Climatology of Shear Line and Related Rainstorm over the Southern Yangtze River Valley Based on an Improved Intelligent Identification Method

Horizontal shear lines play an important role in regional rainstorms and severe convection in China (Gao et al. ^[1]; Fu et al. ^[2]; Luo et al. ^[3]; Sun et al. ^[4]). Research on the mechanism of shear line and rainstorm revealed that the instability of the vortex sheet along wind shear lines would lead to several torrential rain centers (Gao^[5]). Observational studies and simulations also showed that a long-lived quasi-stationary lower-level shear line ahead of a 500 hPa trough was the direct trigger for a persistent heavy rainfall event (Fu et al. ^[2]); the accumulated rainfall mainly stretched zonally and fell south of shear line, where southerly flow prevailed and convergence and updraft were evident (Shao et al. ^[6]; Wang et al. ^[7]). Moreover, the precipitation centers were associated with the growth of a meso-scale wave along the shear line (Kawashima and Fujiyoshi ^[8]). Statistics indicated that most of the plateau shear lines were maintained for six hours, and they could cause 40% of the local rainstorms from May to October (Zhang et al. ^[9]; Liu and Li ^[10]), but in the Yangtze-Huaihe River Basin, the proportion of shear line rainstorms increased to 72% between June and July (Ma and Yao ^[11]). However, as with other important rainstorm and flooding disaster areas in China (Chen and Zhai ^[12]; Cui et al. ^[13]), the characteristics of shear line rainstorms in the Southern Yangtze River Valley (SYRV) are still unclear at present.

Furthermore, objective shear line identification is fundamental to the study of shear line rainstorms and the development of intelligent weather forecast (McGovern et al. ^[14]; Reichstein et al. ^[15]; Ham et al. ^[16]; Xia et al. ^[17]; Liu et al. ^[18-^19]; Fathi et al. ^[20]). Although subjective identification by forecasters is relatively accurate, identification by different forecasters is different due to the influence of personal experience, and it is difficult to obtain quantitative characteristics of shear lines. Since intelligent methods can eliminate subjective differences and quantitatively represent shear lines, they are more advantageous in statistical research and operational applications. In recent years, intelligent methods for shear line identification have rapidly developed, and can be categorized into three types: those based on synoptic physical meaning (Zhang et al. ^[9]; Du et al. ^[21]), those based on computer graphics (Liu and Li ^[10]; Huang et al. ^[22]; Hou and Gao ^[23]), and those based on deep learning (DL) techniques (Ryan et al. ^[24]). They are proven to be effective in the studies on the Tibetan Plateau and Yangtze-Huaihe River Basin, but whether they are suitable for the SYRV remains to be tested since different areas have various flows.

The primary goal of this paper is to study the climatic characteristics of the shear lines over the SYRV and their relationship with summer rainstorms by using an improved intelligent method of shear line identification suitable for complex airflows. The remainder of this paper is organized as follows: Section 2 introduces the data and methodology employed in this study. Section 3 evaluates the wind field adaptability of four reanalysis datasets. Section 4 describes the results of shear line identification. Section 5 investigates the statistical characteristics of shear line rainstorms. A brief summary is provided in Section 6.

The data used in this paper are as follows: (1) The stational sounding data and hourly precipitation data after quality control are provided by the China Integrated Meteorological Information Sharing System (CIMISS, Xiong et al. ^[25]). (2) Four global reanalysis products (Table 1) including CMA-RA, ERA5 (ECMWF Reanalysis version 5), ERA-Interim (ECMWF Reanalysis-Interim) and FNL (Dee et al. ^[26]; Berrisford et al. ^[27]; Albergel et al. ^[28]; Hersbach et al. ^[29]; Chen et al. ^[30]). The CMA-RA is the first generation global atmospheric reanalysis product produced by the China Meteorological Administration (Liu et al. ^[31]; Zhao et al. ^[32]; Ye et al. ^[33]), and its rationality and reliability for analyzing large-scale atmospheric circulation has been demonstrated by Yu et al. ^[34]. Notably, all reanalysis products are adjusted to the same horizontal resolution of 0.25° to ensure the consistency of comparisons before wind assessments.

Given the time span and integrity of data, in this study, wind components are evaluated by using 12-hourly sounding data collected during the summers (June-August) between 2016 and 2018, and shear lines are statistically analyzed by using 6-hourly reanalysis data collected between 2008 and 2018. According to the geographical division by the China Meteorological Administration, the research scope of the SYRV in this paper is selected as 24.5° N-31.5° N, 108.8° E-122° E. This area mainly includes Hunan, Jiangxi, Zhejiang, Fujian, Shanghai, Hubei, and Anhui Provinces with a total of 16 sounding stations (Fig. 1) located in 16 cities including Shaowu, Changsha, Yichang, Wuhan, Nanchang, Hangzhou, Hongjia, Ganxian, Enshi, Baoshan, Guilin, Anqing, Chenzhou, Fuzhou, Quzhou, and Huaihua. To conduct a comparative test with sounding observations, we bilinearly interpolated the wind field of the reanalysis grid data to the 16 sounding stations.

Figure 1.  Distribution of 16 sounding stations (green dots) and 455 national meteorological stations (red dots) in the Southern Yangtze River Valley. The black box marks the Southern Yangtze River Valley in Lambert map projection.

Moreover, meteorological stations may not continuously be measured on an hourly basis over a 10-year period due to station relocation, station number changes, and recent establishments or removals. To increase the number of samples, 455 stations (Fig. 1) with data from eight consecutive years (three-quarters of the research period) were selected as statistical stations for rainstorms.

In this paper, shear line refers to the horizontal transverse wind shear, i.e., the discontinuity interface of zonal wind velocity. It is characterized by a discontinuity line of the cyclonic shear in the wind field.

The effective shear line identification method for both the Tibetan Plateau and the Yangtze-Huaihe Basin (Zhang et al. ^[9]; Ma and Yao ^[11]) is used for verification in the SYRV. The identification criteria are as follows:

where u is the zonal wind velocity, and ζ is the relative vorticity.

A low pressure with cyclonic circulation can be obtained using this formula, including transverse shear line and continuous cyclonic curvatures. Affected by the prevailing southwest flow, cyclonic curvatures are very common in the low levels over the SYRV, but this paper only focuses on horizontal transverse shear line, a typical rainstorm forcing system. Therefore, a new method is needed to adequately identify and address the transverse wind shear line and effectively eliminate the continuous cyclonic curvatures.

On the basis of previous research, and in combination with the regional characteristics of the SYRV, the parameter $ \frac{\partial^2 u}{\partial y^2}=0$ was added to the original shear line criteria as a new criterion to eliminate continuous cyclonic curvatures. This new criterion has the following specifications: (1) from a synoptic point of view, $\frac{\partial^2 u}{\partial y^2}=0 $ is a Rayleigh instability condition and a necessary condition for barotropic instability; (2) from the perspective of mathematical physics, the zero point of the second derivative represents the inflection point of the numerical value, where it can represent the mutated characteristics of the horizontal wind. The introduction of this criterion can theoretically and effectively eliminate the continuous cyclonic curvature and only retain a significant discontinuity wind.

The improved shear line identification criteria are as follows:

where u is the zonal wind velocity, and ζ is the relative vorticity.

Although the new criterion ignores the β effect, it is also reasonable, as the zero point of the second-order wind shear is the inflection point of the wind field, and the zero point of the first-order wind shear is the extreme point of the wind field. These points correspond to the vorticity gradient and the vorticity, respectively. Thus, this criterion adopts barotropic instability. Moreover, since barotropic instability is unrelated to atmospheric baroclinicity, only horizontal shear airflow is necessary. Therefore, the unstable development of the shear wave and the vortex wave is consistent with the necessary conditions required for barotropic instability (Li and Wan ^[35]).

Identification steps for shear lines are as follows. First, find the potential shear line based on the original method, and then judge whether the u and $ \frac{\partial^2 u}{\partial y^2}$ of the adjacent grid points on both sides of the potential shear line are reversed. When both u and $\frac{\partial^2 u}{\partial y^2} $ on the sides are reversed and the potential shear line reaches three longitudes from east to west, it is determined as the shear line. Due to the limitation of data resolution in the actual calculation, two zero lines usually fall between grid points rather than on grid points. At this point, the zero line position is obtained by interpolations of the adjacent four grid points, so the coordinates of the two zero lines after interpolation may not completely coincide. As a result, this paper does not directly use the $ \frac{\partial^2 u}{\partial y^2}=0$ lines; instead, according to the values of grid points on the two sides, it determines whether there is a zero line between grid points using reverse signs. Furthermore, the zero line of zonal wind sometimes falls in the due south (north) wind; therefore, this can be ruled out by judging whether the zonal wind on both sides of the potential shear line is reverse.

To define the shear line: Based on the above identification method, when at least one shear line is identified by the reanalysis data at a certain time (00:00 UTC, 06:00 UTC, 12:00 UTC, and 18:00 UTC) within the SYRV, it is considered that a shear line appears once at that time.

To define a short-duration rainstorm: A short-duration rainstorm is defined as a precipitation event where accumulated precipitation reaches 20 mm within 3 h. Additionally, when at least 1% (five) of the national meteorological stations in the SYRV experience a rainstorm (≥ 20 mm (3h) ^-1), it is considered a regional rainstorm process. Here, the 3 h precipitation at a given time is defined as the amount of rainfall within the 1.5 h before and after the given time.

A shear line rainstorm is defined as a period when there is both at least one shear line and a short-duration rainstorm in the SYRV at a given time.

To determine the relationship between different types of shear lines and short-duration rainstorms, we classified the weather pattern of shear line rainstorm by using two unsupervised machine learning algorithms including the t-distributed stochastic neighbor embedding (t-SNE) method and the k-means clustering method. The t-SNE method was used to reduce dimensions by data similarity and showed a better effect than the empirical orthogonal function (EOF) dimension reduction method (Fischer et al. ^[36]; Linderman and Steinerberger^[37]). The k-means method was used to classify data with similar characteristics after dimension reduction. Three variables after standardization disposal were used for dimension reduction, including 850 hPa gridded (42 × 23) wind fields (meridional and zonal wind) and the stational precipitation at 455 stations. The calculation area was the SYRV, and the optimal classification number was determined by using the maximum silhouette coefficient (Rousseeuw ^[38]; Zhou and Gao ^[39]).

Based on the observational data from 16 sounding stations in the SYRV two times per day from 2016 to 2018 (00:00 and 12:00 UTC), the 850 hPa wind fields (wind direction and wind velocity) of four reanalysis datasets were evaluated. The dataset with the best reproduction ability of the wind field was selected to identify low-level shear lines over the SYRV.

There were three assessment indicators (Table 2) including the correlation coefficient (CC), root mean square error (RMSE), and mean absolute error (MAE). First, the assessment indicators of each station were calculated twice per day. Then, the monthly average of each station was calculated, and the evaluation results for each station in summer (June-August) were obtained for the four reanalysis datasets.

Figure 2 illustrates the assessment results of the 850 hPa wind fields among the four products. In contrast, CMA-RA has the highest CC, lowest RMSE, and lowest MAE for both wind direction (Fig. 2a, 2c, and 2e) and wind velocity (Fig. 2b, 2d, and 2f) than those of the other three products, indicating that CMA-RA exhibits the best reproducibility of the wind field in the SYRV. Therefore, this paper adopted CMA-RA to conduct the identification and statistical analysis of the shear line over the SYRV.

Figure 2.  Evaluation of 850 hPa wind direction (left panels) and speed (right panels) of different reanalysis datasets over the Southern Yangtze River Valley between 2016 and 2018. (a, b) Correlation coefficients, (c, d) root mean square error, and (e, f) mean absolute errors. The small dots represent the monthly average value distribution of different sounding stations in summer (JuneAugust), and the red dots represent the average value.

Figure 3 shows three examples of results of shear line identification using the old method and the new method, respectively.

Figure 3.  Three identification cases of shear line. Left panels are Case Ⅰ, middle panels are Case Ⅱ, and right panels are Case Ⅲ. (a, b, c) Zero line of u (brown line, units: m s^-1), u shear (< 0, colored area, units: 10^-5 s^-1) and vorticity (> 0, slash area, units: 10^-5 s^-1); (d, e, f) shear line of original method; (g, h, i) second-order u shear (colored area, units: 10^-10 m^-1 s^-1); (j, k, l) shear line of improved method. The black boxes represent the Southern Yangtze River Valley.

In Case I, the two methods recognized the same shear line (Fig. 3d and 3j), and its wind shear, rotation and discontinuity were strong. The vorticity reached 2×10^-4 s^-1, $\frac{\partial u}{\partial y} $ exceeded 10^-4 s^-1 (Fig. 3a), and the $ \frac{\partial^2 u}{\partial y^2}$ strength of both sides was ±10^-9 m^-1 s^-1 (Fig. 3g).

In Case II, there were three shear lines defined by the old method, but two of them were diminished by the new method (Fig. 3e and 3k). The two false"shear lines" belonged to continuous cyclonic curvatures involving a northerly wind, and their dynamic rotational and shear characteristics, such as vorticity (10^-5 s^-1), $ \frac{\partial u}{\partial y}$ (10^-5 s^-1) and $\frac{\partial^2 u}{\partial y^2} $ (10^-10 m^-1 s^-1), accounted for only onetenth (10%) of the remaining shear line (10^-4 s^-1, 10^-4 s^-1, 10^-9 m^-1 s^-1, Fig. 3b and 3h).

In Case III, the shear line defined by the old method was diminished by using the new method (Fig. 3e and 3k). This false "shear line" belonged to continuous cyclonic curvatures involving southerly wind. Similar to the two false"shear lines"in Case II, its rotational and wind shear were very weak (Fig. 3c and 3i).

Overall, the observation results of the three examples were consistent with the theoretical analysis. Before $\frac{\partial^2 u}{\partial y^2} $ was introduced, the shear line obtained by using the original method contained a low pressure with cyclonic circulation and certain intensity. Only horizontal wind shear with a significant discontinuity wind could be identified by using the new method in this study. Moreover, when using $ \frac{\partial^2 u}{\partial y^2}$ to eliminate the continuous curvature, there was, in some instances, an excessive elimination phenomenon of 1 to 2 grid points (30-60 km) at both ends. This might be the accuracy error caused by two central differences and bilinear interpolations in the calculation process. This error could be reduced with increased data resolution.

According to the definition of shear line, wind variation is a key factor that presents the characteristics of shear lines. Here, the dynamic convergence and rotational characteristics of the shear line were presented by the divergence minimum and relative vorticity maximum of the shear line. Moreover, the minimum value of the Okubo-Weiss (OW) parameter was used to present the deformation characteristics of the shear line (Okubo ^[40]; Weiss ^[41]; Liu and Li ^[42]).

According to statistics, during the summers from 2008 to 2018, a total of 383 shear lines appeared over the SYRV, with an average of 11.6 times per month (Fig. 4a). Furthermore, 193 cases were observed in the daytime (00: 00 UTC and 06: 00 UTC), and 190 cases were observed at nighttime (12:00 UTC and 18:00 UTC).

The occurrence frequency of shear lines appearing in the daytime and nighttime was similar, but the intensity had a clear diurnal variation: the convergence, rotation, and deformation in the daytime were stronger than those at nighttime (Fig. 4b, 4c, and 4d).

Figure 4.  Frequency distribution of shear lines over the Southern Yangtze River Valley in the summers from 2008 to 2018 (a) and its intensity characteristics including (b) divergence (units: 10^-6 s^-1), (c) relative vorticity (units: 10^-5 s^-1), and (d) Okubo-Weiss parameter (units: 10^-10 s^-2). The small dots represent the value distribution of each shear line, and the red dots represent the average value.

With year-by-year variation considered, the daytime and nighttime shear line frequency failed to pass the Mann-Kendall trend test with a confidence level of 98%, indicating that the annual variation trend in shear line frequency over the last ten years was not significant.

According to spatial distribution, no matter whether during the day (Fig. 5a and 5b) or at night (Fig. 5c and 5d), high-frequency areas were distributed in the opening area between mountain ranges running from east to west and to the north of the Nanling Mountains. The shape of the shear lines was in a quasi-east-west zonal arrangement. This may be because when cold air moved southward, warm and moist air moved northward, and these winds were blocked by the terrain. Additionally, they were more likely to intersect and form shear lines in areas with relatively flat terrain.

Figure 5.  Spatial distribution of shear lines over the Southern Yangtze River Valley in the summers from 2008 to 2018: (a) 00:00 UTC, (b) 06: 00 UTC, (c) 12: 00 UTC, and (d) 18: 00 UTC. The brown line represents the shear line. The black box marks the Southern Yangtze River Valley.

According to statistics, during the summers from 2008 to 2018, there were a total of 383 shear lines and 1, 158 short-duration rainstorms in the SYRV. Among them, half of the shear lines (181 times, accounting for 47%) were able to cause rainstorms, and shear line rainstorm accounted for one-sixth (16%) of the total short-duration rainstorms. This conclusion was essentially consistent with local forecasters'predictions. In addition to shear lines, other systematic forcings, including westerly trough, vortex, and low-level jet, also greatly contributed to short-duration rainstorms in the SYRV. To analyze the relationship between shear lines and short-duration rainstorms, this section uses two unsupervised machine learning methods to distinguish different types of shear lines rainstorms.

Figures 6 and 7 compare the distribution and synoptic situation of shear line rainstorms with other rainstorms. It can be seen that the rainstorm caused by shear line (Fig. 6a) was stronger than that by other synoptic forcing (Fig. 6b). This was because horizontal transverse wind shear lines were always accompanied by stronger barotropic instability and significantly stronger water vapor transport (confidence level 98%), as well as the interaction of cold and warm airflow at the interface between the horizontal easterly wind and westerly wind (Fig. 7c).

Figure 6.  Rainstorm (color area, units: mm (3h)^-1) with or without shear lines (brown line) in the Southern Yangtze River Valley in the summers from 2008 to 2018. (a) Distribution of shear lines with rainstorm and average precipitation. (b) Average precipitation of rainstorm without shear line. The black box marks the Southern Yangtze River Valley.

Figure 7.  Composite analysis of synoptic situation of shear line rainstorms (a), other rainstorms (b) and their differences (c). The synoptic situation includes wind (green barb, units: m s^-1), equivalent potential temperature (red contour, units: K), and moisture flux (shading, units: g cm^-1 hPa^-1 s^-1) at 850 hPa. The pink dotted areas are significant at the confident level of 98%. The black boxes are the Southern Yangtze River Valley.

Furthermore, daytime shear line rainstorms (109 times) were more frequent than nighttime rainstorms (72 times), and the regional average intensity was significantly stronger in the daytime (2.5 mm) than at nighttime (1.9 mm, confidence level 98%), which was consistent with the diurnal variation in shear line intensity (Fig. 4).

Two unsupervised machine learning methods were used for classifying 133 shear line rainstorms with complete data. Table 3 shows the silhouette coefficient distribution of different k-mean clustering numbers on t-SNE dimension reduction data. It can be seen that the Silhouette coefficient reached the maximum at number 3; therefore, the optimal cluster number was set at 3. At this time, frequencies of shear line rainstorms of three types were 41, 36, and 56, respectively. Table 4 quantitatively describes the intensity of the three type shear lines and rainstorms. Fig. 8 shows the weather patterns corresponding to three type shear line rainstorms.

Figure 8.  Composite analysis of synoptic situation (left panels), and shear line and its precipitation (right panels) based on k-means clustering. The synoptic situation includes geopotential height (blue contour, units: dagpm) at 500 hPa and the wind (green barb, units: m s^-1), equivalent potential temperature (red contour, units: K), and moisture flux (shading, units: g cm^-1 hPa^-1 s^-1) at 850 hPa. The right panels include shear lines (brown line) and average precipitation (color area, units: mm (3h)^-1). (a, b) First category, (c, d) second category, and (e, f) third category.

Overall, both the first and second type of shear lines were stronger in convergence, rotation, and deformation intensities than the third type, as well as concentrated rainstorms (Table 4). They were concentrated in the north and middle of the SYRV (Fig. 8b and 8d). The weather patterns showed that northern area of the SYRV was affected by northerly winds behind the 500 hPa East Asia deep trough, and the southern area of the SYRV was in front of the southern branch trough. The shear line was formed when the cold air from the north met the strong southwest warm moist airflow, and it was essentially consistent with the intensive belt of the potential pseudo-equivalent temperature (θse). Cold-warm and dry-humid differences between the north and south sides were clear. The difference between these two types was that the first type of shear line was farther north with a stronger intensity and higher rainfall (see Table 4). This was because the southwest airflow was stronger and more unstable, whereas the dynamic and thermal conditions were more favorable for precipitation.

The 500 hPa geopotential height of the third type was flatter than those of the first two types (Fig. 8e). The intersection of 850 hPa south and north wind was less intense than that of the first two types (Fig. 8f), and the precipitation intensity was the weakest. (Table 4). Moreover, the shear line rainfall along the southeast coast was stronger than that of shear lines in the opening area, which may be related to the enhancing effect of the coastal mountain terrain.

This paper proposes an improved intelligent method for identifying shear lines by adding a second-order zonal-wind shear and utilizing an intelligent classification method to study the characteristics of shear line rainstorms over the SYRV in the summers from 2008 to 2018, and reaches the following conclusions:

(1) The reproducibility of 850 hPa wind fields over the SYRV from CMA-RA is superior to the reproducibility of ERA5, ERA-Interim, and FNL, as measured by CC, RMSE and MAE.

(2) The occurrence frequency of shear lines over the SYRV is almost the same in the daytime as that at nighttime. However, the intensity has a clear diurnal variation: the convergence, rotation, and deformation in the daytime are stronger than those at nighttime.

(3) Half (47%) of the shear lines can cause short-duration rainstorms in summer, and shear line rainstorms account for one-sixth (16%) of the total number of rainstorms. The comparison between shear line rainstorms and other rainstorms shows that the rainfall caused by shear lines is stronger than that caused by other synoptic forcing, which is related to significantly stronger water vapor transport and barotropic instability. Moreover, the intensity of shear line rainstorms exhibits the same diurnal variation with that of shear lines.

(4) The classification results show that both stronger shear lines and stronger precipitation are concentrated in the north and middle of the SYRV since stronger shear lines are usually accompanied by stronger warm and humid southwest airflows.