–3.21 km), the markedly enhanced depolarization values (∼ 0.17–0.34) indicate the presence of ice crystals and large snowflakes (Sassen and Chen, 1995; Di Girolamo et al., 2012). The water vapor mixing ratio qv showed slight enhancements at altitudes of ∼ 1.5–3.0 km at approximately 02:30 LT compared with that measured at approximately 01:12 LT.

As seen from the X and δv precipitation streaks at altitudes below ∼ 1.5 km (Fig. 2a and b), precipitation that reached the surface was intermittent. During periods without reaching-surface precipitation, our lidars were able to sample both a complete virga (from the rain to the snow regions) and a shallow mixed-phase cloud layer immediately above the virga under weak optical attenuation conditions. Such an example is shown in Fig. 6. The lidar profiles above the dark band clearly exhibit the typical structure characteristics of a liquid-topped mixed-phase cloud (a shallow liquid cloud layer and ice virga below) (see Fig. 6 in Wang and Sassen, 2001). The mixed-phase cloud top layer (at altitudes of ∼ 4.6 km) was of high X values and very low δv values (∼ 0.01), while lower part of the cloud was characterized by significantly lower X values and higher δv values (with a maximum up to ∼ 0.33). Furthermore, the cloud top layer had a maximum water vapor mixing ratio qv and a temperature of ∼-8.5 ∘C (based on radiosonde data at ∼ 20:00 LT on 27 December). Combining with the schematic representation of commonly observed mixed-phase cloud layers (see Fig. 1 in Bühl et al., 2016), the current observations suggest that the cloud top layer should mainly be composed of liquid droplets (that were not dense enough to yield detectable multiple scattering), and the lower part of the cloud was mainly precipitating ice crystals (falling ice virga). The liquid-topped mixed-phase cloud (a liquid cloud layer and ice virga below) (Bühl et al., 2016) might be fundamental monomers that constitute mid-level precipitating stratiform clouds. According to the expressions on the right-hand sides of Inequalities (1) and (2), the δv-based discrimination threshold values were, respectively, 0.09 for spherical particles and 0.17 for nonspherical particles when the lidar backscatter ratio R had a value of 7 (the minimum of the R value range) on the upper part of the precipitation-related virga (Lampert et al., 2010; Cheng and Yi, 2020). Thus, the δv magnitude of the falling virga increased from the liquid-water values of ∼ 0.02–0.07 (<0.09) at an altitude of 4.38 km to the ice/snow values of ∼ 0.21–0.33 (>0.17) at an altitude of 4.02 km. The falling ice crystals yielded a very weak ice bright band at an altitude of ∼ 3.0 km and then melted into liquid drops at an altitude of ∼ 2.76 km (the local δv minimum). During their further descent, the liquid drops fully vanished due to evaporation, leaving a lidar-detectable rain virga (water bright band) without reaching-surface precipitation. In contrast to the situation during precipitation that reached the surface, no clear-cut δv enhancement occurred at an altitude of approximately 0.6 km when there were only virgae suspended in air. Similar results were discerned for other lidar profiles shown in Fig. 2, in which a complete mixed-phase cloud layer could be detected.

During the light warm-front rain event, since the reaching-surface precipitations and virgae occurred alternately on a small timescale from a few minutes to tens of minutes and since their precipitation streaks had nearly the same dark-band structures (Fig. 2a and b), both reaching-surface precipitation and virgae would come from the same source cloud (because a warm-front cloud system is generally widespread and slowly varying). Reaching-surface precipitation (drizzle) arose when the precipitation rate was high below the shallow water-droplet-dominated cloud layer (apparent source cloud), while virgae without reaching-surface precipitation took place when the subcloud precipitation rate was slightly low. Therefore, the current lidar observations reveal the microphysical process of precipitating hydrometeors related to light warm-front rain. Both reaching-surface rainfall and virgae suspended in air began as ice-phase-dominant hydrometeors fell out of a liquid apparent source cloud layer at altitudes above the 0 ∘C isotherm level. The depolarization ratio magnitude of falling hydrometeors increased from the liquid-water values (δv<0.09) to the ice/snow values (δv>0.20) during the first 100–200 m of their descent. Subsequently, the falling hydrometeors yielded a dense layer with an ice/snow bright band occurring above and a liquid-water bright band occurring below (separated by a lidar dark band) as a result of crossing the 0 ∘C level. In the ice/snow bright band, large particles would form via the cold rain processes (riming and aggregation), because the broad size distributions of the pristine hydrometeors falling out of the apparent source cloud base could lead to local accretion. The production efficiencies of large particles would depend on the magnitude of the rain rate below the apparent source cloud base and size distributions of the pristine falling hydrometeors. The local depolarization minimum (δv≤0.04, far less than the δv-based discrimination threshold value of spherical particles when R≥5) was persistently observed immediately beneath (∼ 100 m below) the lidar dark-band minimum (X minimum). This displayed that the completion of the melting process of most falling ice particles took place at altitudes (hundreds of meters) below the 0 ∘C isotherm level. The liquid-water bright band (with a geometrical thickness of ∼ 1 km) just below the lidar dark band was characterized by enhanced X values and small δv values. There existed a high-concentration moisture (large qv values) in this bright band. These features indicate that the liquid-water bright band resulted from gravitationally falling, dense evaporating liquid drops. In terms of the lidar-measured profiles during reaching-surface precipitation, at altitudes below the water bright band, the precipitation-related lidar backscattering apparently weakened, while δv first increased with decreasing altitude and then decreased after reaching a prominent maximum at an altitude of ∼ 0.6 km. The lidar profiles for the virgae showed narrower and weaker water bright bands than those observed when precipitation reached the surface. Moreover, during virga occurrence, there was no perceptible depolarization enhancement at an altitude of ∼ 0.6 km. By combining the abovementioned lidar observations, a picture on the microphysical processes of falling hydrometeors in liquid-phase stage emerged. After going through the dark band, most falling raindrops shrank or vanished in the water bright band due to evaporation, whereas a few large raindrops survived and fell out of the water bright band when the rain rate below the apparent source cloud base was high enough. The large raindrops might come from both the complete melting of large falling ice/snow particles and collision–coalescence formation in the dense water bright band. Sparse, large raindrops with high fall velocities further grew in size by collecting smaller raindrops along their fall paths. At an altitude of ∼ 0.6 km, the large raindrops grew to the sizes at which spontaneous breakup could occur, yielding reaching-surface precipitation. When the rain rate below the apparent source cloud base was low, nearly none of the large raindrops fell out of the water bright band. Consequently, there were only virgae suspended in air (without reaching-surface precipitation).

The conventional radiosonde profiles associated with the moderate warm-front precipitation and its precursor clouds and the 1 h mean lidar profiles obtained during the radiosonde launches are plotted in Fig. 8. At ∼ 08:00 LT on 4 March 2019, the sky was nearly cloudless (Fig. 8d, blue), and high relative humidity occurred only at altitudes below 1.2 km (Fig. 8b, blue), while the northwesterly wind prevailed at altitudes from 1.7–10.5 km. This indicated that the warm front had not yet reached our lidar site. At ∼ 20:00 LT on 4 March, a moist layer occurred at altitudes ranging from ∼ 4.8 to 8.0 km with increased relative humidity over water of 80 %–95 % (Fig. 8b, green). An evaporating ice virga was observed at altitudes from ∼ 3.6–4.6 km (Fig. 8d and e, green), just below the moisture layer peak. The apparent source cloud of the virga was invisible by lidars due to strong optical attenuation. A potential occurrence region for the apparent source cloud ranged in altitude from 4.8–6.0 km, where the relative humidity was larger than 90 % (Fig. 8b, green). The southerly wind prevailed at altitudes from 0–12 km (Fig. 8f, green), indicating that the moisture layer and altocumulus (photo III in Fig. 7) were precursors of the warm-front precipitation event. The radiosonde profiles obtained at 08:00 LT on 5 March 2019 showed the meteorological conditions during the moderate warm-front precipitation event after the lidar measurements had already terminated (at 00:51 LT on 5 March). As shown in Fig. 8b (orange), the relative humidity over water had values of 97 %–98 % at altitudes from 0–5.65 km, corresponding to a precipitation rate of ∼ 1.8 mm h-1 (rain gauge record) at approximately 08:00 LT on 5 March. The air pressure from the radiosonde data at Wuhan showed a persistent decrease (by ∼ 2–4 hPa at altitudes of ∼ 0–5 km) during the observational period from the precursor clouds to precipitation (between 08:00 LT on 4 March to 08:00 LT on 5 March) that reflected the warm-front passage.

Figure 9 presents three 1 min lidar X and δv profiles measured from 22:20 to 22:22 LT on 4 March and a 1 h averaged lidar qv profile centered at 22:21 LT on the same day; these profiles exhibit the vertical structure of the X and δv precipitation streaks as well as the water vapor mixing ratio at the onset of the moderate warm-front precipitation event. The lidar profiles obtained at 22:20 and 22:21 LT show nearly identical dark-band locations (the X minima is located at ∼ 2.04 km, and the local δv minima is located at ∼ 1.96 km). The dark-band minima appeared ∼ 960 m below the 0 ∘C level at a radiosonde temperature of ∼ 6.0 ∘C. Such a long survival time of falling ice crystals at altitudes below the 0 ∘C level was due to cooling of the surrounding air during their evaporation and melting. At 22:22 LT, a weak X peak occurred at the dark-band altitudes with δv values ranging from ∼ 0.21–0.29, indicating that partially melted large particles passed through the dark band. As seen from Fig. 9, the X and δv precipitation streaks had complicated vertical structures at altitudes below the dark band and showed strong variations on the timescale of minutes. In particular, enhanced depolarization (0.07–0.12) occurred within the water bright band. These profile details confirm that large-sized particles sometimes fell out of the ice bright band during the moderate warm-front precipitation event, concealing the lidar dark band produced by the melting effect of most relatively small-sized particles in precipitating hydrometeors. This effect appears to explain why the lidar dark band became fuzzy for the moderate precipitation event (Fig. 7). Note that the δv maxima (∼ 0.2), which occurred at an altitude of approximately 0.6 km, were slightly larger in the moderate warm-front rainfall than those observed in the light-rainfall warm-front example. This suggests a larger concentration of raindrops of spontaneous breakup sizes around this altitude. The water vapor mixing ratio qv had values ranging from 3.4–4.4 g kg-1 at altitudes from 1.0–3.0 km (from the bottom of the water bright band to the 0 ∘C level).

Figure 10 gives three 1 min lidar X and δv profiles representing the period from 23:37 to 23:39 LT on 4 March 2019 and a 1 h averaged lidar qv profile centered at 23:38 LT on the same day; these profiles exhibit the vertical structures of the X and δv precipitation streaks as well as the water vapor mixing ratio observed when the surface precipitation rate was highest (3.2 mm h-1) during the moderate warm-front precipitation event (yielding thick liquid-water accumulation on the roof windows of the lidars; see photo V in Fig. 7). The ice bright band, dark band and water bright band were roughly discernible in the three 1 min X profiles despite the considerable fluctuations that occurred on the timescale of minutes. Large ice/snow particles occurred on the ice bright band (at an altitude of approximately 2.5 km), because the δv values were larger than 0.3 therein. The dark band located ∼ 700 m below the 0 ∘C level (3.0 km) had δv values ranging from 0.13–0.19 and a temperature of 4.3 ∘C, reflecting that there were partially melted large particles present in the dark band. In the height range of the water bright band, the depolarization ratio increased from ∼ 0.04–0.06 at an altitude of approximately 2.09 km to ∼ 0.12–0.15 at an altitude of 0.9 km, indicating that more large raindrops formed via collision–coalescence processes therein than in the light-rainfall warm-front example (Figs. 4 and 5). The δv maxima observed at an altitude of ∼ 0.6 km were as high as ∼ 0.27–0.35 corresponding well to the high measured rainfall rate of 3.2 mm h-1 (rain gauge record on the ground). As mentioned above, for the 1 mrad receiver FOV, if such large δv values (∼ 0.27–0.35) came from the multiple scattering by a dense water-droplet cloud layer around 0.6 km altitude, the cloud layer would be optically opaque. It would conceal the vertical structure of the precipitation streaks at altitudes above 0.6 km. In contrast to this situation, as seen from Fig. 10, the vertical structure of the precipitation streaks at altitudes above 0.6 km was clearly discerned by our ground-based polarization lidar, indicating that the enhanced depolarization ratios around 0.6 km altitude cannot be caused by multiple scattering from dense spherical water droplets therein. Furthermore, since most falling raindrops evaporated and vanished in the liquid-water bright band as indicated by the enhanced water vapor mixing ratio therein and rapidly decreasing lidar signal on the bottom of the water bright band, small droplets at altitudes below the water bright band were hardly dense enough to generate a strong multiple scattering with δv≥0.1. Therefore, it is suggested that the prominent δv peak at an altitude of approximately 0.6 km reflected the collision–coalescence growth of falling large raindrops and their subsequent spontaneous breakup. The qv values at altitudes from ∼ 0.7–3.0 km ranged from 5.3–7.3 g kg-1 (Fig. 10c), indicating overall moisture enhancement compared to those values measured at the onset of the moderate warm-front precipitation event (Fig. 9c).