A Climatological Study of Southwesterly Flows and 1 Heavy Precipitation in Taiwan during Mei-yu Seasons 2 from 1979 to 2018 3

17 This paper examined southwesterly flows and rainfall around the Taiwan area during the mei- 18 yu seasons from 1979 to 2018. The occurrence percentage of the southwesterly flow events in 19 southern Taiwan was highly correlated with 6-h accumulated rainfall and heavy precipitation in 20 Taiwan, while those in northern Taiwan showed little correlation. Low pressure to the north of 21 Taiwan and high pressure to the south exerted a large northward pressure gradient force across 22 the Taiwan area, favoring the formation of southwesterly flows and rainfall in southern Taiwan. 23 During an active year of southwesterly flow events, the Pacific high weakens and moisture is 24 transported along two paths in the early mei-yu season: one from the Bay of Bengal and the other 25 from the south of the Pacific high. The moisture-laden air results in high equivalent potential 26 temperature near Taiwan, which in turn creates a large equivalent potential temperature gradient 27 to the north of Taiwan. This setting favors the activity of mei-yu fronts and produces a low 28 pressure environment. The pressure gradient thus increases, supporting the formation of 29 southwesterly flows. The southwesterly flows then help to transport more moisture toward the 30 Taiwan area, resulting in heavy rainfall as well as a further increase of equivalent potential 31 temperature. This kind of positive feedback produces more fronts, stronger southwesterly flows, 32 and heavier rainfall during the mei-yu season. The study also suggests that the meridional 33 component of water vapor transport over the South China Sea and the 34 Philippines in the early mei-yu season can be used to predict the occurrence of southwesterly 35 flows and heavy rain for the entire mei-yu season.


Introduction 38
Taiwan, a subtropical island southeast of China, has a unique climate that is strongly 39 regulated by East Asian monsoons. The monsoonal flows, including northeasterly in cold 40 seasons and southwesterly in warm seasons, largely determine the seasonal rainfall in Taiwan 41 (Ding and Chan 2005;Chen and Chen 2003;Chen et al. 1999). Based on studies employing 42 long-term historical data, Chen and Chen (2003) categorized five different rainfall regimes in 43 Taiwan: winter (December-February), spring (March to mid-May), mei-yu (mid-May to mid-44 June), summer (mid-June to August), and autumn (September-November). This paper primarily 45 deals with rainfall of the warm season regimes from April to August, especially mei-yu, which is 46 predominantly related to the southwesterly monsoonal flow. 47 Mei-yu fronts, which are quasi-stationary fronts, often develop over southeastern China and 48 move southward to Taiwan in mei-yu seasons, resulting in long periods of rainfall (e.g., Wang et 49 al. 2014;Lin et al. 1992;Ding 1992;Chen and Liang 1992;Ray et al. 1991). The fronts usually 50 appear on satellite images as elongated cloud bands that extend from southern China 51 northeastward to Japan (Chen 2004). The cloud bands are typically embedded with mesoscale 52 convective systems (MCS) that can bring heavy rainfall and cause serious disasters to the region 53 (e.g., Ding and Chan 2005;Qian et al. 2004;Zhang et al. 2003;Kuo and Chen 1990). 54 Furthermore, the mei-yu front is usually associated with low-level jets (LLJs) to their south 55 which play an important role in transporting warm moist air from the tropical ocean to the frontal 56 zone (Chen et al. 2008;Chen, C.-S. et al. 2005;Li et al. 1997;Chen and Yu 1988). The wind 57 direction of the LLJs is typically west-southwesterly to southwesterly, with wind speeds stronger 58 than 15 m s -1 at low levels (e. g., 850 hPa) and decreasing both upward and downward (Chen et 59 al. 2006;Chen, G. T.-J. et al. 2005;Chen et al. 2000). However, it is more common to find 60 June 2016, the axis of strong southwesterly winds was located around northern Taiwan at the 120 850-hPa level (Fig. 2a). There were 11 boxes surrounding Taiwan with a mean wind speed 121 greater than 12 m s -1 , and 14 boxes had a southwesterly wind direction; therefore, this was 122 defined as an SW event because both of the first two aforementioned conditions were met. 123 Judged by the third condition, this event was further identified as an SWn event because there 124 were more boxes in northern Taiwan (7) than in southern Taiwan (4) that had a mean 125 southwesterly wind speed exceeding 12 m s -1 . The 6-h accumulated rainfall from 3 hours before 126 to 3 hours after the event time, using rain gauge observations, shows that rainfall occurred 127 mostly over the southern and central mountain regions with maxima around 40-50 mm (Fig. 2b). 128 On the other hand, the second example at 0000 UTC 12 June 2012 (Fig. 2c) was designated as an 129 SWs event because there were more boxes in southern Taiwan (7) than in northern Taiwan (0) 130 that had a mean southwesterly wind speed exceeding 12 m s -1 , in addition to the overall 131 conditions that 7 boxes met the first criterion and 11 boxes met the second criterion. The axis of 132 strong southwesterly winds obviously extended from southern Taiwan toward the northeast at the 133 850-hPa level. This event produced heavy rainfall in the 6-h time period over the entire Island,134 especially in the mountain areas (Fig. 2d). The maximum rainfall reached about 200 mm in 135 several places. 136 In order to examine the relationship between the SW events and rainfall, we collected 40 137 years  of rainfall observations from the 28 Central Weather Bureau of Taiwan 138 (CWB) weather stations. We defined the 6-h rain intensity (hereafter, R; unit: mm/6h) as the 139 accumulated rainfall from 3 hours before to 3 hours after the event time. Heavy precipitation was 140 designated for a weather station when its R exceeds 20 mm/6h. The coverage percentage of such 141 heavy precipitation in Taiwan was defined as heavy precipitation percentage (hereafter, HPP; 142 unit: %), 143 where NHP is the number of stations in Taiwan which observed heavy precipitation, and N is the 145 total number of stations in Taiwan that had rain observation during the event. 146 Figure 3a shows the average occurrence percentage of SWs, SWn, and HPP, and the average 148 R of all 28 CWB stations in 40 years  from 1 April to 31 August using 11-d running 149 mean. The mean (μ) and standard deviation (σ) of each curve are shown in Table 1. It is clear that 150

Relationship between SWs and rainfall 147
SWs was highly correlated with R and HPP. They were all small before May 12, and 151 subsequently began to increase sharply. After reaching their maxima around June 7, all the three 152 8 curves started to decline until early July when the typhoon season began in Taiwan. On the other  153 hand, SWn exhibited a quite different pattern compared with the aforementioned three curves; it 154 had a higher/lower percentage than SWs and HPP in spring/summer. Furthermore, the peak of 155 SWn had a lag of ~11 d compared with that of SWs due to the fact that southwesterly flows 156 originated first in the south and then gradually propagated toward the north. Figure 3b shows the 157 same plot, but using 5-d running mean. The mean and standard deviation which were close to 158 those of the 11-d running mean are shown in Table 1. The curves in the 5-d running mean 159 basically present the similar trend as aforementioned, but with more details than those in the 11-d 160 running mean. They also clearly show that peaks of the SWs event, R, and HPP almost occurred 161 on the same date. Table 2 further shows high correlation coefficients between SWs and R 162 (0.86/0.81), and between SWs and HPP (0.88/0.83) for 11-d/5-d running means. However, 163 correlation coefficients between SWn and R and between SWn and HPP were all negative and 164 small no matter whether for 11-d or 5-d running means. It is therefore evident that rainfall in 165 Taiwan is closely related to southwesterly flow events in southern Taiwan, but not in northern 166 Taiwan. The reason for the high correlation is that, as documented in the literature shown in the 167 introduction (e.g., Chen et al. 2008;Ding and Chan 2005;Chien 2015), the southwesterly flows 168 usually developed ahead of mei-yu fronts and transported moisture-laden air to the Taiwan area. 169 Moisture flux convergence occurred and MCS developed over ocean southwest to Taiwan. When 170 the MCS moved overland, it was further enhanced by the lifting effect of complex terrain and 171 brought heavy rainfall in Taiwan. 172 According to the season definitions by Chen and Chen (2003) and Chien and Chiu (2019), 173 we examined the average 6-h rain intensity (R) and (HPP) for the all, nonSW, SW, SWs, and 174 SWn events during the entire warm season (Apr 1-August 31), spring (Apr 1-May 14), mei-yu 175 (May 15-June 15), and summer (June 16-August 31) from 1979 to 2018 (Fig. 4). The results 176 showed that for the entire warm season and the three different seasons, R and HPP of the SW 177 events were both higher than those of the nonSW events. Except in spring, the magnitudes of the 178 former were all about double those of the latter, suggesting that southwesterly flows can provide 179 favorable conditions for heavy rainfall in Taiwan, which is similar to the 11-year analysis 180 presented in Chien and Chiu (2019). When the SW events were further grouped into the SWs and 181 SWn events, it is interesting to see that R and HPP of the SWs events were both significantly 182 larger than those of the SWn events. R of the SWs events was about 4 times larger than that of 183 the SWn events, and HPP of the SWs events was even about 6 times larger. Furthermore, the 184 SWn events and the nonSW events had about the same R and HPP. We can therefore conclude 185 that during the warm season, the SWs events were the major periods of rain and heavy 186 precipitation in Taiwan. 187 The above analyses show that no matter whether in spring, mei-yu, or summer, SWs events 188 contributed the most to rainfall in Taiwan. It is therefore important to examine the relationship 189 between the SWs events and rainfall. Since the probability of SWs events was low in spring and 190 rainfall was greatly affected by typhoons besides the monsoonal flow in summer, we decided to 191 focus on studying the relationship between the SWs events and rainfall in the mei-yu season 192 when the SWs events played an important role in rainfall (see Figs. 3 and 4). 193 Figure 5 shows scatterplots of SWs occurrence percentage versus the mean 6-h rain intensity 194 (R, mm/6h) over Taiwan and HPP (%) averaged for the entire mei-yu season from 1979 to 2018. 195 When the mean SWs occurrence percentage was larger than 10% in a particular year, all the 196 mean R were larger than 3.0 mm/6h (Fig. 5a). When the mean percentage was lower than 10%, 197 the mean R was mostly smaller than 3.0 mm/6h; their correlation coefficient was about 0.58, and 198 it passed the 99% confidence level of the t-test. This result suggests that the SWs occurrence 199 percentage can have a positive impact on rainfall in Taiwan. The scatterplot of SWs occurrence 200 percentage versus HPP (Fig. 5b) shows basically the same result as Fig. 5a, with a similar 201 correlation coefficient of 0.57. It is thus concluded that for the mei-yu season in a particular year 202 that has more SWs events, there would be more rainfall and more heavy precipitation events. 203 Figure 6 shows the correlation coefficients between the mei-yu-seasonally averaged variables 204 and SWs occurrence percentage in 40 years . A positive/negative correlation 205 coefficient represents that when a particular year has more SWs events, the corresponding 206 variable is relatively larger/smaller in that year, and vice versa. indicates that stronger southwesterly winds at low levels (e.g., 850 hPa) over the SCS would 216 increase the occurrence percentage of the SWs events in Taiwan. Because the southwesterly 217 winds could help transport moisture toward the Taiwan area, a positive correlation coefficient 218 was found between the 850-hPa specific humidity and the SWs events over southeastern China, 219 Taiwan, and extending toward the northwestern Pacific. In the same areas, correlation 220 coefficients between the 500-hPa omega and the SWs events were mostly negative (Fig. 6d), 221 signifying that there would be stronger upward motion when more SWs events appeared in a 222 particular year. Combining Figs. 6c and 6d suggests that during an SWs event, rainfall would be 223 very likely to happen in Taiwan and HPP would be large. 224 The pattern of low-level geopotential height is an important factor that determines the 225 distribution of southwesterly flows near the Taiwan area (Chien and Chiu 2019). We therefore 226 further examined correlation coefficients between the 850-hPa geopotential height and the SWs 227 events (Fig. 6e), and found that negative correlation coefficients were located in the north and of 850-hPa geopotential height gradient was smaller in these areas, the SWs events would 237 increase owing to the geostrophic adjustment process. Because the gradient was negative, a 238 smaller gradient represented a larger southward-pointing pressure gradient in these areas. The 239 positive correlation coefficients extending from southeastern China to southern Japan suggest 240 that when the SWs events increased, the y-component of 850-hPa geopotential height gradient 241 would be larger (or smaller southward-pointing pressure gradient) in these areas because the 242 major pressure gradient had shifted to the south of Taiwan. As a matter of fact, the correlation 243 coefficient pattern in Fig. 6f was similar to that in Fig. 6a, except that the signs were reversed, 244 due to the geostrophic relationship. A mei-yu front was usually associated with high equivalent 245 potential temperature to its south (Ding 1992). Figure 6g shows that a long belt of positive 246 correlation coefficients between the 850-hPa equivalent potential temperature and the SWs 247 events extended from southeastern China, Taiwan, toward the Pacific. This indicated that when a 248 mei-yu front was located north of the Taiwan area, there was more chance for the SWs events. It 249 is noted that the correlation coefficient pattern was similar to that in Fig. 6c because equivalent 250 potential temperature is closely related to the moisture. The correlation coefficients between the 251 y-component of 850-hPa equivalent potential temperature gradient and the SWs events ( Fig. 6h) 252 show that a narrow belt of negative correlation coefficient was located over the ocean north of 253 Taiwan, extending from southeastern China to south of Japan. This corresponded to the region of 254 frontal activity in mei-yu seasons where the equivalent potential temperature gradient was large 255 negative. The negative correlation coefficients suggest that when a mei-yu front was located to 256 the north of Taiwan, chance of the SWs events was higher. 257

Composite analyses of SWs active and inactive years 258
In order to understand characteristics of the SWs events, we examined the seasonally-average 259 occurrence percentage of SWs events in the mei-yu season for each year from 1979 to 2018 ( and was also clearly present in the equivalent potential temperature gradient of the climatological 283 mean, as shown in Fig. 8b. In order to examine the amount and direction of total water vapor 284 transport in the southwesterly flows, we calculated the integrated water vapor transport (IVT) 285 according to Ralph et al. (2019). IVT is basically a vertically integrated water vapor flux from 286 1000 to 200 hPa. The IVT of the CM clearly shows that the primary moisture source in Taiwan 287 during a mei-yu season was southwesterly flows from the SCS (Fig. 8b). The southerly or 288 southeasterly flow produced by the Pacific subtropical high could also play a secondary role. 289 These moisture transports provided the major moisture supply of the mei-yu season and helped 290 to maintain large moisture and equivalent potential temperature gradients to the north of Taiwan, 291 corresponding to the mei-yu frontal activity region. This region, extending from southeastern 292 China, north of Taiwan, and to the south of Japan, favored the formation of upward motion 293 shown in 500-hPa omega (Fig. 8c). 294 Differences between the AYM and CM present a large area of negative geopotential height 295 anomaly extending from southeastern China to the south of Japan. This pressure anomaly which 296 was caused by the weaker Pacific subtropical high in AYM was associated with a cyclonic 297 circulation in the wind anomaly. To the south of the pressure anomaly, a low-level southwesterly 298 wind anomaly associated with positive moisture anomaly extended from the SCS to Taiwan (Fig.  299 9a). Further upstream over southern Indo-China Peninsula, moisture exhibited a negative 300 anomaly because it had been transported to the downstream regions, such as the SCS and the 301 Taiwan area. This is clearly shown in the IVT vector anomaly in Fig. 9b. As a result, equivalent 302 potential temperature also exhibited a large positive anomaly over the same areas. Consequently, 303 the equivalent potential temperature gradient to the north of Taiwan became more significant and 304 the mei-yu frontal activity increased in the active years. The mid-level vertical velocity (omega) 305 showed a negative anomaly (Fig. 9c) which corresponded to the frontal area over the northern 306 SCS, the Taiwan area, and toward south of Japan. 307 As for differences between the IYM and CM (Figs. 9d-f), they were nearly opposite to 308 those between the AYM and CM (Figs. 9a-c). At 850 hPa, the IYM showed a large positive 309 pressure anomaly over the northwestern Pacific due to the northwestward extension of stronger 310 Pacific subtropical high. The pressure anomaly resulted in the easterly to northeasterly wind 311 anomaly to the south of Taiwan and over the SCS (Fig. 9d). Without the southwesterly flow, 312 moisture could not be transported northward. Instead, it accumulated over the southern SCS and 313 the Indo-China Peninsula. Around the Taiwan area, there was a dry anomaly, associated with a 314 negative equivalent potential temperature anomaly (Fig. 9e). The equivalent potential 315 temperature gradient became smaller during the inactive years, such that the mei-yu frontal 316 activity was significantly reduced. It is also evident that the IVT anomaly was directed westward 317 from the relatively dry central Pacific to the SCS. All of these patterns were unfavorable to 318 rainfall in Taiwan, such that the 500-hPa omega exhibited a positive anomaly which was 319 associated with a downward motion (Fig. 9f). 320 In order to understand the process of establishing a southwesterly flow environment in 321 Taiwan, we examined the time evolution of the SWs events, R, and HPP in May and June for 322 CM, AYM, and IYM using the 40-year data (Fig. 10). Black curves in Figs. 10a-c are basically 323 the same as the black, red, and blue curves, respectively, in Fig. 3. During the first half of mei-yu 324 season (May 15-31), the occurrence percentage of the SWs events in AYM was not much 325 different to that in CM (Fig. 10a). However, this percentage in AYM sharply increased from 5% 326 in late May to 30% in mid-June. Although the occurrence percentage of the SWs events in CM 327 also increased during the same period, it was only a small rise from about 4% to 8%. After June 328 15, the occurrence percentage of the SWs events in AYM sharply dropped to almost zero in late 329 June. R and HPP (Figs. 10b, c) showed a similar pattern of difference between AYM and CM in 330 the second half of the mei-yu season, but the slope of the curve was rather gentler than those of 331 the occurrence percentage of the SWs events. From May 6 to 20, R and HPP of the AYM were 332 relatively larger than those of CM, but starting in the first half of the mei-yu season the 333 difference between AYM and CM became smaller. The fact of the latter, that AYM was close to 334 CM, implies that in the first half of the mei-yu season an active year was nothing special in terms 335 of the occurrence percentage of the SWs events, R, and HPP. Only until the second half of the 336 mei-yu season did the conditions favor the formation of the SWs events and heavy rainfall 337 appeared for an active year. For inactive years (IYM), the occurrence percentage of the SWs 338 events, R, and HPP were all very low during the mei-yu season, and they started to increase 339 slightly only after June 16, later than in the active years (AYM). It is thus clear that during an 340 inactive year, the atmosphere condition was unfavorable for the formation of the SWs events and 341 rainfall. 342 Hovmoller diagrams along a cross section AB (location shown in Fig. 9a)  an active year was caused by the increased pressure gradient around the Taiwan area. 357 Figure 11b shows the same diagrams as Fig. 11a, except for the equivalent potential 358 temperature gradient. A large negative of such gradient can be a good indicator of an active 359 region of the mei-yu front. The climatological mean shows that the equivalent potential 360 temperature gradient was large negative to the north of Taiwan (>1200 km) and remained almost 361 the same during the first half of the mei-yu season (Fig. 11b). The fact that the largest negative 362 gradient was found near point B was due to the front frequently forming in the north and moving 363 southward into the cross section from time to time. After June 1, the frontal active region started 364 to shift northward over time. AYM shows two minima of negative anomaly to the north of 365 Taiwan, one appeared early before the mei-yu season, the other appeared in late May to early 366 June. This finding means that the mei-yu front was more active in an active year of southwesterly 367 flow events because the frequently-formed front was associated with a low-pressure environment 368 in the north, which produced a larger pressure gradient around the Taiwan area and favored the 369 formation of the southwesterly flows. 370 In order to check whether there really were more fronts in an active year of the SWs events, 371 we examined daily weather maps at 0000 UTC issued by Japan Meteorological Agency. When a 372 front appeared in the box north to Taiwan (115°E-125°E; 25°N-28°N), that particular date was 373 counted as a frontal date. For active years in 2001, 2005, 2006, 2012, and 2017, the frontal days 374 were 12, 14, 12, 11, and 8, respectively, in 32 days of the mei-yu season (May 15 to June 15). On 375 the other hand, for inactive years in 1980, 1987, 1989, 1991, and 1992, the frontal days were 7, 376 7, 4, 0, and 7, respectively. This finding was consistent with those discussed in the last paragraph 377 and suggests that low pressure associated with the frequent frontal formation favors the 378 formation of southwesterly flows.  (Figs. 12a, b). The first one extended from Indo-China, through southern China, north of 384 Taiwan, east-northeastward to south of Japan. The second one originated from south of the 385 Pacific subtropical high, westward through the Philippines and the northern SCS, and northward 386 to the Taiwan area. Another path of large IVT was far over the Bay of Bengal, which more or 387 less connected to, and supplied moisture for, the first path in CM. In AYM, however, this third 388 path of moisture transport became an important moisture supplier in the mei-yu season (Fig.  389 12d). Before the mei-yu season, the first path of moisture transport was weaker, while the second 390 path was stronger in AYM (Fig. 12c) than in CM. After the onset of the mei-yu season (Fig. 12d), 391 the Pacific subtropical high weakened and the first path disappeared in AYM (compare Figs. 12c 392 and d). The third path strengthened and extended to the SCS, where it combined with the second 393 path and transported moisture northward to Taiwan. As a result of increased moisture in Taiwan, 394 the equivalent potential temperature increased. Consequently, the equivalent potential 395 temperature gradient increased to the north of Taiwan, which favored the mei-yu frontal activity. 396 In turn, the pressure gradient around the Taiwan area increased with a greater chance for 397 southwesterly flow formation, which would then further increased the moisture transport. In 398 IYM, the Pacific subtropical high strengthened and the first and second paths both became 399 stronger than in CM, but they circled around Taiwan such that not much moisture affected 400 Taiwan before the mei-yu season (Fig. 12e). After the onset of mei-yu season (Fig. 12f), there 401 was still not much moisture near the Taiwan area and the IVT pattern was actually very similar to 402 that of the climatological mean. 403 To illustrate more clearly the aforementioned formation process of the SWs events in an 404 active year, Figs. 13a, b present time evolution of several box-averaged variables in AYM from 405 May 1 to July 1. The boxes are shown in Fig. 13c with corresponding colors. The red curve in 406 Fig. 13b presents the same occurrence percentage of the SWs events as that in Fig. 10a for 407 comparison purpose. Since the y-component gradients of equivalent potential temperature and 408 geopotential height are both negative, they are multiplied by minus one for easy comparison. We 409 use the average 850-hPa geopotential height in a box to the east of Taiwan and the northern 410 Philippines (brown box in Fig. 13c) as an index of the strength of the Pacific subtropical high. 411 The brown curve in Fig. 13a shows that the Pacific subtropical high was weakening from early 412 May to June 6 in AYM. Figure 12d already shows that northward moisture transport over the 413 SCS and the Philippines was large. We therefore compute the averaged y-component of IVT in a 414 box over these regions (black box in Fig. 13c) to examine the moisture transport. The y-415 component of IVT (black curve in Fig. 13a) started to increase after May 8 and peaked 416 approximately on May 18, which corresponded to the process shown in Figs. 12c, d. As a result 417 of such moisture transport, the equivalent potential temperature (purple curve) around the Taiwan 418 area increased after May 21, and so did the minus y-component of equivalent potential 419 temperature gradient north of Taiwan (green curve in Fig. 13b). This environment favored the 420 formation of mei-yu fronts and low pressure to the north of Taiwan such that the minus y-421 component of pressure gradient (blue curve) around the Taiwan area increased with a greater 422 chance for southwesterly flow formation (red curve). The southwesterly flow then resulted in a 423 further increase of moisture transport (black curve in Fig. 13a) from June 1 to June 16. Lastly, 424 two additional notes are needed for the minus y-component of equivalent potential temperature 425 gradient (green curve). First, the peak around May 15 which was also found in R/HPP (Figs. 10b,  426 c) was mainly associated with the SWn events (green curve in Fig. 3a) that contributed to the 427 large moisture transport around northern Taiwan in early May, and was thus irrelevant to current 428 discussion. However, the moisture transport (black curve) and equivalent potential temperature 429 increase (purple curve) before May 15 might have contributed to its formation to some extent. 430 This peak and the second peak around May 28 corresponded to the two minima in AYM of Fig.  431 11b. Second, the drop of the minus y-component of equivalent potential temperature gradient 432 after June 1 was a result of northward shift of frontal activity (see CM in Fig. 11b). The low 433 pressure environment was still present to the north of Taiwan. 434 mei-yu seasons. The spatial distribution of correlation coefficients is presented in Fig. 14a for the 443 5-d period of May 12-16 as an example. It is clear that over the eastern part of the SCS and the 444 Philippines the correlation coefficients were all high (>0.5) and exceeded the 99% confidence 445 level, suggesting that we could use the y-component of IVT over these regions in the early mei-446 yu season (e.g., May 12-16) to predict rainfall in Taiwan for the entire mei-yu season. 447 Since the correlation coefficients over the SCS and the Philippines were relatively high, we 448 further took an average of the y-component of IVT first in a box over these regions (location 449 shown in Fig. 14a, and is the same as the black box in Fig. 13c) for a chosen 5-d period before 450 and during the early mei-yu season for each year. The correlation coefficient was then calculated 451 between the 5-d areal mean IVT and the seasonally-averaged occurrence percentage of the SWs 452 events/R/HPP for the 40 mei-yu seasons. The correlation coefficients are presented in Fig. 14b  453 for 11 5-d periods, including May 7-11, May 8-12, …, and May 17-21. The y-component of 454 IVT was highly correlated with the occurrence percentage of the SWs events, R, and HPP. For 455 the three periods including May 11-15, May 12-16, and May 13-17, the correlation coefficients 456 were all higher than 0.57. Even for the period before the mei-yu season, May 10-14, the 457 correlation coefficients were still higher than 0.49 which exceeded the 99% confidence level of 458 the t-test. This finding suggests that we could use this IVT variable in early mei-yu season as an 459 index to predict the occurrence percentage of the SWs events, rain, and heavy precipitation in 460 Taiwan for the entire mei-yu season. 461

Summary and conclusions 462
This paper presents a climatological study on southwesterly flows and heavy rainfall around 463 the Taiwan area during the mei-yu seasons from 1979 to 2018. The relationship between 464 southwesterly flows and rainfall in Taiwan was investigated using ERA-interim reanalysis data 465 from the ECMWF and rainfall data from the 28 CWB weather stations. Constituting a 466 continuous study of Chien and Chiu (2019), this paper divided the southwesterly flow events into 467 two groups according to the axis location of southwesterly flows. The event is defined as an 468 SWn/SWs event if the axis is located over northern/southern Taiwan. It was found that the 469 occurrence percentage of the SWs events was highly correlated with 6-h accumulated rainfall 470 and heavy precipitation, and the rainfall of the SWs events was significantly larger than that of 471 the SWn events. It is clear that during the warm season, the SWs events were the major periods 472 of rain and heavy precipitation in Taiwan. Correlation coefficient analyses show that stronger 473 southwesterly winds at low levels (e.g., 850 hPa) over the SCS would increase the occurrence 474 percentage of the SWs events in Taiwan. These airflows were associated with plenty of moisture 475 in the low level and upward motion in the middle level, indicating that rainfall and heavy 476 precipitation would very likely happen in Taiwan in the second half of the mei-yu season. This is owing to the fact that the Pacific subtropical high 488 weakened and the mei-yu fronts were more active in an active year and the frequently-formed 489 front was associated with a low-pressure environment in the north, producing a larger pressure 490 gradient around the Taiwan area and favoring the formation of southwesterly flows. More 491 specifically, Fig. 15 presents a schematic diagram to summarize the formation process of 492 southwesterly flows in the environment surrounding the Taiwan area. It all started with the 493 weakening of the Pacific subtropical high in the early mei-yu season. Moisture was then 494 transported along two major paths south of the Taiwan area. The first one extended from the Bay 495 of Bengal, Indo-China, the SCS, and to the Taiwan area. The second one originated from the 496 south of the Pacific subtropical high, westward through the Philippines and the northern SCS, 497 and northward to the Taiwan area. The airflows converged over the SCS and moved northward to 498 southern Taiwan. As a result of the moisture transport, the equivalent potential temperature 499 increased in the Taiwan area. Consequently, the equivalent potential temperature gradient 500 increased to the north of Taiwan, which favored mei-yu frontal activity, creating a cyclonic 501 circulation and low-pressure anomaly. The pressure gradient around the Taiwan area thus 502 increased, which supported the formation of southwesterly flows. The southwesterly flows then 503 helped transport more moisture toward the Taiwan area, resulting in heavy rainfall and increased 504 equivalent potential temperature in Taiwan. This kind of positive feedback can produce more 505 fronts, stronger southwesterly flows, and heavier rainfall in the mei-yu season. Based on the 506 moisture flux study, we suggest that the meridional-component of IVT over the eastern part of 507 the SCS and the Philippines in the early mei-yu season (e.g., May 12-17) can be used to predict 508 the occurrence percentage of the SWs events and heavy rainfall for the entire mei-yu season. 509

Acknowledgements 510
The data used in this study are obtained from the European Centre for Medium-Range contributions by different weather systems over Taiwan. J. Appl. Meteor. Climatol., 47, 588 2068-2080. 589 Zhang, Q., K.-H. Lau, Y.-H. Kuo, and S. J. Chen, 2003: A numerical study of a mesoscale 590 convective system over the Taiwan Strait. Mon. Wea. Rev., 131, 1150-1170 592 593    and June for the CM (black), AYM (red), and IYM (blue) using the 40-year data from 1979 to 746 2018. An 11-d running mean is taken for all curves. Black curves in (a-c) are the same as the 747 black, red, and blue curves, respectively, in Fig. 3a. 748 a b c 749 Figure 11: Hovmoller diagrams of 850-hPa (a) geopotential height gradient (gpm/100 km) (b) 750 equivalent potential temperature gradient (K/100 km) along line AB (see Fig. 9a for location). 751 Top: CM, middle: difference between AYM and CM, and bottom: difference between IYM and 752 CM. The ordinate is distance (km) from A to B, and the abscissa is time from May 1 to July 1. 753 An 11-d running mean is taken for both variables. A light blue line in left denotes the range of 754 Taiwan.