Characteristics of Radar Echo Parameters and Microphysical Structure Simulation of a Short-Time Heavy Precipitation Supercell

As one of the most violent forms of local severe convective storms, supercell storms have been attracting the attention of many meteorologists (Yu et al. ^[1]; Davies-Jones ^[2]). With the development of radar detection technology, some special features in the interior structure of supercells have been further confirmed (Wade et al. ^[3]; Peters et al. ^[4]; Coffer et al. ^[5]). A typical supercell storm usually shows a relatively fixed appearance, while supercells of severe precipitation vary in shapes and are sometimes featured with multi-cell storms or typical supercells (Martín et al. ^[6]; Mulholland et al. ^[7]). China is vast in territory, having complicated and diverse topography, so the environmental conditions for supercell storms to form and the characteristics of radar observation are quite different (Dai et al. ^[8]; Cai et al. ^[9]; Zhang et al. ^[10]; Yu et al. ^[11]). Analyzing radar characteristics and environmental conditions is helpful for understanding the internal structures of supercells and the causes for their formation and development (Zhang et al. ^[12]; Diao et al. ^[13]). It has been found that some scattered and small-scale convective cells tend to develop southward along the surface mesoscale convergence line and coalesce with each other into supercells (Wu et al. ^[14]). Therefore, the coalescence of thunderstorm cells is an important factor to promote the growth, enhancement and persistence of thunderstorm systems (Yi et al. ^[15]). However, the coalescence is a complicated non-linear process, which is understood mostly based on qualitative observation of radar images, rarely involving the analyses on the quantitative changes of thunderstorm systems and the physical process in the progress of coalescence. At present, there exist large deviations in the evolution of thunderstorm systems between the observation and that extrapolated by a quantitative precipitation estimation algorithm of dual-polarization radar (Noh et al. ^[16]). Constructing echo parameters with clear physical meanings based on observations and quantitatively analyzing their evolution features of thunderstorm system can indirectly help comprehend the in-cloud convection and particle changes and other physical processes (Yi et al. ^[17]; Yi et al. ^[18]; Xu et al. ^[19]). However, to reveal the microphysical mechanism of the evolution of supercells needs further studying combined with numerical simulation methods (Wang et al. ^[20]; Pan et al. ^[21]). Simulation studies on microphysical structures of the supercells of heavy precipitation showed that significant discrepancy exists between the characteristics of thunderstorm clouds in southern and northern China (Gan et al. ^[22]; Lou et al. ^[23]; Feng et al. ^[24]). During the supercell-induced heavy rainfall in southern China, there is usually abundant liquid water content, which is about twice the maximum cloud-water content of thunderstorm clouds in northern China (Liu et al. ^[25]; Xu et al. ^[26]). Another research denoted that, in terms of the structure and development, convective storms are very sensitive to changes in low-level wind fields (Wu et al. ^[14]). The strong updraft in clouds can make the coexistence zone of ice and graupel particles become very large, and make the falling water droplets reach the surface quickly, which is the important mechanism for convective storms to produce heavy precipitation (Wang et al. ^[27]). Using the deep learning method, which has emerged in recent years to integrate numerical prediction products and multi-source observation data, can improve the forecast lead time of convective storms effectively (Yu and Zheng ^[28]). In addition, a short-time heavy precipitation forecast model based on the XGBoost ensemble learning method has good performance in forecasting short-time heavy precipitation (Han et al. ^[29]). Liu et al. ^[30], Li et al. ^[31] and Gao et al. ^[32] carried out some tests on the forecasts of short-time heavy precipitation caused by convective storms by using the Kalman filter ensemble method, achieving ideal results. All these studies are of great significance for understanding the structure and formation mechanism of heavy rainfall supercells, and for the forecasting and warning of severe convective weather. However, the research on quantitative prediction of supercell-induced heavy precipitation is still insufficient.

Impacted by a squall line in transit, Pingshun County in Changzhi City, Shanxi Province experienced a record-breaking short-time extreme precipitation process from 19:00 to 20:00 (Beijing Time, the same below) 24 June 2020, but due to insufficient estimation, this extreme precipitation event was missed in forecast. Thus, in this study, we will investigate the evolution characteristics of the supercell that caused the extreme precipitation in Pingshun County in the squall line process based on radar data, and also analyze quantitatively the evolution of parameters and the relationship between the physical process change of these parameters and the supercell development by means of constructing radar echo parameters. We also plan to build a comprehensive prediction model of echo parameters relying on the random forest principle, and determine a quantitative prediction index and warning lead time of short-time heavy precipitation by the methods of hindcast and comparison. In addition, combining with the NCEP/NCAR FNL 1°×1° reanalysis data and numerical simulation data, we will analyze the microphysical structure of severe precipitation supercell and verify the results of radar echo parameters, and finally summarize the prediction models of short-time heavy precipitation so as to provide some technical supports for future's forecasting and early warning of such extreme weather events.

On 24 June 2020, a severe convective weather process occurred in Shanxi Province from northwest to southeast lasting for nearly 10 h (13: 00-22: 00). Thunderstorms and gales swept 11 cities of the province with 11 counties being hit by hails. There were four counties in the southern and central parts of the province caught by short-time intense rainfall from 17:00 to 20:00 (Fig. 1a), during which the hourly precipitation in Pingshun County reached 92.1 mm from 19:00 to 20:00 (Fig. 1b), breaking the historical extreme value (74.9 mm h^-1 in the 8 August 2019 short-time severe precipitation in Zuoquan County) of Shanxi Province, and the most intense precipitation in Pingshun concentrated in the 19:10-19:40 spell (Fig. 1b).

Figure 1.  (a) Severe convective weather observations (Big green dots represent short-time severe precipitation, triangles represent hail, and the small gray, blue and red dots are for the northwesterly wind at the Beaufort scales of 8, 9, 10 and above, respectively), and (b) evolution of 10 min precipitation (units: mm) from 19:00 to 20:00 in Pingshun County on 24 June 2020.

To further understand the characteristics of the severe convective system, four observing stations (Fangshan, Fenyang, Lucheng and Pingshun) on the moving path of the system are selected to analyze the changes of meteorological elements with time from 10: 00 on the 24th to 02:00 on the 25th (figure omitted). The result shows that the changes of meteorological elements observed at the four stations all had abrupt characteristics within an hour, such as sharp increases of more than 40% in relative humidity, more than 3.5 hPa in pressure and more than 8 m s^-1 in wind speed, a sudden change of wind from easterly to northerly, and a steep drop of temperature by more than 10℃ in an hour. These features conform to the typical characteristics of a squall line, which has a large impact range, slow moving speed, long duration and strong intensity. Therefore, the Pingshun event was a squall line process rarely seen in early summer in Shanxi Province.

Figure 2a shows a wide westerly jet in the area of 30°N-40° N at 200 hPa at 08:00 on the 24th, with the jet axis (≥40 m s^-1) passing through the central part of Shanxi and a high-level trough located near 98° E. On that very day, Shanxi Province lay in the divergence area of an upper-level jet outlet, having strong suction effect in the upper-level air. Corresponding to the 500 hPa situation (Fig. 2b), the atmospheric circulation was in a "two troughs versus one ridge"pattern in the high latitudes. A shortwave trough was located over the Hetao area, and, affected by a northwesterly wind at the bottom of a cold vortex, cold advection was very obvious in Shanxi. Warm ridges at the low levels of 700 hPa and 850 hPa (figure omitted) were superimposed on a 500 hPa temperature trough, resulting in potential instability in the atmospheric stratification. However, the lower-level wind field was weak and the air was dry (T - T_d ≥ 12℃). On the surface (figure omitted), a Mongolian cyclone was stable and less active from 08:00 to 20:00, with the central strength maintained between 995 and 997.5 hPa, and the accompanying front barely moved. However, the near-surface humidity in southeastern Shanxi continued to increase in the afternoon.

Figure 2.  The circulations at (a) 200 hPa (Black solid lines represent geopotential height, units: dagpm, shadow is jet stream, units: m s-¹, and wind vector, units: m s-¹) and (b) 500 hPa (Black solid lines represent geopotential height, units: dagpm, and dotted line is isotherm, units: ℃) at 08:00 on 24 June 2020.

A sounding analysis reveals that at 08: 00 on the 24th (figure omitted), the wind below 850 hPa over Taiyuan blew clockwise with altitude and there was warm advection, while the wind between 850 hPa and 500 hPa flew anticlockwise with altitude, and cold advection existed. The stratification curve had an "upward trumpet mouth"structure; the air above the 1.5 km height was dry; the convective available potential energy (CAPE) was 111.6 J kg^-1; the downdraft convective available potential energy (DCAPE) was 150.3 J kg^-1; and the wind vertical shear of the 0-6 km height was at a moderate strength. With the vertical structure being conducive to the strengthening of convective instability, and poor conditions of water vapor, energy and wind shear conditions, it is thus not beneficial to the development and systematization of convective cells.

To sum up, this squall line process in Shanxi occurred under the synoptic context that prior to the surface cold front, low-level cold air was in filtration and near-surface humidity continuously increased in the afternoon. There was potential instability in the atmospheric stratification, but the surface front did not move southeastward, the low-level shear line was not so distinct, there was no jet stream, and the wind vertical shear at the 0-6 km altitude was not strong enough. With such a synoptic situation, it is difficult to explain how the convective cells got systematized and developed into squall lines, and how they formed supercells and triggered the short-time extremely heavy precipitation in Pingshun County from the perspective of synoptics. Therefore, we will explore the related formation mechanism in terms of microphysical processes and the evolution of their structure.

The short-time extremely heavy precipitation in Pingshun County occurred in the developing and systematizing courses of the squall line, and was caused by a supercell that developed from the thunderstorm cells on the front side of the squall line. Seen from Fig. 3, the supercell developed into the mature stage at 18:22 when the strongest echo was 55 dBZ at the elevations of 0.5° (Fig. 3a) and 1.5° (Fig. 3b), 62 dBZ at 2.4° (Fig. 3c) and 3.4° (Fig. 3d), and 67 dBZ at 6° (Fig. 3e) and 9.89° (Fig. 3f), respectively. In the meantime, the echo area of 65 dBZ enlarged to the peak value at the 9.89° elevation. Therefore, the reflectivity factor (R) was enhanced with rising height, and the strong echo core shifted to the right with the increase of height. There were an obvious V-shaped notch at the elevations below 4.3°, and the front inflow is strong. On the corresponding vertical cross-section (Fig. 3g and 3h), a strong echo of ≥55 dBZ reached the ground surface, the echo top (ET) was at the height of 9-10 km, the strongest echo was located at the height of 4-5 km with an overhanging structure and there was an obvious bounded weak echo zone. From 18: 28 to 18:45, with the stable development of the cell, the height of the strong echo zone extended upward, making the volume of the bounded weak echo region enlarged and the echo overhang significantly enhanced. The forward inflow maintained for 24 min at the low level, which was in line with the persistent strengthening of the southeasterly wind on the surface.

Figure 3.  Reflectivity factors (units: dBZ) at the elevations of (a) 1.5°, (b) 2.4°, (c) 3.4°, (d) 4.3°, (e) 6°, and (f) 9.89° and their cross-sections (g) 1.5° and (h) 6° at 18:22 on 24 June 2020 (White arrow points to the V-notch, black arrow points to the forward inflow, and the cross-section is made along the solid black line).

On the radial velocity chart (Fig. 4), cyclonic velocity pairs appeared at elevation angles below 4.3° at 18: 34 (Fig. 4a-4d), the rotational velocity was between 16.5 m s^-1 and 24.5 m s^-1, and the diameter of the core area was 4-6 km. The mesocyclone extended to the 10 km height lasting for 12 min, and always in the area of strong reflectivity. On the corresponding cross-section (figure omitted), there was obvious convergence in the middle layer. The mesocyclone had a moderate strength (Yu et al. ^[1]). In addition, the vertical integrated liquid water content (VIL) reached its peak value at 18: 22 (figure omitted), and the maximum VIL near the center of the cell became greater than 70 kg m^-2.

Figure 4.  The radial velocity (units: m s^-1) at the elevations (a) 1.5°, (b) 2.4°, (c) 3.4° and (d) 4.3° at 18:34 on 24 June 2020 (Black circle represents the speed pair).

This cell lasted for more than 1 h and developed rapidly at first. During its developing stage, there were characteristics like the V-shaped notch, overhanging and weak echo area, etc., and the forward inflow was strong, a moderate-strength mesocyclone appeared on the velocity chart, and the radial convergence in the middle layer was clear. All these indicate that this cell was a typical supercell. In the mature stage, the VIL increased sharply by more than 40 kg m^-2 in 6 min. About 20 min after the supercell got matured, the reflectivity R at different elevations suddenly increased by more than 36.6 dBZ, and after about 40 min, heavy precipitation occurred at the maximum gradient point of R. With the onset of heavy precipitation, the values of VIL and R declined rapidly. This result has an indicative meaning for quantitatively nowcasting heavy precipitation about 30 min in advance.

Constructing echo parameters with clear physical meanings is an effective means to indirectly understand physical processes such as in-cloud convection, precipitation, and electrification. Based on the construction and calculation of radar echo parameters, Yi et al. ^[18] found some special phenomena of the echo parameters in the process of cells'coalesce. In the present study, we use this method to quantitatively analyze the characteristics of the radar echo parameters of the supercell that caused the short-time extremely heavy precipitation in Pingshun County on 24 June 2020. The used radar data were collected by 14 C-band radars across Shanxi Province. The rectangular coordinate system is used to calculate the parameters from the radar 3D mosaic data. The data has a horizontal resolution of 0.01° × 0.01° and is divided into 21 layers vertically, of which the 0.5-5.5 km height range is divided into 11 layers with a 0.5 km interval, the 6-10 km range into 5 layers with a 1 km interval, and the 11-20 km range into 5 layers with a 2 km interval, respectively.

The constructed parameters and their meanings are as follows: (1) V₄₀ and V₅₀ are the volumes enclosed by the areas with echo strengths ≥40 dBZ and ≥50 dBZ in each of the 21 layers, respectively. These parameters reflect the size of the echo core at a certain time and indirectly reflect the development degree of the cell and the number of hydrometeor particles in thunderstorm clouds (Yi et al. ^[18]). S_6-40 and S_6-50 are the areas enclosed by strong echoes of 40 dBZ and 50 dBZ at the 6 km height, respectively, and they can also reflect the strength of developed cells. (2) V_40up-6 and V_40down-6 are the volumes enclosed by strong echoes ≥40 dBZ in the 10 layers above 6 km and the 11 layers below 6 km, respectively. Because ice particles (graupel, ice crystal and hail) are the major components in clouds at or above the 6 km height, they can reflect the upward motion strength of the particles in clouds and the concentration of ice particles to a certain extent. (3) FV₄₀ and FV₅₀ express the variable quantity before and after the 6 min interval in V_40up-6 or V_40down-6 and in V_50up-6 or V_50down-6, respectively, i.e., variability.

where t represents a certain moment, (t + 6 min) is the moment at which t plus 6 min. A positive FV₄₀ value indicates that the cell is developing while a negative FV₄₀ value means that the cell has stopped developing or that despite developing it has a trend of weakening. The calculation of the above elements is obtained on the basis of Yi, et al. ^[18] (units: grid).

It can be seen from Fig. 5a that at 17:48 the storm cells began to form. After about 48-54 min, the echo parameters V_40up-6 and V_40down-6 reached their peak values, and the latter was 6 min later than the former, which is different from the research conclusion of Yi et al. ^[18]. During this period, the two parameters were both continuously increasing (arrow 1). Heavy rainfall began to occur about 30 min after the two parameters reached their peaks. In the process of heavy precipitation, however, the parameters showed a decreasing trend (arrow 2). What is worth noting is that the growing trend of the cells slowed down 6 min before their maturing.

Figure 5.  Temporal evolution of radar echo parameters of convective cells from 17:48 to 19:48 on 24 June 2020. (a) V₄₀ (Solid line is for V_40up-6 and dashed line for V_40down-6; numbers correspond to the left and right coordinate axes, respectively; horizontal arrow indicates the time for the parameter to increase to its maximum value; and the arrows 1 and 2 indicate the continuous increase and decrease of parameters, respectively); (b) FV₄₀ (Solid line is for FV_40up-6 and dashed line for FV_40down-6; numbers correspond to the left and right coordinate axes respectively; and the arrows 3 and 4 indicate the"sudden drop"before the parameter increases to its maximum value).

The curves in Fig. 5b indicate that, at 18:12 and 18:36, both FV_40up-6 and FV_40down-6 suddenly decreased, that is, they had so-called"sudden drop, "which means that when V₄₀ slowed down, FV₄₀ experienced a"sudden drop"(arrows 3 and 4), in other words, the FV₄₀ value reached its peak 6 min earlier at the height above 6 km than it did below 6 km. Therefore, we can see that the developing and maturing of storm cells began with the enhancement of the upward motion of particles in clouds and the concentration increase of ice particles, while the increase in hydrometeor particles took a lag of 6 min. At this "sudden drop" moment, no precipitation was observed on the surface, indicating that the number of hydrometeor particles was not reduced. Thus, this "sudden drop"was not caused by precipitation, but by the cells'coalescence and rapid development. Then, this "sudden drop"can be taken as a precursor signal to judge the coalescence and development of cells. However, after the beginning of precipitation, V₄₀ declined continuously as a result of the obvious decrease in the number of in-cloud hydrometeor particles, which was caused by the precipitation release. In addition, the weakening of the strong echo core indirectly reflected the weakening of the updraft, which indicates that the overall upward motion of the thunderstorm system was weakened when the cells approached maturity. After the precipitation started, FV₄₀ above 6 km changed inversely with that below 6 km, indicating that the volume of the strong echo core was still enlarging above 6 km while getting smaller below 6 km, which is the typical performance of overhanging structure. Moreover, they were all negative, suggesting that the strong echo core was weakening and the precipitation was no longer developing.

The strong echo core areas of 40 dBZ and 50 dBZ experienced abrupt enlargement simultaneously at the 6 km height 1 h before the cell coalesced and developed (arrow 1 in Fig. 6a). The area of the 50 dBZ echo core increased even more, by 4 times within 12 min reaching 319 grids, which was 3.1 times the area of strong echo core of short-time heavy precipitation cells (≥30 mm h^-1). The area of the 40 dBZ echo core developed to its maximum after about 45 min, while that of the 50 dBZ echo reached the largest size after 30 min, which indicates that the echo growth started from the core and expanded outward. During the period of cell coalescence and development (box in Fig. 6), the echo core areas with 40 dBZ (lasting for 30 min) and 50 dBZ (lasting for 48 min) both covered more than 200 grids. This infers that the in-cloud hydrometeor particles aggregated unceasingly and the updraft strengthened continuously. After the heavy precipitation started, the strong echo core areas with 40 dBZ and 50 dBZ saw a"sudden drop"at the same time (arrows 2 and 3 in Fig. 6a), and the shrink in the area of the 50 dBZ echo core was more sharply, cut by half within 6 min (arrow 2 in Fig. 6a). The greater the decrease in the core area, the stronger the precipitation intensity. Therefore, changes in the echo core area at the height of 6 km could heavily influence the coalescence and development of convective cells, rapid growth of hydrometeor particles in clouds, and precipitation intensity.

Figure 6.  Evolution of (a) the strong echo area of supercell at the 6 km height (solid line is 40 dBZ, dashed line is 50 dBZ) and (b) the cross-section area of the convective cell at the 3-11 km height from 17: 48 to 19: 48 on 24 June 2020 (units: grid).

Profiles of convective cells are made along different heights (3, 6, 9 and 11 km), and their cross-sectional areas (S) are calculated (Fig. 6b). These profiles can reflect the changes in the intensity of convection and also indirectly reflect the changes in the numbers of hydrometeor particles and ice particles in clouds. By comparatively analyzing their time series we have found that, as the height increased, the cross-sectional area decreased and thus the cell developed to be spire-shaped. Before the cells coalesced and developed, the cross-sectional areas showed a continuously increasing trend, but before their developing and maturing, there appeared obvious "sudden increase"in the cross-sectional areas, which start from the upper layer and then expand downwards. This indicates that the convection kept strengthening and the numbers of hydrometeor particles and ice particles in clouds were continuously increasing. Moreover, the increase of high-level ice particles preceded the increase of low-level hydrometeor particles. After the cells grew into the mature stage, the S value for the 3-11 km altitude was between 359 and 1136 grids, decreasing with altitude. The S value for the 3-6 km height in the lower level was 3.5 times that of the short-time severe rainfall of 20-30 mm h^-1 in Shanxi Province. In 6-12 min prior to the start of heavy precipitation, the S in the upper layer began to decrease while the S in the lower layer was still increasing, which indicates that ice particles had begun to fall and decreased in number. As the result of collisions, the number of water particles in the lower layer increased, which was a signal of the onset of heavy precipitation. When it approached the heavy precipitation, the S in the upper layer decreased still earlier than in the lower layer. Its decrease rate was approximately 20% at the 9-11 km height, and less than 6% at the 3 km height, which means that the ice particles decreased rapidly while the hydrometeor particles decreased slowly. This is an important factor in the maintenance of heavy precipitation.

A total of 132 samples of short-time heavy precipitation in recent 10 years are selected, and the short-time heavy precipitation (expressed by SR) is divided into four grades in order: [20, 50) mm h^-1, [50, 80) mm h^-1, [80, 100) mm h^-1 and ≥100 mm h^-1. The lead time (expressed by T) is also divided into four levels: 20-30 min, 30-40 min, 40-50 min and ≥50 min. First, the above 11 echo parameters and their lead time are normalized and the natural breakpoint method is used to classify the corresponding grades. Then, a comprehensive parameter (CP) prediction model for the echo intensity and lead time is constructed according to the random forest principle (Table 1). The echo intensity is expressed by ECP, and the lead time by TCP.

A hindcast validation is performed on the above technical thresholds, and the obtained historical fitting rate reaches 95.7%. According to the model prediction result, the TCP of this process is 40-50 min and the SR is [80 mm, 100 mm), which means that the CP can be used to predict the future short-time heavy precipitation value of 80 mm h^-1≤SR < 100 mm h^-1 40-50 min in advance.

The analysis of the densely observed surface data shows that a mesoscale dew point front always existed during the formation, development and coalescence of the convective cells and was accompanied by strengthening southerly or easterly winds on the surface. The mesoscale convergence line formed by strengthening northerly wind and its meeting with southerly or easterly winds was an important factor for the convective cells to form and develop and for triggering heavy precipitation. However, the continuously strengthening northerly wind and the rapidly weakening southerly or easterly wind led to the intensity reduction of the cells

Now we focus on the change of 1-h maximum wind in the supercell-affected area. Before 11: 00, the wind direction was disorderly with wind speed ≤2 m s ^-1. From 11:00 to 14:00 (Fig. 7a), the westerly wind was blowing mainly with the wind speed increasing to about 4 m s^-1. From 15:00 to 16:00 (Fig. 7b), the southwesterly wind kept strengthening. From 16: 00 to 17: 00 (Fig. 7c), the southwesterly wind joined the northeasterly wind in the region from Lucheng County to Pingshun County causing the maximum wind speed to exceed 12 m s^-1, and then weak cell activity began to appear in this region. From 17:00 to 18:00 (Fig. 7d), a convergence of the southwesterly, northerly, and easterly winds happened from Jincheng to Lucheng and Pingshun County. As a result, wind speed strengthened to 8-14 m s^-1, and convective cells were generated along the convergence line, developing rapidly. From 18:00 to 19: 00 (Fig. 7e), the southwesterly and northeasterly winds blowing on both sides of the convergence line continued to increase, and the wind speeds reached 14 m s^-1 and 16 m s^-1, respectively. Meanwhile, both the near-surface moisture and cold air peaked. These were important factors for supercells to mature and trigger the short-time heavy rainfall. From 19:00 to 20:00 (Fig. 7f), the southerly wind decreased rapidly to 1 m s^-1 while the northerly wind maintained its speed at 12-14 m s^-1. At this time, there was cold air spreading across the lower troposphere, the water vapor conditions disappeared, the cells weakened, and the heavy rainfall tended to end.

Figure 7.  Evolution of 1 h maximum wind (units: m s-¹) observed densely by automatic weather stations at (a) 14:00-15:00, (b) 15: 00-16:00, (c) 16:00-17:00, (d) 17:00-18:00, (e) 18:00-19:00, (f) 19:00-20:00 on 24 June 2020 (Black solid line represents frontal surface, arrow represents wind direction, double dotted line represents convergence line, and double dotted line represents mesoscale vortex).

In addition, in the process of the squall line moving eastward, the mesoscale low pressure appeared in front of the squall line, and a wake mesoscale high formed after the squall line, forming a strong pressure difference between them (figure omitted), thus increasing the pressure gradient before and after the squall line and leading to an increase in wind speed. The interaction between the mesoscale convergence line and the severe convection made the echo structure more obvious.

In summary, the mesoscale system on the surface plays an important role in the formation, development and coalescence of the convective cells. The formation of the mesoscale convergence line and the strengthening of the wind elements on both sides of the convergence line are meaningful in predicting the formation and development of convective cells 2 h in advance, thus having a reference value for the nowcasting of heavy precipitation.

Radar echo parameters can only roughly estimate the changes in ice particles and hydrometeor particles. To further understand the distribution and evolution of different types of particles in clouds and their roles in precipitation processes, and reveal the microphysical mechanism of supercells that produce short-time heavy precipitation, it is necessary to adopt numerical simulation.

In this study, we use WRFV3.7 with a three-layer nesting scheme for simulation. The resolution of the innermost layer is 3 km. The improved NSSL two-parameter scheme is used in the microphysical parameterization scheme. The outer two layers of the cumulus scheme have the Tiedtke scheme, but the inner layer does not. The longwave radiation is calculated by the RRTM scheme. The shortwave radiation is calculated by the Dudhia scheme. The Yonsei University scheme is used in the boundary layer. The land surface is resolved by a five-layer thermal diffusion scheme. The NCEP/NCAR FNL 1° × 1° reanalysis data with 6-h intervals are used as the initial field and boundary condition. The start time of the simulation integration is set at 08:00 on 24 June and the integration time is 18 h with 90 s time steps in all the three layers. According to the large-scale synoptic situation simulated by the model (figure omitted), the 500 hPa circulation in the upper air and the 700 hPa and 850 hPa wind fields in the lower air at 08: 00 on 24 June are basically consistent with the observations. The focus of our analysis is on the simulated composite reflectivity, whose shape, developing trend and strong echo distribution are close to the actual situation. From the moment when the supercell matures and produces the short-time heavy rainfall (Fig. 8b), the area of the echo ≥35 dBZ is slightly smaller than the observed by 12.1 km² (Fig. 8a). The simulated supercell in front of the squall line is located at the junction of Licheng County and Pingshun County, around 4 km east of the actual location, and the maximum intensity of the supercell is ≥55 dBZ, slightly smaller than that of the actual situation by about 3 dBZ. In general, the model simulation of this process is successful and the simulation results are reliable. The simulated high-resolution data are reliable and can be used in analysis.

Figure 8.  (a) The observed and (b) simulated composite reflectivity at 19:00 on 24 June 2020 (units: dBZ) (Black dotted line is the squall line in the process of development and systematization, and the blue arrows point to the supercell).

A radial vertical cross-section is made along 113° E at which the center of the matured supercell is. The evolution characteristics of water vapor in the ambient atmosphere, mixing ratio of particles in the cells and vertical wind speed during their lifetime are analyzed.

During the period of the cells'formation and development, the updraft (red isoline in Fig. 9) was primarily between 850 hPa and 200 hPa (at the 1.5-12 km height), and its maximum upward speed was only 6 m s^-1. There appeared large-value areas respectively at the altitudes of 3 km and 6 km, and the value at the higher altitude (6 km) was 2 m s^-1 larger than the value at the lower height (3 km). Compared with the squall line process in southern China, the updraft in the supercell had lower extension height and lower vertical velocity (Gan et al. ^[22]). The water vapor (Fig. 9a) in the ambient atmosphere was concentrated below 500 hPa, and the large-value area of its vertical gradient was near 700 hPa (3 km in height). The near-surface specific humidity exceeded 14 g kg^-1, approaching the specific humidity threshold in the local area for heavy torrential rain. In the cell area, however, corresponding to the strong vertical ascending motion, the water vapor rapidly extended upward in the cold and warm cloud areas and developed even higher, up to 400 hPa, in the cold cloud, getting closer to the wet layer thickness of the local heavy rain. This was an important signal that the cell obtains moisture from the ambient atmosphere and develops rapidly. The cloud water (Fig. 9b) was distributed over a large vertical range, primarily between the heights of 2 km and 12 km. The warm cloud base was located at the height of about 2 km, having a large-value area near the 0 ℃ layer. The top of the cold cloud extended upward to the 12 km height, and its large-value area was upright, stretching from the -3 ℃ layer to the - 30 ℃ layer, narrow and deep, much stronger than the supercell over the city of Guangzhou (Gan et al. ^[22]). There was vertical ascending motion in both cold and warm clouds, but the ascending motion in the cold cloud was strong and deep. The center of the maximum ascending motion (6 m s^-1) was located near the -10 ℃ layer at 500 hPa, which was favorable for the rapid increase of ice crystals.

Figure 9.  The cross-sections of simulated distribution of (a, c) water vapor and (b, d) cloud water at (a, b) at 18:00 and (c, d) 19:00 on 24 June 2020 along 113°E (units: g kg^-1) (Black line is isotherm, red line is vertical velocity; black ellipse dotted line is the largevalue area of vertical gradient of water vapor, and black dotted arrow indicates the upward extension of water vapor in the cell).

In general, the ambient atmosphere and clouds has abundant water vapor, which can be quickly acquired by the cells so that they can grow rapidly with the nutrient support of their mother cell. Compared with previous studies (Liu et al. ^[25]; Xu et al. ^[26]), the cloud water content of thunderstorm clouds in northern China was much larger than and closer to that in southern China (Gan et al. ^[22]), which further indicates that liquid water was very rich in this convective precipitation.

Around the beginning of the heavy precipitation, the updraft (red isoline in Fig. 9) still maintained two large-value centers, and their heights descended slightly. The intensities of the two centers also rapidly decreased to less than 1 m s^-1, and the center at high levels had more rapid decline in intensity. Meanwhile, a downdraft occurred vertically below the 0 ℃ layer at the 4 km height. The water vapor in the ambient atmosphere (Fig. 9c) was still distributed below 500 hPa, and the large-value area of vertical gradient area had dropped to 750-800 hPa, but the gradient and horizontal scale had increased. The near-surface specific humidity exceeded 13 g kg^-1, which was still close to the specific humidity threshold of local heavy rainfall. Inside the cells, however, the water vapor in the cold and warm cloud areas no longer developed upward. The cloud water (Fig. 9d) still had two gathering areas, but they moved southward fast, accompanied by rapid decline in height and intensity, especially the intensity decreased sharply to a quarter of its original values. Thus, this was a very important precursory signal indicating an immediate occurrence of heavy rainfall.

The profiles of the mixing ratios of hydrometeor particles indicate that during the period of cell formation and development, the ice crystal particles (Fig. 10a) have a high distribution height mainly above - 20 ℃ with a wide horizontal scale, and their mixing ratio center (with the maximum value about 0.27 g kg^-1) lies between the -40 ℃ and -30 ℃ layers, corresponding to the upper part of the ascending motion. The mixing ratio of ice crystals is slightly bigger than that of hail particles. Since snow particles (Fig. 10b) are ice crystal aggregates, their distribution characteristics are similar to those of ice crystals, but their heights in the air are lower than that of ice crystals, mainly distributed above the -15 ℃ layer. Their mixing ratio (with the maximum value about 0.9 g kg-¹) is bigger than that of ice crystal particles, with the center located between the -35 ℃ and -25 ℃ layers, but in the front of the strong upward motion. The snow and ice crystals produce graupel via the riming process, and accordingly, the graupel mixing ratio center is vertically beneath those of snow and ice crystals. A large number of graupel particles (Fig. 10c) are distributed below 270 hPa and above 700 hPa (the - 40 ℃ to 5 ℃ layers), and the mixing ratio center (maximum > 2 g kg^-1) is between the -20 ℃ and -10 ℃ layers. The graupel mixing ratio is the largest among the mixing ratios of all hydrometeor particles. Compared with the supercell in southern China, the vertical distribution of graupel particles is more upright, with the extension height also more than 1 km higher than that of the former (Pan et al. ^[21]). The graupel grows through the collision and freezing process, and some are transformed into hail particles. Therefore, the hail particles are distributed at the center of the graupel mixing ratio. Between -30 ℃ and 0 ℃, there are a small number (maximum 0.13 g kg-¹) of hail particles (Fig. 10d), whose vertical structure is similar to the structure of graupel particles, but the structure bottom is lowered to about 1 km, which is close to the altitude of Pingshun County, corresponding to small hail on the surface. The distribution areas of graupel and hail both match with the strong ascending motion, and the vertical upward motion reached the strongest strength at the 8-10 km height, creating an even larger coexistence area for graupel, ice crystals and snowflakes. The raindrops (Fig. 10e) are distributed below the 0 ℃ layer, and their mixing center extends to the 4 km height, corresponding to the part below the upward motion in clouds.

Figure 10.  The simulated cross-sections (units: g kg^-1) of all hydrometeors along 113°E at 18:00 on 24 June 2020. (a) ice crystals; (b) snowflakes; (c) graupel; (d) hail; (e) raindrops (Black line is isotherm, red line is vertical velocity)

It can be seen that, during the developing period of convective cells, graupel has a large vertical distribution range and hail is the major consumption form of graupel. Graupel and snow share large areas of coexistence. Raindrops collect cloud water via the gravity collision-coalescence process and also collect the small droplets formed by melting graupel and hail below the 0 ℃ layer, so that the raindrops can reach the biggest sizes under the graupel and hail in vertical direction. The height of raindrops is close to the surface, with the largest horizontal scale and high precipitation efficiency. Therefore, graupel is the most important source of rain water in severe convective cloud precipitation process. Approaching the onset of heavy precipitation (figure omitted), the mixing ratios of all hydrometeor particles are reduced significantly and their heights decline rapidly.

Based on the above comprehensive analysis, prediction models are obtained for the development of heavy precipitation supercell (Fig. 11a) and the short-time heavy precipitation (Fig. 11b) in the squall line process. The steady southward movement of the mesoscale dew point front on the surface was an important factor for the squall line to be systematized for many times. The surface mesoscale convergence lines (or vortices) promoted the generation of new cells, and the interaction of these two mesoscale systems allowed the cells to develop, coalesce, strengthen and get systematized. In the afternoon, as the southerly and easterly winds in front of the squall strengthened persistently, the near-surface humidity continued to increase as well. In the evening, newborn cells were excited again near the surface mesoscale convergence line, so that the cells rapidly captured moisture from the ambient atmosphere and got fed by the mother cell. While the mesoscale dew point front moved southward, convergence of winds from three different directions (or mesoscale vortexes) occurred, resulting in a strong mesoscale upward motion and triggering the rapid upward development of the supercell. As a result, the cold cloud area continued to extend upward, the vertical upward motion in clouds was deep and strong, the ice particles continued to grow, and the liquid water was very rich. At the same time, the coexisting areas of snow and graupel particles were also rapidly enlarged. The graupel kept growing through the colliding and freezing process, and near-surface raindrops captured the melted particles falling from upper layers, keeping growing and landing on the ground efficiently, and finally resulting in the Pingshun County short-time extreme precipitation with intensity surpassing the historical extreme value in Shanxi Province.

Figure 11.  The prediction models for (a) super-cell formation and development and (b) induced short-time heavy precipitation on 24 June 2020.

The analysis of the synoptic-scale background can only show the possibility of severe convection, but the analysis on the surface mesoscale system can help forecast the generation and development of convective cells 2 h in advance. The conventional radar products can denote the probability of short-time heavy precipitation 23-30 min in advance. However, a fine analysis of the radar echo parameters and microphysical structure constructed can be used to judge the intensity of heavy precipitation about 1 h in advance.

The severe squall line process in Shanxi Province on 24 June 2020 led to the occurrence of thunderstorm gale, hail, and a localized record-breaking short-time extreme precipitation event. In this paper, the physical process and microphysical structure of supercell evolution have been analyzed by constructing radar echo parameters combined with numerical simulation.

(1) The surface meteorological elements of the squall line process of interest, rarely seen in early summer, changed violently with a large influence range, long duration, and strong intensity. It occurred under the background that it was ahead of a surface cold front with infiltration of low-level cold air, and there was continuous increase of near-surface humidity in the afternoon. Different from the weather condition of usual squall lines, the synoptic scale for this squall line was specially characterized by the eastward moving of the frontal surface, the absence of obvious shear line or jet at low levels, and a weak vertical wind shear in the lower troposphere. Therefore, the severe convection was difficult to be categorically forecast, and the forecast signal was even weaker for short-time severe precipitation exceeding the historical extreme value.

(2) In the process of the squall line's development and systematization, the cells in front of the squall line developed rapidly, persisting more than 1 h, with the features of a V-notch, an overhang structure, and a weak echo zone, etc. The forward inflow was strong, forming a moderate-intensity cyclone, and the radial convergence was obvious at the middle level. Thus, it was a supercell that has been rarely captured by C-band radars in the central and western provinces of northern China. The supercell survived 36 min in Pingshun County, incurring extreme short-time heavy rainfall that broke the historical record in this region. The characteristics of sharp increase in VIL, large-magnitude increase in R, and persistent rise in ET, etc. were observed 23-30 min before the precipitation event began, which is indicatively meaningful in qualitative nowcasting of severe precipitation.

(3) The quantitative analysis of the radar echo parameters reveals that the development and maturity of the convective cells began with the strengthening of the upward motion of the in-cloud particles and the growth of the ice particle concentration. When the growth of V₄₀ slowed down, the"sudden drop"in FV₄₀ was a precursor signal of cells'coalescence and rapid development; when the FV₄₀ values above 6 km varied inversely with those below 6 km, precipitation would no longer develop. The changes in the core area of strong echoes at different heights denote that the development of echoes started from the core and expanded outward. The simultaneous occurrence of"sudden drop"in the strong echo cores with 40 dBZ and 50 dBZ was an important signal for the development of precipitation. The great reduction in the echo cores meant the intensity of precipitation was strong. In the vertical direction, the convective cells developed in the form of a"spire", and the increase of ice particles at upper levels was earlier than that of liquid particles at lower levels. The decrease of the high-level cross-sectional area plus the increase of the low-level cross-sectional area was an important indication for the onset of heavy precipitation. The rapid decrease of cross-sectional area at the high level occurred earlier than at the low level, which was caused by the rapid decrease in the concentration of ice particles and slow decrease in the concentration of hydrometeor particles, being considered an important reason for the maintenance of heavy precipitation. The echo comprehensive parameter (CP) prediction model constructed based on the random forest principle can quantitatively predict the precipitation of 80-100 mm h^-1 40-50 min in advance.

(4) An analysis of the densely-observed surface data indicates that, during the generation, development and coalescence of the thunderstorm cells, there was always a mesoscale dewpoint front on the surface, accompanied by a strengthening southerly or easterly wind. The mesoscale dewpoint front and mesoscale convergence lines interacted with the severe convection, making the thunderstorm cells highly systematized. The mesoscale convergence line formed by the increasing northerly wind and its meeting with the southerly wind or easterly wind was an important element causing the formation and development of cells and triggering heavy precipitation. This finding is indicative for the prediction of cell development 2 h in advance.

(5) The numerical simulation analysis has verified the results of the radar echo parameters. The ambient atmosphere was rich in moisture, and the convective cells developed quickly by rapidly acquiring moisture from the ambient atmosphere and being fed by the mother cell. The cloud water distribution had a large vertical range. The cold cloud top extended to 12 km, the large-value area was upright, narrow and deep, and there was a strong and deep vertical upward motion in clouds, which was conducive to a rapid increase of ice crystals. The graupel mixing ratio center was located vertically below the mixing ratio centers of ice crystals and snowflakes, and its vertical distribution range was wide. The graupel mixing ratio was the biggest among all the mixing ratios of water-borne particles. Graupel and hail corresponded to strong ascending motion, creating a large coexistence area of graupel, ice crystals and snowflakes. The raindrop mixing ratio center was below the ascending motion, extending from the near-surface layer to the 4 km height with wide horizontal scale and high precipitation efficiency. These may be the important reasons for the occurrence of the short-time heavy precipitation that surpassed the historical extreme value.

In the forecasting operation, comprehensive analyses of different-scale systems should be strengthened, especially the development and evolution of surface mesoscale systems, radar parameters and microphysical features (Thurai et al. ^[33]; Nzeukou et al. ^[34]; Gorgucci et al. ^[35]; Tang et al. ^[36]; Chen and Diao ^[37]; Janapati et al. ^[38]; Patade et al. ^[39]; Homeyer and Kumjian ^[40]; Stith et al. ^[41]; Wang et al. ^[42]). With the wide use of polarimetric radar and raindrop disdrometer, the understanding of the microphysical characteristics of supercells will be further deepened (Didlake Jr and Kumjian ^[43]; Janapati et al ^[44]; Seela et al ^[45]; Kumjian ^[46]; Kumjian and Ryzhkov ^[47]; Ryzhkov et al. ^[48]; Brandes et al. ^[49]; Raupach and Berne ^[50]). In the future, other physical parameters will be added to test and update the model when comprehensive parameters are constructed so as to improve the prediction of lead time warning and extremity of short-time heavy precipitation.