2.1 X-band Doppler dual-polarisation radar

MXPol was installed on top of a building at Gangneung–Wonju National University (GWU) at the coast of the East Sea (Fig. 1). MXPol operates at 9.41 GHz with a typical angular resolution of 1^∘, range resolution of 75 m, non-ambiguous range of 28 km and a Nyquist velocity of 39 m s⁻¹ in dual-pulse pair (DPP) mode or 11 m s⁻¹ in fast Fourier transform (FFT) mode (see Schneebeli et al., 2013, for more details). The scan cycle was composed of three hemispherical range height indicators (RHIs) at 225.8^∘ (in FFT), 233^∘ (in DPP) and 325.7^∘ (in DPP) azimuth. The first two are towards MHS, while the third is perpendicular to this direction following the coast (dashed red lines in Fig. 1). The RHIs were performed with a range resolution of 75 m and an extent of 27.2 km. The cycle was completed by one plan position indicator (PPI) in DPP mode at 6^∘ elevation (dashed red circle in Fig. 1) and one PPI in FFT mode at 90^∘ elevation (for Z_DR monitoring). The PPI at 6^∘ elevation had a range resolution of 75 m and an extent of 28.4 km. The scan cycle had a 5 min duration and was repeated indefinitely. The main variables retrieved from MXPol measurements are the reflectivity factor at horizontal polarisation Z_H (dBZ), Z_DR (dB), K_dp (${}^{\circ }{\mathrm{km}}^{-\mathrm{1}}$), the mean Doppler velocity (m s⁻¹) and the Doppler spectral width SW (m s⁻¹). During the FFT scans the full Doppler spectrum at 0.17 m s⁻¹ resolution was retrieved. A semi-supervised hydrometeor classification (Besic et al., 2016) was applied to the polarimetric variables. A recently developed demixing module of this method (Besic et al., 2018) was also used to estimate the proportions of hydrometeor classes within one radar volume, which allows us to study mixtures of hydrometeors.

2.2 W-band Doppler cloud profiler

WProf was deployed at MHS, 19 km inland from GWU at 789 m a.s.l. WProf is a frequency-modulated continuous-wave (FMCW) radar operating at 94 GHz used to sample the vertical column above the radar using typically three vertical chirps (Table 1). It consists of one transmitting and one receiving antenna. A comparison of MXPol and WProf specifications can be found in Table 2. WProf has an integrated passive radiometer at 89 GHz, which provides the brightness temperature of the vertical column above the radar (see Küchler et al., 2017, for more details).

Table 1Description of WProf chirps.

Download Print Version | Download XLSX

Table 2Specifications of MXPol and WProf.

Download Print Version | Download XLSX

2.3 Multi-angle snowflake camera

The MASC was deployed in a double-fence wind shield at MHS. It is composed of three coplanar cameras separated by an angle of 36^∘. As hydrometeors fall in the triggering area, high-resolution stereographic pictures are taken, and their fall speed is measured. A complete description of the MASC can be found in Garrett et al. (2012). The MASC images were used as input parameters to a solid-hydrometeor classification algorithm. Each individual particle is classified into six solid-hydrometeor types, namely small particles (SPs), columnar crystal (CC), planar crystal (PC), a combination of column and plate crystal (CPC), aggregate (AG), and graupel (GR). In addition, the maximum dimension (diameter of the circumscribed circle) of particles is measured to characterise the size of the snowflakes. A detailed explanation of the algorithm is provided in Praz et al. (2017). One challenge during the measurement of snowflakes in free fall is the contamination from blowing snow. Schaer et al. (2020) developed a method to automatically identify blowing snow particles in MASC images. Despite the presence of a double-fence wind shield during ICE-POP 2018, 31 % of the particles were identified as blowing snow and removed for this study. We make use of particle size distributions estimated from the maximum diameter of particles following the work of Jullien et al. (2019).

Figure 2Sea level pressure (grey contours, labels in hectopascals), dynamical tropopause on the 315 K isentrope (purple lines), IVT (brown contours, labels in kilograms per metre per second; only values greater than 500 $\mathrm{kg}{\mathrm{m}}^{-\mathrm{1}}{\mathrm{s}}^{-\mathrm{1}}$ are shown) and precipitation rate (PR) in millimetres per hour (colour-filled) at (a) 18:00 UTC on 27 February 2018 and (b) 00:00 UTC, (c) 06:00 UTC and (d) 12:00 UTC on 28 February 2018 from ERA5 data. The warm and cold fronts at 06:00 UTC were manually drawn based on an analysis of equivalent potential temperature at 850 hPa. The red star shows the location of GWU.

2.4 Warm conveyor belt trajectory computation

We define the large-scale WCB ascent over the Korean Peninsula from a Lagrangian perspective as a coherent ensemble of trajectories with an ascent rate of at least 600 hPa in 48 h (Wernli and Davies, 1997; Madonna et al., 2014). The 48 h trajectories were computed with the Lagrangian analysis tool LAGRANTO (Wernli and Davies, 1997; Sprenger and Wernli, 2015) based on the 1-hourly 3D wind field of the hydrostatic model IFS (model version Cy43r3, operational from July 2017 to June 2018; ECMWF, 2017). The details of the microphysics scheme can be found in Forbes et al. (2011). The IFS is run with a spatial resolution of O1280 (approximately 9 km) and 137 vertical hybrid pressure–sigma levels (ECMWF, 2017). The IFS data set combines operational analyses at 00:00, 06:00, 12:00 and 18:00 UTC with hourly short-term forecasts in between. The data set was interpolated to a latitude–longitude grid with 0.5^∘ spatial resolution for trajectory computation. To analyse the air parcel ascent in the vicinity of the measurement site, we combine 24 h backward and 24 h forward trajectories starting every hour between 00:00 UTC on 27 February and 00:00 UTC on 1 March 2018 every 50 hPa from 1000 to 200 hPa in the proximity of Pyeongchang. WCB trajectories are subsequently selected as trajectories exceeding an ascent rate of 600 hPa in 48 h. In addition to this Lagrangian representation (i.e. following the trajectories), we projected the WCB trajectories in an Eulerian reference frame above Pyeongchang. This gives for each time step and for all vertical levels the position of trajectories above Pyeongchang which ascended with ascent rates of at least 600 hPa in 48 h within the period between 00:00 UTC on 27 February and 00:00 UTC on 1 March 2018. This highlights the heights at which WCB air parcels (i.e. with strong ascent at any given time) occur.

Figure 3WCB trajectories with an ascent rate of at least 600 hPa in 48 h that are located near the measurement site at 06:00 UTC on 28 February 2018. Colours indicate (a) the pressure level and (b) the LWC along the WCB air parcels. Also shown are sea level pressure (grey contours, every 5 hPa) and the dynamical tropopause at 315 K (purple). Only the WCB trajectories close to the site (green asterisk in panel a and red in panel b) are shown.

5.1 Vapour deposition: 03:00 to 04:00 UTC

The period from 03:00 to 04:00 UTC is dominated by crystals above 2000 m (Fig. 5c). At this time Pyeongchang is located ahead of the warm front. The vertical profiles of polarimetric variables (Fig. 7) show an increase in Z_H of 2 dBZ from 6000 to 2000 m, while Z_DR is almost constant from 6000 to 3000 m and then subsequently decreases slightly. This likely indicates the onset of aggregation at 3000 m, below which temperatures are greater than −10 ^∘C and hence represent favourable conditions for aggregation (Hobbs et al., 1974). K_dp values are almost zero, suggesting that the number concentration of oblate crystals is low. In summary, this period is characterised by the presence of crystals in limited concentration, which grew by vapour deposition and likely aggregated below 3000 m.

Figure 7Vertical profiles of quantiles of Z_H, Z_DR, K_dp and SW from MXPol from 03:00 to 04:00 UTC on 28 February 2018. The red line shows the median, while the solid blue lines show the 25th and 75th percentiles and the dotted blue lines the 5th and 95th percentiles. The statistics are based on 11 RHIs towards the MHS site. The data within a horizontal distance of 7 to 20 km from MXPol are selected.

Download

Figure 8(a) Selection of in-focus MASC images. (b) Normalised size distribution and classification of 55 particles observed from 03:00 to 04:00 UTC. SP represents small particles, CC columnar crystals, PC planar crystals, AG aggregates, GR graupel, and CPC a combination of planar and columnar crystals. The size distribution is normalised by the number of particles. The pictures of panel (a) were selected to be representative of the classification. The average temperature was 1.5 ^∘C (Fig. 5e).

Download

The selection of snowflake images (Fig. 8b; collected at an average temperature of 1.5 ^∘C) mainly shows small aggregates and crystals of about 2 mm in their maximum dimension. They are partly melted (liquid water is less reflective than ice and creates the dark areas on the snowflake pictures), and riming is indicated by the brighter areas. The size distribution shows that most particles are below 5 mm in size, with a median of 2.2 mm. The classification shows that 64 % of the observed hydrometeors are aggregates, while above 2000 m, the MXPol hydrometeor classification shows predominantly ice crystals (Fig. 5c). This supports our previous conclusion that below 3000 m, when the temperature increases and aggregation is more efficient, a large fraction of crystals aggregate. A total of 7 % of particles were identified as graupel (Fig. 8b), which we could confirm by a visual analysis. Again the riming could have taken place below 2000 m, which explains why no rimed particles are present from 03:00 to 04:00 UTC in the MXPol hydrometeor classification of Fig. 5c. Figure 9a shows the distribution of the riming index (0= no riming; 1= graupel; Praz et al., 2017). The mode around 1 corresponds to the graupel particles, while half of the particles had a riming index smaller than 0.4. This shows that except for the few graupel particles, the other hydrometeor classes did not feature significant riming in comparison with other periods of the event.

Figure 9Distribution of the riming index for all particles between (a) 03:00 and 04:00 UTC, (b) 06:00 and 08:00 UTC, and (c) 09:00 and 11:00 UTC. The dashed lines show the lower and upper quartiles; the solid line shows the median of the distribution.

Download

5.2 Embedded updraughts, riming and aggregation: 06:00 to 08:00 UTC

The period from 06:00 to 08:00 UTC is characterised by embedded updraughts (Fig. 5b), a layer with strong vertical wind shear at 3800 m (Fig. 6c, d) and significant riming, as seen by MXPol (Fig. 5c). From 6000 to 4800 m the crystals grow by vapour deposition, leading to an increase in both Z_H and Z_DR (Fig. 10) as particles grow mainly along their longest dimension, which results in larger and more oblate crystals (Schneebeli et al., 2013; Andrić et al., 2013; Grazioli et al., 2015). The median of K_dp increases to only 0.4 ${}^{\circ }{\mathrm{km}}^{-\mathrm{1}}$, suggesting that the number concentration of oblate particles is small. The temperature in this layer varies from −23 ^∘C to −16 ^∘C, and the air is slightly above saturation with respect to ice (Fig. 6a, b). This represents favourable conditions for depositional growth of sectored plates (Lohmann et al., 2016; Fig. 8.15), while aggregation is unlikely to dominate within this temperature range according to Hobbs et al. (1974). However, we cannot rule out the formation of early aggregates, which at this stage would be oblate and hence contribute to the increase in Z_DR (Moisseev et al., 2015).

At 4800 m we observe a peak and subsequent decrease in Z_DR, which marks the end of growth dominated by vapour deposition. We hypothesise that aggregation starts at this altitude. First, snowflakes tend to be less oblate and less dense after aggregation, which explains the decrease in Z_DR. Second, aggregation increases the size of snowflakes, and hence Z_H continues to increase.

The observed increase in K_dp below the peak in Z_DR is a commonly observed but not fully understood feature. Andrić et al. (2013) proposed that secondary ice generation of small oblate crystals could explain the observed enhanced K_dp values. Concentration of secondary ice particles can be much larger than the number of snowflakes they originate from, which would affect K_dp more strongly than Z_DR since the former is more sensitive to concentration. Moisseev et al. (2015) suggested that it is the result of the onset of aggregation, producing early aggregates that are relatively oblate. Our hypothesis is that first, the generation of secondary ice by droplet shattering (Mason and Maybank, 1960) and by ice–ice collisions (Vardiman, 1978; Takahashi et al., 1995; Yano and Phillips, 2011) contribute to the increase in K_dp below 5000 m a.s.l. Droplet shattering shows a maximum occurrence at −17 ^∘C (Leisner et al., 2014), which corresponds to the altitude where LWC is converted into IWC (Fig. 4 around 5000 m a.s.l.). Secondary ice generation by ice–ice collision is most effective at −15 ^∘C (Takahashi, 1993) and may also take place at around 5000 m a.s.l. Second, riming of already-oblate crystals will tend to increase K_dp because in the early stage of riming the cavities in the crystals are filled, increasing the density (and thus the dielectric response) of the hydrometeors without changing their aspect ratio. Third, rime splintering by the Hallett–Mossop process (Hallett and Mossop, 1974) below 2500 m a.s.l. (temperature above −8 ^∘C) can contribute to maintain high K_dp values. Finally, Korolev et al. (2020) recently suggested that secondary ice produced by shattering of freezing droplets transported above the melting layer could be lifted to higher levels. This may enhance the concentration of secondary ice in regions of strong updraughts. Note that the higher K_dp values compared to the period 03:00–04:00 UTC are primarily due to the increase in precipitation intensity, but the fact that K_dp increases below the onset of aggregation cannot be explained by precipitation intensity only since aggregation decreases the number concentration and the oblateness of the particles. Riming will initially increase K_dp by first filling cavities and hence increasing the density of particles but will later lead to a decrease in K_dp as the rime mass will smooth the particles' shape. There has to be a mechanism which produces a high number concentration of oblate particles to explain an increase in K_dp in a layer dominated by aggregation and riming, and secondary ice production is a good candidate.

Figure 10Same as Fig. 7 for 06:00 to 08:00 UTC on 28 February 2018. The statistics are based on 21 RHIs towards the MHS site. The dashed black line shows the lower limit of the WCB (black contour in Fig. 5a, b, c).

Download

At 3800 m a strong vertical wind shear in the lower part of the WCB (black contour in Fig. 5a, b, c at 06:00 UTC) can be observed in both the radiosoundings (Fig. 6c, d) and the spectral-width profiles (Fig. 10). The jet at 6500 m at 06:00 UTC (Fig. 6c) can be seen as an enhancement of Doppler velocity between 4000 and 6000 m a.s.l., with a maximum of 45 m s⁻¹ at 5000 m in the RHI of Fig. 11. This is in good agreement with the wind speed measured by the radiosonde at 06:00 UTC since the RHI is almost aligned with the wind direction. The vertical wind shear at 3800 m is visible as the Doppler velocity decreases in the lower part of the WCB and reaches a value of 0 m s⁻¹ at 3000 m (combined effect of a decrease in wind speed and change in wind direction from parallel to perpendicular to the radar beam). This vertical wind shear may generate Kelvin–Helmholtz instabilities, which can trigger embedded convection (Hogan et al., 2002). Moreover orography might also play a role in lifting the easterly low-level flow, which directly impinges the Taebaek mountains from the East Sea. These sources of lifting together with the moist neutral layers below 3000 m (Fig. 6a) can lead to the observed strong updraughts (Fig. 5b), which promote aggregation by increasing the probability of collision between particles. The effect of turbulent cells on aggregation has been discussed thoroughly in Houze and Medina (2005), Medina et al. (2005), and Medina and Houze (2015). Houze and Medina (2005) suggested that overturning cells promote both aggregation and riming. First, they can sustain the production of SLW necessary for riming. Second, the turbulence increases the probability of collision between particles. Finally, aggregates are larger targets for the collection of SLW droplets, which again enhances growth by riming. While the cause of the turbulence is different here, the processes described are consistent with our measurements.

Figure 11RHI of Doppler velocity at 11^∘ azimuth at 06:25 UTC from MXPol radar. The dashed black line shows the lower limit of the WCB (black contour in Fig. 5a, b, c). The Python ARM Radar Toolkit (Py-ART; Helmus and Collis, 2016) was used to plot the radar data.

Download

Figure 12Same as Fig. 8 for the period 06:00 to 08:00 UTC. The number of particles is 589. The average temperature is 0.1 ^∘C (Fig. 5e).

Download

The MASC images (Fig. 12a) show mainly rimed aggregates of about 10 mm in their maximum dimension and two graupel particles. The average temperature of collection was 0.1 ^∘C, and hence the particles should not be as melted as during the period 03:00–04:00 UTC. The classification shows a majority of aggregates (77 %). The hydrometeor classification from Besic et al. (2018) classifies rimed aggregates and graupel as rimed particles, whereas the MASC-based classification from Praz et al. (2017) classifies only fully rimed particles as graupel, and the aggregate class also contains rimed aggregates. Therefore a direct comparison of the class aggregates between the two classification methods is difficult. The size distribution is much broader than from 03:00 to 04:00 UTC, with particles reaching 13 mm in their maximum dimension. The median amounts to 2.8 mm as there is still a significant proportion of small particles. The total number of particles is 589, while it was 55 in the previous period, showing that this period features more intense precipitation, which also makes the empirical size distribution more robust. The 75th percentile is 4.7 mm compared to 3.0 mm for the period 03:00 to 04:00 UTC, showing that the particles from 06:00 to 08:00 UTC are significantly larger. These large rimed aggregates can be attributed to the strong updraughts, which enhance the growth by aggregation and riming. Note that while the previous period (Sect. 5.1) featured riming below 2000 m, the precipitation rate was much smaller than from 06:00 to 08:00 UTC (Fig. 5e), and hence this riming did not contribute significantly to the total precipitation accumulation. The important message here is that the flow conditions in the WCB promoted rapid precipitation growth by aggregation and riming above 2000 m and are thus responsible for the large precipitation accumulation between 06:00 and 08:00 UTC. Moreover, most of the particles have higher quartiles of riming index (Fig. 9b) than between 03:00 and 04:00 UTC despite the lower proportion of graupel, which is due to the enhanced aggregation favouring rimed aggregates at the expense of pure graupel. We conclude that this period features the most riming both in absolute mass and in relative terms over all hydrometeor classes.

Figure 13Range spectrogram from WProf averaged (a) from 07:42 to 07:47 UTC and (b) from 07:57 to 08:02 UTC. Positive velocity values represent upward motions. The dashed white line shows the lower limit of the WCB (black contour in Fig. 5a, b, c).

Download

Figure 13 shows a range spectrogram from WProf averaged from 07:42 to 07:47 UTC and from 07:57 to 08:02 UTC. The updraught present in Fig. 5b can be seen as a strong shift in the mode of the spectrum from about −0.5 m s⁻¹ at 6000 m to above 2 m s⁻¹ at 5000 m (Fig. 13a). This updraught goes along with strong turbulence, which is visible as an increase in spectral width between 4000 and 6000 m. By 07:57 UTC (Fig. 13b), the increase in spectral reflectivity together with a decrease in Doppler velocity from 4800 to 4000 m suggests that large aggregates likely formed in the turbulent layer and further aggregate during their fall. This is consistent with the onset of aggregation below 5000 m observed in Fig. 10. The enhanced aggregation in the updraughts present from around 07:30 to 08:00 UTC leads to an increase in precipitation rate from 07:55 UTC to the maximum at 09:40 UTC (Fig. 5e). Particles that started falling between 4000 and 5000 m will take 85–105 min to fall to the ground, with an average effective fall speed (absolute fall speed plus updraught) of 0.8 m s⁻¹. This is consistent with the increase in precipitation intensity from 07:55 to 09:40 UTC, and the maximum could correspond to aggregates that formed in the updraught between 07:35 and 07:55 UTC. This hypothesis assumes a certain horizontal homogeneity, supported by the increase in precipitation in both rain gauge measurements at MHS and GWU that are separated by 19 km (Fig. 5e). It would imply that the embedded updraughts are responsible for the period of strongest precipitation, which also features intense riming (Fig. 5c), consistent with the suggestion in Oertel et al. (2019, 2020).

Figure 14Same as Fig. 8 for the period 10:00 to 11:00 UTC. The number of particles is 69. The average temperature is 0.0 ^∘C (Fig. 5e).

Download

Figure 15Same as Fig. 7 for 09:00 to 11:00 UTC on 28 February 2018. The statistics are based on 20 RHIs towards the MHS site. The dashed black line shows the lower limit of the WCB (black contour in Fig. 5a, b, c).

Download

Figure 16Conceptual model in a Lagrangian reference frame (i.e. along the WCB) summarising the key findings. The temperature indications come from the radiosounding at 06:00 UTC (Fig. 6a). The time is indicated as hours from the start of the ascent. The ascent is representative of a strong wintertime WCB.

Download

5.3 Shear-induced turbulence: 09:00 to 11:00 UTC

The period from 09:00 to 11:00 UTC features turbulence and intense precipitation rates (Fig. 5b, e). Due to a malfunction of the MASC, only pictures between 10:08 and 10:50 UTC were collected, leading to only 69 particles during this period (Fig. 14). There are substantially more crystals (10 %) than during the other periods. Figure 14a shows a few small aggregates, columnar and planar crystals, and one graupel particle. The median of the size distribution is 2.7 mm, and the 75th and 95th percentiles are 3.9 mm and 6.6 mm, respectively, indicating that the particles are smaller than in the previous period. Figure 15 shows a less pronounced increase in Z_DR compared to the period between 06:00 and 08:00 UTC (Fig. 10), suggesting that the depositional growth rate is smaller than during the previous period. The median of Z_DR increases to 0.7 dB around 4200 m, where crystals and aggregates probably dominate. The median of Z_H reaches a maximum of 22 dBZ at 2000 m compared to 25 dBZ at 06:00–08:00 UTC (Fig. 10). This is consistent with the size distribution (06:00–08:00 UTC in Fig. 12) that shows more large particles than Fig. 14. Below 4200 m, larger aggregates start to form as temperatures exceed −15 ^∘C and Z_DR decreases slightly. It is collocated with the increase in spectral width (Fig. 15), reflecting the vertical wind shear below the maximum wind speed (Fig. 6c) at 09:00 UTC. The shear layer was between 4000 and 5000 m in the period from 06:00 to 08:00 UTC, while it is now between 3000 and 4000 m. This vertical wind shear is collocated with the turbulent cells observed in Fig. 5b from 08:00 to 10:00 UTC around 4000 m, which suggests that they originate from Kelvin–Helmholtz instabilities. This is supported by values of the gradient Richardson number (not shown here) of 0.2 where the turbulent cells are present. The decrease in the height of the wind shear is consistent with the decrease in the height of maximum wind speed between 06:00 and 09:00 UTC (Fig. 6c) and explains why the altitude of aggregation enhancement by turbulence decreases with time. This was also observed by Keppas et al. (2018), who attributed this altitude decrease in the maximum wind speed to the passage of the warm front. On the other hand, the altitude of the onset of aggregation could increase with time as the warm front passes because the altitude of the −15 ^∘C isotherm (relative maximum aggregation in Hobbs et al., 1974) is higher at 09:00 UTC than at 06:00 UTC. Our interpretation is that the enhancement of aggregation by turbulence dominates the polarimetric signatures in our case: first, because crystals were likely growing as sectored plates and not as dendrites, the latter being more effective to aggregate at −15 ^∘C, and second, because the intense aggregation taking place in the shear layer leads to larger aggregates than the early aggregation at −15 ^∘C, which makes the former more visible in the polarimetric profiles. The distribution of the riming index (Fig. 9c) is broader than for the other periods due to the higher proportions of crystal-like particles and the presence of graupel and rimed aggregates.