A Modeling Study on the Mechanism of Convective Initiation of a Squall Line over Northeastern China

Deep moist convection (DMC) and its accompanied heavy rain, lightning, strong winds bring many hazards each year, especially during the summer months when both the thermodynamics and dynamic conditions favor DMC. Yet despite of a number of studies focused on convective initiation (Wilson et al. ^[1]; Kingsmill ^[2]; Ziegler ^[3]; Wakimoto and Murphey ^[4]), predicting the initiation of DMC is still a challenge (Bennett et al ^[5]; Hane et al. ^[6]; Olson et al. ^[7]; Weckwerth and Parsons ^[8]), with major difficulty being the prediction of the exact timing and location of convective initiation.

Convective initiation (CI) links closely to convergence. Studies have shown that convergence can be detected 15-90 min prior to the convective rainfall (Atkins et al.^[9]; Ulanski and Garstang ^[10]), and the time between which is just about the time convection needs to reach mature stage after initiation. Wilson and Schreiber demonstrated that 71% of the thunderstorms initiated in Colorado are associated with boundary-layer convergence lines using radar images ^[11]. And the forcing of surface convergence could be classified into seven categories: frontal, gust front, trough lines, drylines, colliding, bore and unknown (Wilson and Roberts ^[12]). Convergence lines are sometimes precursors to convective initiation; therefore, much attention has been paid to convergence lines detection in predicting where and when the convection would initiate. Yet convergence lines do not necessarily trigger convection. For example, convection is not triggered along the entire convergence line. Bluestein et al. ^[13] showed that convection is often initiated at the"triple point", which is the intersection between two air mass boundaries where three distinct air masses converge (air mass boundaries includes synoptic fronts and drylines). Other studies showed that the regions along convergence lines where misocyclones existed were areas favorable for convective initiation(Lee et al. ^[1]; Marquis et al. ^[14]). However, CI does not necessarily occur along convergence lines. For example, on 24 May 2002 during International H₂O Project (IHOP), field operation focused on the triple point, but the convection was initiated somewhere else (Xue and Martin ^[15]). In addition, some convergence lines failed to trigger convection. Markowski and Hannon showed that CI failed near the triple point due to a lack of persistent and spatially mesoscale ascent (Markowski et al. ^[16]). Weckwerth ^[17] demonstrated that the stability parameters, midlevel moisture, and vertical shear obtained by sounding showed no difference in the horizontal convective rolls between storm and non-storm day. Thus, it is hard to predict CI by merely detecting convergence.

Some studies focused on the inhomogeneity along convergence lines to explain why convection was triggered at some specific spots. Updrafts might be enhanced at the intersection of boundaries and the updrafts of horizontal convective rolls (HCRs) and thus favored CI (Wilson et al. ^[2]; Xue and Martin ^[18]; Fovell ^[19]). Many studies showed that mesocyclones along the convergence line were accountable for the place where the convection initiated (Marquis et al. ^[14]; Lee and Wilhelmson ^[20]; Buban and Ziegler ^[21]). Other possible mechanisms include low level moisture variations (Droegemeier and Wilhelmson ^[22]), wind shear (Droegemeier and Wilhelmson ^[23]; Bluestein and Weisman ^[24]), land and sea breeze (Kingsmill ^[25]; Xing and Li ^[26]), soil moisture (Holt et al. ^[27], Meng et al. ^[28]), and gravity waves (Liu et al. ^[29]; Mccarthy and Koch ^[30]).

Compared with observations, numerical simulations have huge advantage of their spatial and temporal resolution in studying the mechanism of CI. Lee et al. concluded that the timing and intensity of convection was affected by boundary layer moisture convergence using a two-dimensional model ^[1]. Ziegler et al. showed that thermal driven secondary circulations along the dryline destabilized local environment and supported convection using a three-dimensional model ^[31]. Lee and Wilhelmson presented evolution of a series of mesocyclones along an outflow boundary and their effect on CI ^[20]. Several studies showed that CI is related to the downdrafts of HCRs and moisture bulge using high resolution model output (Xue and Martin ^[18]; Murphey et al. ^[32]). The present study uses high resolution simulation data to investigate mechanism of CI of a squall line, with the hope that this research will enhance our understanding of the role that mesocyclones and moisture field play in CI.

The rest of this paper is organized as follows. In section 2, an observational overview of the case is discussed. Section 3 introduces the numerical experiment and validates it against the observational data. In section 4, the mechanism of the CI in this case is presented. The final section is the conclusion and discussion.

The squall line of interest occurred in northwestern Liaoning Province, China along a convergence line in front of a surface low on 26 June, 2014. A series of convective clouds were triggered around 0230 Universal Time Coordinated (UTC). These clouds developed into NE-SW oriented consecutive convective line around 0900 UTC as they moved southeastward. Afterwards, the northern part of the convective line dissipated gradually while the southern part of it kept developing and became an intense squall line. The squall line reached its mature stage and evolved into a bow echo around 1200 UTC at the coast of Liaodong Gulf, bringing heavy rains, lightning and strong winds. The populated region of Liaoning locates near the coast. Dalian, the provincial capital city of Liaoning, locates near the coast and thus suffered from strong winds of more than 20m s^-1 and heavy rain. A few trees and advertising boards were torn down and streets were flooded. Some electronic equipment in Dalian airport were destroyed by the lightning and a number of flights were canceled. The squall line gradually decayed as it continued moving southeastward. Then it broke into discrete convective clouds and dissipated eventually around 1600 UTC in Liaodong Gulf.

The squall line formed in an environment favorable for convection. At 0000 UTC 26 June 2014, Liaoning was in the left-front flank of the 200 hPa jet stream. Ascending motion usually exists in the left-front flank of jet steam due to ageostrophic circulations and provides favorable conditions for convection (Fig. 1a). At 500 hPa level (Fig. 1b), Liaoning was in front of a deep trough. The deep trough was NE-SW oriented, the trough of temperature was in the west of the trough of geopotential and cold air was advected towards Liaoning from the rear of the trough. The trough developed and moved southeastward. At 850 hPa level (Fig. 1c), Liaoning located at the south of a low vortex. A branch of northeasterly low-level jet stream was right across Liaoning. The jet stream intensified between 0000 UTC and 0600 UTC. The low level jet induced cyclonic and anticlonic vorcitity and favored the convection. At the surface level (Fig. 1d), a narrow low was trapped in northeast China by a continental and a marital high. This low elongated and moved southeastward. Liaoning located south of this surface low. A branch of northwesterly flow in the rear of the low constantly moved dry and cold air to Liaoning and a branch of southerly flow originated from Liaodong Gulf brought moist and warm air into the land. These two branches of strong surface flow encountered in the northwest of Liaoning and caused convergence and wind shift zone there. Convection was triggered in this zone.

Figure 1.  (a) The temperature (dash lines; units: ℃), geopotential height (solid lines; units: dgmp) and wind speed (shaded; units: m s^-1) at 200 hPa; (b) the temperature (dash lines; units: ℃), and geopotential height (solid lines; units: dgmp) at 500 hPa; (c) the temperature (dash lines; units: ℃), geopotential height (solid lines; units: dgmp) and wind (vector, units: m s^-1, larger than 12m s^-1 is plotted) at 850 hPa; (d) the wind at 2m (vector) and sea level pressure (solid lines; units: hPa).

The signals that triggered the convection might be just a few kilometers, which is not resolvable in observation because of the density of observation stations. On the other hand, numerical model has huge advantage in their spatial and temporal resolution in studying the mechanism of CI. Thus, a numerical experiment was carried out. The numerical experiment was performed by version 3.5 of the Weather Research and Forecasting (WRF) model using two-way nesting with the coarse and fine resolution of 4km and 1.33km, respectively (Skamarock et al.^[33]). Both domains had 51 vertical layers with fixed top at 50 hPa and 910 × 892 horizontal grid points. The simulation area is shown in Fig. 2. With initial and boundary conditions provided by 6-hourly 1°×1° Final Operational Global Analysis (FNL) data from National Centers for Environmental Prediction (NCEP), the simulation was integrated for 24h, starting at 0000 UTC 26 June 2014. The schemes used for physics and dynamical parameterization were Morrison 2-moment for the microphysics and asymmetrical convective model version 2 (ACM2) for the planetary boundary layer respectively (Morrison et al.^[34]; Pleim ^[35]). No cumulus parameterization was used in either of the two domains. The fine resolution outputs were used to investigate the mechanism of CI.

Figure 2.  Model domains and terrain.

The simulation outputs were verified against observational data, including radar reflectivity, hourly rainfall, surface wind and sounding. The hourly rainfall was the merged precipitation product from China (Shen et al. ^[36]). Results showed that the simulation adequately reproduced the life cycle of the squall line, such as the time and location of the initiation, motion speed, structure, evolving and dissipating process.

The radar nearby did not work for some reason during the first few hours after the convection was triggered. Therefore, no radar reflectivity image could be used to verify the simulation in the first few hours. However, the 1-hour accumulated rainfall at 0400 UTC, i. e., the rainfall collected 30min to 90min after the CI, could be used to verify the simulation outputs during the triggering time in another aspect. Fig. 3a shows the observed 1-hour accumulated rainfall at 0400 UTC. In the convergence zone in northwest Liaoning, a center of rainfall can be easily identified, with the maximum 1-hour accumulated rainfall of 4-6mm. The 1-hour accumulated rainfall (Fig. 3e) at the same time of simulation reproduced the observation field quite reasonably, with the rainfall center located in almost the same region in northwest Liaoning, except that the maximum 1-hour accumulated rainfall of simulation was a bit stronger, reaching 8-10mm, and the simulated rainfall was more scattered compared with the observation. The observation also showed two other rain centers in northeast and east of Liaoning which did not exists in simulation. But the rain in these two areas were caused by a decaying mesoscale convective system which was beyond the scope of this study. At 0700 UTC, the observed 1-hour accumulated rainfall (Fig. 3b) showed mainly two rain belts in northwest Liaoning related to the convection that we were interested in. One located near the northwest border of Liaoning with two rainfall centers, while the other in the west of Liaoning with several scattered rainfall centers. The simulated 1-hour rainfall captured the location and the pattern of the observed rainfall well (Fig. 3f), though the rainfall center located in the west of Liaoning was more scattered and stronger compared with the observed. By 1200 UTC, the squall line reached its mature stage and brought heavy rain (Fig. 3c). Heavy rainfall mainly located near the convective region of the squall line in Liaodong Gulf, reaching more than 25mm h^-1. The simulation result matched the observation well in location and intensity, except that the coverage area was a bit smaller (Fig. 3g). At 1500 UTC, the squall line broke into several segments and dissipated gradually (Fig. 3d). The 1-hour accumulated rainfall was much smaller compared with that of 1200 UTC, with the max center located in the east part of Liaodong Gulf. The simulated squall line dissipated more quickly than observation. However, the simulated 1-hour accumulated rainfall still matched well with observation in location with slightly lower intensity compared with the observation at that time (Fig. 3h). Besides, the simulation reproduced the surface wind quite well. As mentioned earlier, two branches of strong surface flow encountered in the northwest of Liaoning and caused convergence and wind shift zone there, which can be clearly observed in Fig. 3. In the northwest of Liaoning, the wind was mainly northwestly while in the middle and east of Liaoning, the wind was mainly southly or westsouthly.

Figure 3.  (a-d) The observed and (e-f) simulated 1 h accumulated precipitation (color shading denotes areas where the 1 h accumulated precipitation was larger than 1 mm) superimposed with surface wind: (a, e) 0400 UTC; (b, f) 0700 UTC; (c, g) 1200 UTC; (d, h) 1500 UTC.

The simulated radar reflectivity captured the location and pattern of the observed radar reflectivity very well, too. At 0900 UTC (Fig. 4a), the first time we could acquire observed radar reflectivity, the convection along the convergence zone ahead of the trough broke into two parts. Convection in the north was scattered and not so well-organized, so it dissipated gradually. Convection in the south was much better-organized. A consecutive convective line (composite reflectivity greater than 40 dBZ) can be easily identified, accompanied by a wide area of stratiform precipitation (composite reflectivity less than 20 dBZ). The convection in the south kept growing and evolved into a squall line. The simulated radar reflectivity reproduced this feature quite well (Fig. 4a), the convection also broke into two parts, and the pattern of each part was similar to the observation, except that the stratiform area was smaller than the observation. We also noticed that an extra convective zone located between the north and south convective zones, which did not exist in observation. However, it dissipated quickly afterwards and did not affect our analysis much. At 1200 UTC (Fig. 4b), the northern part of the previous convection almost disappeared, while the southern part evolved into a mature squall line and somehow showed a bow echo pattern, with the max radar reflectivity exceeding 60 dBZ. The stratiform area was mainly located behind the convective line, and the squall line can be classified as trailing stratiform type as a result. The simulated radar reflectivity captured the location and the type of the squall line well (Fig. 4e). At 1500 UTC, the convective line in the front of the squall line broke into discrete convection and dissipated rapidly afterwards (Fig. 4c). The simulated radar reflectivity reproduced this feature but was more sporadic and dissipated more quickly compared with the observation (Fig. 4f).

Figure 4.  (a-c) The observed and (d-f) simulated composite radar reflectivity (units: dBZ) at (a, d) 0900 UTC, (b, e) 1200 UTC, and (c, f) 1500 UTC.

The profile of wind, temperature and dew temperature was validated against the sounding data as well. The closest sounding stations to the squall line are in Jinzhou and Shenyang. Jinzhou station was affected by this squall line while Shenyang station was not. The moisture was abundant near the surface and around 800 hPa, and the low level wind shear from 925-700hPa exceeded 12m s^-1 at Jinzhou station (Fig. 5a), favorable for the initiation and maintenance of convection. While in Shenyang station (Fig. 5b), there was a dry layer near the surface and quite wet in the mid-level. The simulated sounding reproduced the characteristics of both stations quite well.

Figure 5.  Observed (red lines) and simulated (black lines) skew T-log p diagram at (a) 0000 UTC at Jinzhou; (b) 0000 UTC at Shenyang, with black solid line and black dash line representing simulated temperature and dew point, and red solid line and red dash line representing observed temperature and dew point, respectively.

Verification results showed that the simulation reproduced the features of squall line well, such as the time and location of the initiation, evolution and dissipation process. Thus, the simulation output can be utilized to examine the mechanism of convective initiation.

As mentioned earlier, the squall line was triggered in a convergence zone ahead of a trough. In synoptic aspect, a branch of dry and cold flow and a branch of warm and moist flow encountered in the northwest of Liaoning. The collision provided low-level convergence while the moist flow brought affluent water vapor, making the environment desirable for CI. Fig. 6 is the area averaged sounding taken at 0000 UTC June 26, 2014. The area that are averaged is roughly the convergence zone in northwest Liaoning. From Fig. 6, it can be seen that low-level environment in northwest Liaoning is humid while relatively dry in the mid-level, and the convective available potential energy (CAPE) is considerable (1378J kg^-1), and the lifting condensation level (LCL) is pretty low, at 893 hPa. Thus, when the air is forced to rise due to convergence, it can reach the LCL easily and become more unstable due to condensation heating. Moreover, when it keeps rising and reaches level of free convection (LFC), convection is easily triggered. The convergence line has always existed in low level; however, the convection was not triggered in the whole convergence zone. Thus, question is raised: why convection was triggered in some specific places. Analyses show that the inhomogeneity of the moist field might be accountable for some preferred locations of CI. The details are as follows.

Figure 6.  Area averaged simulated skew T-log p diagram at 0000 UTC June 26th, 2014, with solid line and short dash line representing the temperature and dew point, respectively. And the area surrounded by the long dash line and the solid line is convective available potential energy (CAPE). Full ticks and pennants of windbarb represent 4m s^-1 and 20 m s^-1 respectively. Area that is averaged is between 41.5°N and 42.2°N, and 120.2°E and 121.2°E.

Figure 7a shows the water vapor mixing ratio, water vapor flux divergence, vertical vorticity, and wind at the surface level as well as vertical velocity at 8th model level (at about 830 hPa). A region of large mixing ratio gradient locates in the convergence zone where the two branches of flow encounter. The convection is first triggered at the large water mixing ratio gradient region, but the water vapor flux divergence in the convergence zone is inhomogeneous. In some places, the water vapor converges more intensely and results in several centers of large water vapor flux divergence. Vertical velocity coincides well with the centers of water vapor convergence. The convection is initially triggered where the vertical velocity is large, which is intuitive: the air is forced to rise in the convergence centers, and once it reaches LFC, convection is easily triggered. In addition, the water vapor convergence destabilizes the low-level environment, making the ambient environment favorable for CI.

Figure 7.  (a) The wind vector (minus by average wind in this area, units: m s^-1), water vapor mixing ratio (blue line, units: g kg^-1, at 0.2 g kg^-1intervel), water vaper flux divergence (shaded, units:10^-4g kg^-1s^-1, positive value indicates convergence), vertical vorticity (black line, units: 10^-5s^-1. The outermost contour is 20*10^-5s^-1, incremented by 10*10^-5s^-1) at the surface, and the vertical velocity at 8st model level (red line, units: m s^-1.The outermost contour is 0.2 m s^-1, incremented by 0.3 m s^-1). The grey box and straight line are the area that are used in Fig. 7b and the cross-section in Fig. 8, respectively. (b) The area average northerly wind component v along longitude.

Figure 8.  Simulated skew T-log p diagram at 0220 UTC June 26th, 2014, with solid line and short dash line representing the temperature and dew point, respectively. The area surrounded by the long dash line and the solid line is CAPE. Full ticks and pennants of windbarb represent 4m s^-1 and 20m s^-1, respectively. Soundings (a) and (b) are taken as indicated in the black and red dots in Fig. 7(a), respectively. The grid points of the dots and their nearest 9 grid points are averaged to represent the features of the dots.

As for the reason why the moisture field is inhomogeneous, we suggest that it is due to the vortices near the surface in the convergence zone. A series of quasi-equally spaced vortices near the surface locate in the convergence zone where the wind shift is large. The vortices there distort the flow nearby and cause the inhomogeneity of moisture field. As can be seen in Fig. 7a, the vortices and the water vapor flux divergence centers do not coincide with each other. The vortices rotate the warm and moist air in their east towards their north while cold and dry air in their west towards their south. As a result, relatively moist air locates in the north of the vortices while relatively dry air locates in the south of the vortices. The moist air is more unstable and can reach LFC easier. Thus, spots in the north of the vortices are preferred locations for CI. Some previous studies focused on mesocyclones' in CI, but they focused on how the mesocyclones inflected or distorted the low-level convergence and thus enhanced convergence that favored CI. For example, Marquis et al. ^[14] found that convergence could be enhanced on only one side, both sides, or neither side of mesocyclones. They also mentioned that it was unclear why convergence was enhanced upstream, downstream, upstream and downstream, or not at all near mesocyclones. However, we suggest here that instead of how the mesocyclones affect the low-level convergence, how they affect the low-level moisture field is more important. Convergence might be enhanced at one or both sides of mesocyclones, but only the areas where the moisture converges are favorable for CI, for example in this case, north (including northeast and northwest) of the mesocyclones.

As for the cause of these series vortices near the surface, we believe it is due to horizontal shear instability (HSI). HSI is also called inflection point instability, because a necessary but insufficient requirement of HSI is$ \frac{{∂^2 \bar u}}{{∂y^2}} $ change sign, which means an inflection point is presented in the u profile. Fjørtoft ^[37] proposed a more stringent condition for HSI, i. e., $\frac{{∂^2 \bar u}}{{∂y^2}} ( \bar u - \bar u_I) $must be less than zero somewhere in the flow, where u_I is the wind at the inflection point. In this case, convergence zone was created by the encountering of two branches of northwesterly and southerly flows. Northerly wind component (v) changed dramatically. The change of v might result in HSI. Fig. 7b shows the area average v values along longitude. Near 120.3°E, the function is concave function, thus $ \frac{{∂^2 \bar v}}{{∂x^2}} $ > 0. Near 121.1° E, the function is convex function, thus $ \frac{{∂^2 \bar v}}{{∂x^2}} < 0$. Therefore, it can easily be inferred that $\frac{{∂^2 \bar v}}{{∂x^2}} $ must change sign somewhere between. Moreover, it is easy to find some point where $ \frac{{∂^2 \bar v}}{{∂x^2}} ( \bar v - \bar v_I)$ is less than zero. Thus, the northerly wind component in this region satisfies stringent condition for HSI. In addition, Miles and Howard ^[38] derived that the fastest growing mode of HSI was around 7.5 times of the shear level in ideal case. In real case, it is hard to measure the width of the shear level, so we are unable to testify whether the fastest growing mode of HSI is around 7.5 times of the shear level in this case. However, it can be clearly seen that the vortices along the convergence zone are quasiequally spaced, implying that there is a fastest growing mode for these series of vortices. These features of the vortices suggest that they might be caused by HSI in another aspect.

Figure 8 shows two simulated soundings taken in the north (where the convection firstly triggered, the black dot location in Fig. 7a) and south (the red dot location in Fig. 7a) of the vortices, respectively. Though these two sounding spots are pretty close to each other, they show some differences. CAPE seems not to be accountable for which spot to trigger first. In the north sounding, CAPE is even a bit less than that in south sounding (1675 J kg^-1 vs 1729 J kg^-1). What might be accountable for triggering is LCL. LCL in north sounding is 40 hPa, which is lower than that in the south (878 hPa vs 838 hPa). That means in the north of the vortices, air particles can reach LCL more easily and become more unstable because of condensation heating. The water vapor mixing ratio in the surface of these two spots are almost the same, so the difference of LCL is the result of the difference in temperature lapse rate in the low level. The temperature lapse rate in the north of the vortice is large than that in the south, making the environment more unstable and the air can reach LCL more easily, thus convection can be triggered more easily in the north of the vortice.

Figure 9 is the cross-section taken from where the first convection is triggered. As shown in Fig. 9 (a), an upward moisture bulge can be easily identified at the northeast of the low-level vortices. The moisture bulge destabilizes the environment nearby and about 20 min later (Fig. 9b), the first convection is triggered when the air is forced to rise by low-level convergence and reaches the LFC. The ambient wind advects the convection towards the east (southeast) where the environment is moist and more unstable, favorable for the development of the convection. In addition, the updraft itself creates vertical vorticity by tilling the horizontal vorticity into vertical vorticity, as indicated by the tilling term ω_h · ▽w in vertical vorticity equation (Eq. (1), neglecting Coriolis force and viscous effect, where v and ω_h are horizontal velocity and horizontal vorticity vector, respectively).

Figure 9.  Cross-sections of wind (vectors, units: m s^-1, with vertical velocity w amplified 5 five times), equivalent potential temperature (blue lines; units: K, interval: 2 K), vertical vorticity (black line, units: 10^-5s^-1, starts at±30*10^-5s^-1, interval: 15*10^-5s^-1), vertical velocity (red line, units: m s^-1, interval: 0.5m s^-1) and water vapor mixing ratio(shaded, units: g kg^-1, at 0.5 g kg^-1intervel) at (a) 0200 UTC, and (b) 0220 UTC.

The maximum updraft locates from around 800 hPa to 600 hPa, and the vertical vorticity created by the tilling term also shows a maximum center about the same levels as a result. The horizontal vorticity in these levels is mainly southerly (not shown), so a dipole of negative and positive vertical vorticity is created aside the updraft core. The negative vertical vorticity is in the north of the updraft while the positive is in the south. With the development of the convection, these mid-level vertical vorticity cores extend toward higher and lower levels, reaching near ground. The positive vertical vorticity cores collaborate with the pre-existing near ground vortices (created by HSI) and become vortexes from ground to mid-level. These vortexes distort the moisture field by rotating the moist air toward their north where the updraft is, promoting the development of convection in turn by providing low and mid-level water vapor.

This paper presents the results of simulation of a squall line along a convergence zone northeast of China during June 26, 2014. Version 3.5 of WRF was used to perform the simulation to investigate the mechanism for convective initiation. The simulation output was verified against observed data including 1-h accumulated rainfall and radar reflectivity. Results showed that the simulation reproduced the life cycle of the squall line well, such as the timing and location of the initiation, evolution and dissipation process, as well as structure, intensity and so on.

Simulation output was used to investigate the process of convective initiation of the squall line in this case. Results showed that the squall line was triggered in the convergence zone, where the horizontal gradient of moisture was large because of the encountering of a branch of dry air and a branch of moist air. However, the moisture field in the convergence zone was not homogeneous. Convection was first triggered in the spots where the surface moisture convergence was high. Air was forced to ascend due to convergence, and when it reached LFC, convection was triggered. The inhomogeneity of the surface moisture field was caused by a series of vortices near the surface, which themselves were a result of HSI. Wind shear in the convergence zone, especially the shear of component, resulted in these quasi-equally spaced vortices. These vortices distorted the surface moisture field by rotating moist air into their north while dry air into their south. As a result, moisture in the north of the vortices converged more intensely, creating preferred spots for convection initiation (see Fig. 10 for the conceptual model). After the convection was triggered, the updraft tilled the horizontal vorticity into vertical, creating large vertical vorticity dipole in the mid-level. The positive vertical vorticity cores in the mid-level collaborated with the pre-existing vortices near the surface and became vortexes from near the surface to mid-level. These vortexes promoted the development of convection in turn by rotating the moisture into the updraft.

Figure 10.  Conceptual model for the mechanism of the convective initiation in this case.

Some previous works suggested the vortices might enhance convergence, and thus favor the convective initiation (Marquis et al. ^[14]; Lee and Wilhelmson ^[20]; Buban and Ziegler ^[21]); however, they did not illustrate why the convergence is enhanced at specific spots and triggers convection. Our results also suggested that vortices is accountable for the convective initiation, but instead of enhancing convergence, the vortices'role in reshaping the moisture field and enhancing the convergence of moisture is critical in convective initiation. Convection is triggered at the spots where the moisture converge intensely. However, regarding the generality of these results, as the findings are made from a single case study, uncertainty remains. Results should be cautiously applied to other cases involving convergence zone and wind shear. Besides, other processes that might affect convective initiation, such as the vertical structure of moisture and temperature, wind shear, land surface variation, should be thoroughly investigated to get a profound understanding of convective initiation and further improve the model's ability of predicting the timing and location of convective initiation.