Invited Review Articles
日本で集中豪雨をもたらす線状降水帯と名付けられた準停滞線状降水システム
2020 年 98 巻 3 号 p. 485-509
詳細
2020 年 98 巻 3 号 p. 485-509
In Japan, localized heavy rainfall events producing accumulated three-hour precipitation amounts larger than 200 mm are often observed to cause severe landslides and floods. Such events are mainly brought by quasistationary band-shaped precipitation systems, named “senjo-kousuitai” in Japanese. Senjo-kousuitai is defined as a band-shaped heavy rainfall area with a length of 50–300 km and a width of 20 - 50 km, produced by successively formed and developed convective cells, lining up to organize multi-cell clusters, and passing or stagnating at almost the same place for a few hours. The formation processes of senjo-kousuitai are categorized mainly into two types; the broken line type in which convective cells simultaneously form on a quasi-stationary local front by the inflow of warm and humid air, and the back building type in which new convective cells successively forming on the upstream side of low-level winds linearly organize with pre-existing cells.
In this study, previous studies of band-shaped precipitation systems are reviewed, and the numerical reproducibility of senjo-kousuitai events and the favorable conditions for their occurrence are examined. In a case of Hiroshima heavy rainfall observed in western Japan on 20 August 2014, the reproduction of the senjo-kousuitai requires a horizontal resolution of at least 2 km, which is sufficient to roughly resolve the formation and development processes of convective cells, while a resolution of 250–500 m is necessary to accurately reproduce their inner core structures. The 2-km model quantitatively reproduced the Hiroshima case when initial conditions 10 hours before the event were used, but the predicted amounts of maximum accumulated precipitation were considerably reduced as the initial time became closer to the occurrence time of the senjo-kousuitai. This reduction was brought from the excessive inflow of low-level dry air that shifted occurrence areas of new multi-cell clusters.
Six favorable occurrence conditions of senjo-kousuitai events for their diagnostic forecasts were statistically constructed from environmental atmospheric fields in previous localized heavy rainfall events. Two conditions of (1) large water vapor flux amounts (> 150 g m−2 s−1) and (2) short distances to the level of free convection (< 1000 m) were chosen representatively for the low-level water vapor field that is judged based on 500-m height data. Four other favorable conditions are selected; (3) high relative humidity at midlevels (> 60 % at 500 hPa and 700 hPa), (4) large vertical shear estimated from the storm relative environmental helicity (> 100 m2 s−2), (5) synoptic-scale ascending areas (400 km mean field at 700 hPa), and (6) the exclusion of warm air advection frequently appearing at 700–850 hPa and inhibiting the development of convection (i.e., an equilibrium level > 3000 m).
In Japan, when localized heavy rainfall is observed, band-shaped rainfall areas with a length exceeding 50 km and accumulated three-hour precipitation amounts exceeding 200 mm (see Figs. 1, 4) are often captured in distributions of radar/raingauge-analyzed (R-A) precipitation amounts with a horizontal resolution of 1 km (Nagata 2011), which is hourly accumulated amounts produced every 30 minutes by the Japan Meteorological Agency (JMA). Previous studies (e.g., Watanabe and Ogura 1987; Kato and Goda 2001; Kato 2006; Kawabata et al. 2011; Takemi 2018; Tsuguti et al. 2019) showed that such areas were produced by band-shaped precipitation systems stagnating at almost the same place for a few hours. In Japan, such band-shaped rainfall areas have recently been given the name “senjo-kousuitai” in Japanese because of their characteristic shape; the term of “senjo” means band-shaped, and that of “kousuitai” means a rainfall area. In this review, quasi-stationary band-shaped precipitation systems that result in band-shaped rainfall areas captured in distributions of accumulated precipitation amounts are named “senjo-kousuitai”, since corresponding phenomena are rarely recognized outside of the Japanese islands due to the small number of observations. Making a simple translation of senjo-kousuitai into English (e.g., band-shaped rainfall area), the image on localized heavy rainfall with accumulated three-hour precipitation amounts exceeding 200 mm, usually recognized in Japan, would be vanished.
Before introducing the characteristics of the shape and appearance of senjo-kousuitai events observed in Japan, previous studies on band-shaped precipitation systems observed in the United States, Japan, and other countries are reviewed in this section. In addition to the review, the numerical reproducibility of senjo-kousuitai events for the horizontal model resolution and initial conditions is examined, and favorable occurrence conditions of the events are proposed in this paper.
Mesoscale convective systems (MCSs) have been classified from characteristics of their shapes and formation patterns. Bluestein and Jain (1985) showed four formation types of squall lines observed in Midwestern United States; broken line type, back building type, broken areal type, and embedded areal type. Parker and Johnson (2000) proposed three types of trailing-stratiform (TS), leading-stratiform (LS), and parallel-stratiform (PS) for the formation of squall lines occurred in central United States by focusing on the distributions of associated stratiform clouds. Schumacher and Johnson (2005) showed two patterns of MCS organization in the area east of the Rocky Mountains: One has a line with trailing convective cells and adjoining stratiform, and the other has a back building or quasi-stationary area of convection that produces a region of stratiform rain downstream. They indicated that the formation processes of the latter pattern strongly depend on storm-generated cold pools. However, cold pools are not necessary for the occurrence of senjo-kousuitai events because low-level humidity is very high in the vicinity of the Japanese islands (e.g., Kato and Goda 2001; Kato et al. 2003). Gallus et al. (2008) also classified MCSs observed in central United States into nine morphologies; three types of cellular convection (individual cells, clusters of cells, and broken squall lines), five types of linear systems (bow echoes (BE), TS, LS, PS, and lines with no stratiform rain (NS)), and nonlinear systems (NL). Zheng et al. (2013) classified MCSs observed in east central China into seven morphologies of NL and six linear modes; NS, TS, LS, PS, BE, and embedded lines. In Japan, Seko (2010) examined inner structures of band-shaped precipitation systems and classified them into three types; squall lines, back building type, and back and side-building type. It should be noted that the classifications made in previous studies were based on snapshots of radar observations, while senjo-kousuitai events can be recognized in distributions of accumulated precipitation amounts.
The formation mechanisms and structures of the band-shaped precipitation systems that result in senjokousuitai events have been examined using radar observations and numerical simulations in many case studies. Watanabe and Ogura (1987) emphasized the orographic effect, i.e., the surface convergence generated by the deflected flow over the land and undeflected flow over the sea was responsible for the formation of a senjo-kousuitai event observed in Shimane Prefecture (see Fig. 5) in 1983. The locations of other senjo-kousuitai events introduced below are also shown in Fig. 5. Yoshizaki et al. (2000) also examined an orographic rainband observed in the Nagasaki Peninsula in 1998, and showed that both a strong southwesterly jet at a height of 3–4 km and strong vertical wind shear in the lower troposphere were essential for its formation. In relation to the orographic effect, quasi-stationary convective bands observed outside of the Japan islands have been studied (e.g., Kirishbaum et al. 2007; Barret et al. 2015), but total accumulated precipitation amounts caused by these events were not large (< 100 mm). Kato (1998) studied a senjo-kousuitai event observed in Kagoshima Prefecture in 1993, and successfully reproduced precipitation systems with the formation of back building type by using a cloud-resolving model with a horizontal resolution of 2 km. He also showed that vertical wind shear was significant for the successive generation of convective cells upstream of the precipitation systems. Seko et al. (1999) and Kato and Goda (2001) also examined senjo-kousuitai events with the formation of back building type observed over the Kanto Plain in 1994 and in Niigata Prefecture in 1998, respectively. Kato (2006) clarified the structure of a band-shaped precipitation system resulting in a senjo-kousuitai event observed in Fukuoka Prefecture in 1999 (see Fig. 4b). This precipitation system had a hierarchical structure consisting of different scales; a quasi-stationary band-shaped precipitation system was organized by several MCSs, with each MCS consisting of a few convective cells that formed successively on its upstream side.
The numerical reproducibility of senjo-kousuitai events strongly depends on the horizontal model resolution and the accuracy of initial conditions. Kato and Aranami (2005) showed that a horizontal resolution of at least 1.5 km was necessary for the successful reproduction of a senjo-kousuitai event observed in Niigata and Fukushima Prefectures in 2004. They also indicated that a senjo-kousuitai event observed in Fukui Prefecture in 2004 could not be reproduced due to uncertainty in the initial low-level wind conditions. Oizumi et al. (2018) investigated the dependence of the reproducibility of a senjo-kousuitai event observed across Izu-Oshima Island in 2013 (see Fig. 1a) on the horizontal resolution (5 km, 2 km, 500 m, and 250 m) and planetary boundary schemes. An experiment using a 500-m grid spacing and the Deardorff scheme (Deardorff 1980) gave the best precipitation results according to the fractions skill score. Takemi (2018) performed a series of 167-m-resolution numerical experiments, and indicated that the representation of model terrains was critically important in simulating stationary MCSs and quantifying the resulting heavy rainfall in the northern part of Kyushu Island in 2017. Kato et al. (2003) tried to improve the reproducibility of a senjo-kousuitai event observed in Kagoshima Prefecture in 2001 by using modified initial conditions in which total precipitable water (TPW) data derived from the Tropical Rainfall Measuring Mission Microwave Imager was additionally assimilated by the four-dimensional variational (4DV) data assimilation technique. It was found, however, that a method that manually increased the humidity of the lower atmosphere to correspond with observations better reproduced the event. Kawabata (2011) applied a cloud-resolving nonhydrostatic 4DV data assimilation system to a senjo-kousuitai event observed over the Kanto Plain in 2005 (see Fig. 4c), and showed that additional assimilation of TPW derived from global positioning system data, surface and wind profiler data, and Doppler radial wind and radar-reflectivity data improved the reproducibility.
The environmental conditions that prevailed when senjo-kousuitai events occurred in Japan have been also clarified in many case studies. Kato and Goda (2001) and Kato (2006) showed that the continuous and abundant inflow of warm and humid air below a height of 1 km is necessary for the formation of convective cells that organize into the band-shaped precipitation systems that result in senjo-kousuitai events. They also showed that conditionally unstable atmospheric conditions, i.e., a low level of free convection (LFC) and a high equilibrium level (EL), are favorable for the formation of senjo-kousuitai events. Kato (2006) examined the tops of cumulonimbus clouds simulated by a numerical model with a horizontal resolution of 2 km, and noted that the inflow of midlevel dry air inhibited the further development of convective cells. Seko (2010) pointed out that suitable vertical wind shear is necessary for the organization of the band-shaped precipitation systems that cause senjo-kousuitai events. The characteristics of the favorable atmospheric conditions required for the formation of senjo-kousuitai events have also been statistically ascertained in cases of warm-season quasi-stationary convective clusters observed in Japan (Unuma and Tatsumi 2016a, b) and in cases of quasi-linear convective systems in the United States (e.g., Schumacher and Johnson 2008; Peters and Schumacher 2015).
In this review paper, first of all, examples of specific senjo-kousuitai events are introduced and the statistical characteristics of senjo-kousuitai events are studied based on distributions of accumulated three-hour R-A precipitation amounts by following the method of Tsuguti and Kato (2014, hereafter abbreviated to TK14), and the scale of senjo-kousuitai is then roughly defined. These are described in Section 2. In Section 3, the formation types of senjo-kousuitai are introduced and the detailed processes of organization for a heavy rainfall event observed in Hiroshima on 20 August 2014 are explained (see Fig. 1b). The Hiroshima event is also used to examine the numerical reproducibility of senjo-kousuitai by focusing on the horizontal model resolution and initial conditions in Section 4. Favorable occurrence conditions of senjo-kousuitai, operationally used in diagnostic forecasts by the JMA from 2016, are introduced in Section 5. The last section includes a summary and issues for consideration in future studies.
Shape characteristics of senjo-kousuitai events found in distributions of accumulated three-hour R-A precipitation amounts are introduced using two specific cases observed across Izu-Oshima Island on 16 October 2013 (Fig. 1a) and in Hiroshima on 20 August 2014 (Fig. 1b). In the Izu-Oshima event, Typhoon Wipha (T1326) was approaching the Kanto Plain, and a stationary front is analyzed on the northeastern side of T1326 on the surface weather map. The senjo-kousuitai occurred with extending from the southwestern edge of the stationary front to the Kanto Plain. The maximum three-hour precipitation amount is 333 mm, and the length and width of the area with three-hour precipitation amounts exceeding 100 mm are about 200 km and 30 km, respectively. In the Hiroshima event, the senjo-kousuitai occurred 200–300 km south of a stationary front located over the Sea of Japan. Such an occurrence location of heavy rainfall is often found in western Japan during the rainy season, which is called the Baiu season (e.g., Kato 1998; Yoshizaki et al. 2000). The maximum three-hour precipitation amount is 232 mm, and the length and width of the area with three-hour precipitation amounts exceeding 100 mm are about 80 km and 20 km, respectively. Rainfall areas are localized around the senjo-kousuitai in the Hiroshima event, while moderate intense rainfall areas extend widely in the Izu-Oshima event.
Another case, which cannot be easily judged as a senjo-kousuitai event from distributions of accumulated three-hour R-A precipitation amounts, is introduced in the following. This event occurred about 100 km south of the Baiu front with a low pressure system in Kumamoto Prefecture on 20–21 June 2016 (Fig. 2). The Baiu front is a quasi-stationary front extending in the east-west direction that usually appears around the Japanese islands during the Baiu season. No band-shaped area can be recognized in distributions of three-hour precipitation amounts exceeding 100 mm; however, two band-shaped areas with amounts exceeding 200 mm can be picked up — the maximum precipitation amounts within these areas are 255 mm and 246 mm. This means that two quasi-stationary band-shaped precipitation systems formed adjacent to each other at almost the same time. In this case, an hourly accumulated precipitation amount of 150 mm, the third largest in the JMA's historical records, was observed at Kousa, associated with the southern senjo-kousuitai.
Examples of senjo-kousuitai events for cases of (a) direct precipitation associated with typhoons and (b) a stationary front. The left-hand panels are the surface weather maps at (a) 0300 JST 16 October 2013 and (b) 0300 JST 20 August 2014. The right-hand panels are the distributions of accumulated three-hour R-A precipitation amounts at (a) 0500 JST 16 October 2013 and (b) 0400 JST 20 August 2014. The values in the right-hand panels are the maximum accumulated three-hour precipitation amounts. Red crosses denote the positions corresponding to the hodographs shown in Fig. 15. The solid black rectangle in (b) marks the model domains of 5kmNHM, 2kmNHM, 1kmNHM and 500mNHM, and the dashed black box marks that of 250mNHM.
Same as Fig. 1, but for an example of a senjo-kousuitai event with two band-shaped precipitation systems. The left-hand panel is the surface weather map at 2100 JST 20 June 2016, and the right-hand panel is the distribution of accumulated three-hour R-A precipitation amounts at 0100 JST 21 June 2016.
The specific senjo-kousuitai events introduced above accompanied band-shaped areas of heavy rainfall with a length of 50 - 300 km and a width of 20–50 km, as found in the distribution of accumulated three-hour R-A precipitation amounts, which is notable characteristics in shape.
2.2 Appearance characteristicsBefore making a definition of senjo-kousuitai, senjo-kousuitai events are objectively extracted and their appearance characteristics are statistically studied following the method of TK14. The extraction of senjo-kousuitai events is based on accumulated three-hour R-A precipitation amounts with a horizontal resolution of 5 km, covering the Japanese Islands, in the 1989–2015 warm seasons (April–November), while TK14 treated the 1995–2009 warm seasons. In addition to the statistical period, the procedure for extracting senjo-kousuitai events and a threshold value for aspect ratios of rainfall areas are modified from TK14, and a new exclusion condition for small rainfall areas is also adopted in this study. The other procedures used are the same as those in TK14.
Figure 3 shows the procedures for extracting heavy rainfall and senjo-kousuitai events that are used in this study. In the first step in the extraction procedures of heavy rainfall events, for previous accumulated 24-hour precipitation amounts, heavy rainfall points within top 50 in the amounts that exceed 12 % of the annual mean amount in the 15 warm seasons of 1995–2009 are extracted at every 5 km grid over the land at hourly intervals. The period of 1995–2009 warm seasons is adopted to hold consistent with the results of TK14. In the second step, grid points with previous accumulated three-hour precipitation amounts exceeding 130 mm are extracted from the grid points obtained in the first step. In the third step, extracted points that lay within a time-frame of 24 hours and a distance of 150 km are judged as belonging to the same heavy rainfall event, and the grid point with the maximum accumulated precipitation amount for the previous 24 hours and the time corresponding to the maximum 3-hour precipitation amount are then set to represent the location and time of the occurrence of the event, respectively.
Extraction procedures of heavy rainfall events and senjo-kousuitai events, modified from Fig. 2 of Tsuguti and Kato (2014). 3-h PA means an area of accumulated three-hour precipitation amounts.
In this study, senjo-kousuitai events are automatically extracted using areas with accumulated three-hour precipitation amounts exceeding 50 mm and aspect ratios exceeding 2.0, while TK14 manually extracted areas with aspect ratios exceeding 3.0. It should be noted that the extracted numbers of senjo-kousuitai events obtained in this study and those extracted in TK14 are almost the same for the 15 warm seasons of 1995–2009, which might be due to the manual judgement used in TK14. Furthermore, a new condition, not used in TK14, is adopted in this study; events in which the length of areas with accumulated three-hour precipitation amounts exceeding 30 mm is less than 50 km are excluded. Those events could have been caused by isolated multi-cell clusters consisting of a few convective cells.
The extracted heavy rainfall and senjo-kousuitai events are categorized using the synoptic fields depicted on weather maps. As described in TK14, the following five synoptic fields are also represented for each event; low pressure (LP), cold front (CF), stationary front (SF), direct precipitation associated with tropical cyclones (DTC), and indirect precipitation associated with tropical cyclones (ITC). The locations of typhoons are obtained from the JMA's best-track data and those of the other disturbances are manually judged using a weather map for the time corresponding to, or just before, the occurrence of the maximum three-hour precipitation amount. The distance between the disturbance and the representative point of rainfall events satisfies the following conditions; within 500 km (within 200 km for CF), and more than 500 km but within 1500 km for ITC (see Table 1). The synoptic field with the closest disturbance to the representative point of each rainfall event is chosen, but this categorization always gives priority to DTC and ITC against the other disturbances. Examples of senjo-kousuitai events categorized into five synoptic fields are shown in Fig. 4a (LP), Fig. 4b (CF), Fig. 1b (SF), Fig. 1a (DTC), and Fig. 4c (ITC). No disturbance cases, such as shown in Fig. 4d, are additionally categorized. In a sample LP event on 27 October 1999, an hourly accumulated precipitation amount of 153 mm, the highest in the JMA's records, was observed at Katori on the Kanto Plain.
Examples of senjo-kousuitai events for cases of (a) low pressure, (b) cold front, (c) indirect precipitation associated with typhoons, and (d) no disturbance. The left-hand panels are surface weather maps at (a) 2100 JST 27 October 1999, (b) 0900 JST 29 June 1999, (c) 2100 JST, September 2005, and (d) 0900 JST 09 August 2013. The right-hand panels are distributions of accumulated three-hour R-A precipitation amounts at (a) 2100 JST 27 October 1999, (b) 1000 JST 29 June 1999, (c) 2300 JST 04 September 2005, and (d) 1200 JST 09 August 2013. Values in the right-hand panels are the maximum accumulated three-hour precipitation amounts.
Table 1 shows the number of heavy rainfall and senjo-kousuitai events detected in the Japanese islands during the 1989–2015 warm seasons, categorized by the synoptic fields. The results of this categorization are based on Imamura (2018) in which the same method as that in Tsuguti and Kato (2014) was used, but the statistical period (i.e., the number of extracted events) is different. The total number of heavy rainfall events is 715, and among these, about half are associated with tropical cyclones (DTC + ITC); in particular, about one-third of the events are classified into DTC. Senjo-kousuitai events are also extracted at a rate of about 50 % from heavy rainfall events. The lowest rate of senjo-kousuitai extraction is for DTC (∼33 %), which indicates that intense precipitation systems usually move associated with tropical cyclones and do not stagnate at almost the same place. In contrast, the rates for CF (∼72 %) and SF (∼68 %) exceed two-thirds, and the number of 108 for SF is the largest. This high rate for SF is notable in Kyushu Island during the Baiu season (not shown). Kato (2005) showed statistically that quasi-stationary band-shaped precipitation systems often appear on the western side of Kyushu Island. These appearance characteristics in rates are consistent with the results of TK14, and don't change a lot even when other aspect ratios of rainfall areas (1.5, 2.5, 3.0) and different threshold values of accumulated three-hour precipitation amounts (30 mm, 70 mm) are adopted for the extraction of senjokousuitai events.
The orientations of band-shaped areas with accumulated three-hour precipitation amounts exceeding 50 mm in the detected senjo-kousuitai events are divided into four types; north-south (N), northeast-southwest (NE), east-west (E), and southeast-northwest (SE). The NE orientation is the most apparent (∼44 %), while the SE orientation is the least (∼ 7 %). The rates of N and E orientations are 18 % and 31 %, respectively. The distribution of the orientation of the detected senjo-kousuitai, shown in Fig. 5, has regional characteristics. The dominant orientations are E and NE on the northwestern side of Kyushu Island; NE on the ocean side of Shikoku Island and in the Kii Peninsula; E on the southern side of Kyushu Island, in Shimane Prefecture and in the Kinki region, as well as on the Japan Sea side of the Tohoku region; and N in the Nobi Plain.
Distribution of predominant orientations of the senjo-kousuitai detected in the 1989–2015 warm seasons (April–November). Prefecture names are in italics.
In senjo-kousuitai events, the inflow of low-level warm and humid air into the Japanese islands is often found along the edge of Pacific Ocean high pressure systems (e.g., Kato et al. 2003; Kato 2018). This inflow can produce the warm advection associated with the vertically veering wind shear. As a result, the movements of convective cells are shifted clockwise compared to the direction of the low-level warm and humid inflow, because they are roughly determined by wind vectors between low and middle levels (Kato 2006). The dominant direction of the inflow along the edge of Pacific Ocean high pressure systems is between south and southwest, which could bring the dominant senjo-kousuitai orientations of E and NE on the Pacific Ocean side of western Japan and the western side of Kyushu Island. It should be noted that the inflow of low-level warm and humid air into the Japan Sea side is usually blocked by high mountains of the Japanese islands, which have heights exceeding 1 km. This inflow, therefore, occurs over the Sea of Japan only when it passes through the Tsushima Strait and its direction is close to southwest, which could bring the dominant orientation of E on the Japan Sea side of the Japanese islands in senjo-kousuitai events. Moreover, some of senjo-kousuitai events are influenced by topography, especially those observed in the Nobi Plain. Kato (2002) and Takasaki et al. (2019) performed numerical sensitivity experiments, and showed that the southerly low-level inflow is blocked or influenced by topography, such as mountains of the Kii Peninsula, resulting in the winds having an easterly component. This change of wind directions brings the dominant orientation of N in the Nobi Plain in senjo-kousuitai events.
2.3 Scale definitionIn the previous subsections, senjo-kousuitai events are introduced for several cases and are extracted using distributions of accumulated three-hour precipitation amounts. This extraction is based on the fact that senjo-kousuitai events are caused by band-shaped precipitation systems stagnating at almost the same place for a few hours, which means that the resulting areas of accumulated rainfall also become band-shaped. It should be noted that accumulated rainfall areas never become band-shaped when band-shaped precipitation systems found in snapshots of radar observations travel at a certain level of speed as ordinary squall lines.
The length and width of areas with accumulated three-hour precipitation amounts exceeding 50 mm in the detected senjo-kousuitai events in this study varied widely within the ranges 30–500 km and 10–200 km, respectively. Some of these areas include precipitation produced by other systems, and consequently moderate intense rainfall (10–20 mm (3h)−1) extended further. Since the degree of extension of these rainfall areas varied a lot for different events, the scale of senjo-kousuitai cannot easily be defined; however, typical events (Figs. 1, 4) indicate that specific senjo-kousuitai can be roughly defined as a band-shaped heavy rainfall area with a length of 50–300 km and a width of 20–50 km in distributions of accumulated precipitation amounts, such as R-A precipitation amounts.
Following the classification of Bluestein and Jain (1985), the formation processes of senjo-kousuitai can be categorized mainly into two types. One is the broken line type, shown in Fig. 6a, in which convective cells simultaneously form on a quasi-stationary local front by the inflow of warm and humid air. The Izu-Oshima event (Fig. 1a) is a typical case of this type. In this event, cold outflow produced by precipitation prior to the event over the Kanto Plain caused and intensified a local front extending from the Kanto Plain that corresponds to the stationary front analyzed on the surface weather map. The other type of the formation is the back building type, in which new convective cells successively forming on the upstream side of low-level winds linearly organize with preexisting cells, as shown in Fig. 6b. These formation processes have been recognized by analyzing time-series of radar observations (e.g., Kato 1998; Kato and Goda 2001; Takasaki et al. 2019). The Hiroshima event (Fig. 1b) is a typical case of the back building type, and it is used to concretely explain the formation processes and structures of senjo-kousuitai in the following.
Major formation processes of senjo-kousuitai; (a) broken line type and (b) back building type. Postscript information is added in Fig. 1 of Bluestein and Jain (1985). Dashed blue lines in (a) and blue arrows in (b) show quasistationary local fronts and movements of convective cells, respectively. The large orange arrow in (a) and large green arrow in (b) denote representative winds at low levels.
Figure 7a shows the time-series of distributions of radar-estimated precipitation intensity between 2340 JST (= UTC + 9 hours) 19 August and 0040 JST 20 August 2014. The vertical cross section at 0040 JST 20 August (Fig. 7c) shows that multi-cell clusters A and B, respectively consisting of several convective cells ➀–➃ and ➄–➈, organize into a band-shaped precipitation system resulting in a senjo-kousuitai event. Hereafter this band-shaped precipitation system is simply called “senjo-kousuitai”. The developed convective cells, i.e. cumulonimbus clouds, reach the height of the tropopause (∼ 16 km). The convective cell ➄ occurring around 2340 JST 19 August travels northeastward, and organizes the multi-cell cluster B with cells ➅–➈ successively forming on its southwestern side at intervals of about 10 minutes. This process, in which new convective cells that successively form on the upstream side of low-level winds linearly organize with pre-existing cells, clearly explains the formation of back building type. Moreover, the senjo-kousuitai has a hierarchical structure with different scales (Fig. 7b); a quasi-stationary band-shaped precipitation system is organized by multicell clusters, with each multi-cell cluster consisting of several convective cells that form successively on its upstream side, as shown in Kato (2006).
(a) Distributions of radar-estimated precipitation intensity (mm h−1) between 2340 JST 19 August and 0040 JST 20 August 2014 at 10-min time intervals. Yellow arrows denote the movements of nine convective cells. Brown arrows represent the wind directions at low and middle levels. (b) Schematic diagram of formation mechanisms and structure of a multi-cell cluster and senjo-kousuitai. (c) Vertical cross section on the black line in (a) at 0040 JST 20 August. Multi-cell clusters A and B consist of convective cells ➀–➃ and ➄–➈, respectively.
As mentioned in the previous section, the occurrence of senjo-kousuitai events requires the stagnation of band-shaped precipitation systems. This stagnation process is also explained for the Hiroshima event by using time-series of distributions of radar-estimated precipitation intensity. The time series with an interval of 30 minutes (Fig. 8a) show that new multi-cell clusters A–G successively form around the boundary between Yamaguchi and Hiroshima Prefectures, and organize and maintain the senjo-kousuitai for about four hours while traveling northeastward. Consequently, the senjo-kousuitai extends from southwest to northeast with a length of about 100 km. This indicates that the formation process of this senjo-kousuitai is also the back building type based on multi-cell clusters that successively form on the low-level upstream side of the traveling direction of the clusters. These formation processes that have a hierarchical structure consisting of convective cells, multi-cell clusters, and senjo-kousuitai (Fig. 7b) are also analyzed in the Kumamoto event (Fig. 2).
(a) Same as Fig. 7a, but between 2345 JST 19 August and 0315 JST 20 August 2014 at 30-min time intervals. Points where the blue lines cross denote the location of Miiri observation station. (b) Time series of 10-min accumulated (blue bars) and total (black line) precipitation amounts between 2100 JST 19 August and 0500 JST 20 August 2014, observed at Miiri.
Figure 8b shows the time-series of precipitation amounts observed at Miiri (the point where the blue lines cross in Fig. 8a), where a large accumulated precipitation amount exceeding 200 mm was recorded over several hours and caused a landslide that killed several people. Five multi-cell clusters (B, C, D, F, and G) successively pass over the area at intervals of about 30 minutes to produce accumulated 10-minute precipitation amounts of 10–20 mm for a continuous period of over two hours. Consequently, the hourly accumulated precipitation amount reaches 101 mm between 0300 JST and 0400 JST on 20 August. As shown above, heavy rainfall with accumulated precipitation amounts exceeding 200 mm was caused by the stagnation of senjo-kousuitai that was organized with successively forming multi-cell clusters.
The dependence of the numerical reproducibility of senjo-kousuitai events on the horizontal model resolution and initial conditions are also investigated for the Hiroshima event (Fig. 1b). The model used in this study is the JMA nonhydrostatic model (NHM) (Saito et al. 2006). Horizontal resolutions of 5 km, 2 km, 1 km, 500 m, and 250 m were tested, and the experiments were denoted 5kmNHM, 2kmNHM, 1kmNHM, 500mNHM, and 250mNHM, respectively. Each NHM employs 50 vertical levels with variable thicknesses, from 40 m near the surface to 886 m at the top of the domain (a height of 21.8 km) and the same dynamical and physical processes, except those for precipitation. For precipitation, the NHMs uses a bulk-type cloud microphysics scheme (Murakami 1990), and 5kmNHM additionally uses the Kain-Fritsch convection parameterization scheme (Kain 2004). The bulktype microphysics scheme predicts both the mixing ratio and number density of ice hydrometeors (i.e., cloud ice, snow, and graupel) but only the mixing ratio of liquid hydrometeors (i.e., cloud water and rain). The turbulence closure scheme of Mellor-Yamada- Nakanishi-Niino level-3 (Nakanishi and Niino 2006) is used. The other settings, including numerical diffusion, are the same for all the NHMs. The other model specifications are detailed in Saito et al. (2006, 2007).
The models all cover the same area of 1000 km × 800 km, as marked by the solid black rectangle in Fig. 1b, except for 250mNHM, which has an area of 412 km × 330 km (the dashed black rectangle in Fig. 1b). The initial and linearly interpolated boundary conditions of the NHMs are obtained from hourly available JMA local analyses with a horizontal resolution of 5 km, in which a three-dimensional variational data assimilation technique is adopted (Japan Meteorological Agency 2013); however, the boundary conditions of 250mNHM are obtained from forecasts of 2kmNHM that successfully reproduced the senjo-kousuitai in the Hiroshima event, as described in the next subsection. The sea surface temperature is obtained from the Merged Satellite and In-situ Data Global Daily Sea Surface Temperature product produced daily by the JMA (Japan Meteorological Agency 2013). The model terrains of NHM5km, NHM2km and NHM1km are generated by interpolation of the global 30 arc-second elevation dataset, which has a horizontal grid spacing of approximately 1 km, while the terrains of NHM500m and NHM250m are interpolated from a digital elevation model with a 50-m grid spacing provided by the Geospatial Information Authority of Japan.
4.2 Dependency of reproducibility on horizontal resolutionsIn order to objectively reproduce individual convective cells organizing into multi-cell clusters, it is necessary that updrafts, and downdrafts in the inner-core structures of the cells are separately resolved through microphysics processes by model grids. Bryan et al. (2003) conducted ideal numerical simulations of squall lines with grid spacings between 1 km and 125 m, and revealed that simulations with a 1-km grid spacing do not produce equivalent squall line structure and evolution as compared to higher resolution simulations. On the other hand, previous studies in real cases (e.g., Miyamoto et al. 2013; Ito et al. 2017) have shown that a horizontal resolution of at least 2 km is necessary to resolve developed convective cells. In this study, therefore, the dependency of numerical reproducibility on the horizontal resolution is examined for the senjo-kousuitai in the Hiroshima event by using 5kmNHM, 2kmNHM, 1kmNHM, 500mNHM and 250mNHM.
Figure 9 shows the distributions of accumulated three-hour precipitation amounts in the analysis (R-A precipitation) and the model forecasts for the initial time of 1800 JST 19 August 2014. Each NHM, except 5kmNHM, successfully reproduces the band-shaped rainfall distribution extending from southeast to northwest and could also quantitatively predict the maximum precipitation amount (∼ 232 mm). On the other hand, the 5kmNHM predicts several rainfall areas with moderate precipitation amounts of 10–30 mm along the observed areas of the senjo-kousuitai, and the predicted maximum precipitation amount is only 32 mm, although predicted near-surface winds on the upstream side are almost the same in all the NHMs. An additional experiment of 5kmNHM without the Kain-Fritsch convection parameterization scheme was performed, and the band-shaped rainfall area was well reproduced; however, this area was shifted by about 50 km to the northeast, and the predicted maximum precipitation amount was 122 mm (3 h)−1 (not shown). This indicates that the explicit treatment of precipitation processes is important in the prediction of senjo-kousuitai events, even if the horizontal resolution is not enough to resolve convective cells.
(a) Distribution of accumulated three-hour R-A precipitation amounts (mm) between 0100 JST and 0400 JST 20 August 2014. Same as (a), but for forecasts of (b) 5kmNHM, (c) 2kmNHM, (d) 1kmNHM, (e) 500mNHM, and (f) 250mNHM forecasts. Arrows denotes near-surface winds. Initial time of each model is 1800 JST 20 August.
The differences of structures in simulated convective cells are examined using the results of 2kmNHM and 250mNHM. Figures 10a and 10b show the distributions of hydrometeors (total mixing ratios of snow, graupel, and rain) and horizontal winds at a height of 2 km at 01 JST 20 August 2014 in 2kmNHM and 250mNHM, respectively. The 250mNHM (Fig. 10b) well reproduces two multi-cell clusters, A′ and B′, corresponding to clusters A and B in the radar observations (Fig. 7a), while several larger convective cells are simulated in 2kmNHM (Fig. 10a). Moreover, the 250mNHM reproduced almost the same back building type formation processes of convective cells as those found in the radar observations (Fig. 7a). The differences in the simulated vertical structures are also examined (Figs. 10c, d). The structure containing several convective cells within multi-cell clusters A′ and B′ are well reproduced by 250mNHM (Fig. 10d), while the multi-cell structures consisting of a smaller number of convective cells (2–3) than seen in the radar observations are found in 2kmNHM (Fig. 10c). The Hiroshima senjo-kousuitai event can, therefore, be reproduced using an even coarser horizontal resolution of 2 km; however, 2 km is not enough for the examination of the inner-core structures of the senjo-kousuitai. It should be noted that the results of 500mNHM were close to those of 250mNHM, and the 1kmNHM had intermediate characteristics between the results of 2kmNHM and 250mNHM (not shown).
Distributions of hydrometeors (total mixing ratios of snow, graupel, and rain (g kg−1)) and horizontal winds (arrows) at a height of 2 km at 0100 JST 20 August 2014, simulated by (a) 2kmNHM and (b) 250kmNHM for the initial time of 1800 JST 19 August. Multi-cell clusters A′ and B′ in (b) correspond to the observed clusters (Fig. 5c). (c) and (d) Vertical cross sections of hydrometeors on the broken blue lines in (a) and (b), respectively. Arrows denote wind vectors projected on these cross sections.
The use of initial conditions closer to the occurrence time of phenomena usually produces better forecasts; however, the forecasts for the initial time of 1800 JST 19 August 2014, 10 hours before the occurrence of the event, are the best in the Hiroshima event. In this subsection, the dependency of the numerical reproducibility of the senjo-kousuitai is examined on initial conditions that are produced from hourly available JMA local analyses at different initial times. Figure 11 shows the distributions of accumulated three-hour precipitation amounts predicted by 2kmNHM for the initial times of 1900 JST, 2000 JST, 2100 JST and 2200 JST 19 August. The forecasts for the first three initial times show the successful reproduction of the senjo-kousuitai, although the maximum precipitation amounts are reduced by about half. The initial conditions closest to the occurrence time (Fig. 11d) fails to predict the senjo-kousuitai, and the maximum precipitation amount is only 30 mm.
Same as Fig. 9c, but for the initial times of (a) 1900 JST, (b) 2000 JST, (c) 2100 JST, and (d) 2200 JST 19 August 2014.
The reason why the forecasts become worse as the initial time becomes closer to the occurrence time is examined focusing on the inflow amounts of low-level water vapor. The reduction of predicted maximum precipitation amounts could be brought from that of the inflow amounts of low-level water vapor. Figure 12 shows the distributions of water vapor flux amounts (FLWV = ρqν |ν|, where ρ is the density, qν is the mixing ratio of water vapor, and ν is the horizontal wind vector) at a height of 500 m at 0300 JST 20 August 2014, depicted from the JMA local analysis and 2kmNHM forecasts for the initial times of 1800 JST, 1900 JST, 2000 JST, 2100 JST and 2200 JST 19 August. The distribution corresponding to the initial time of 1800 JST, not the JMA local analysis for the occurrence time of the event at 0300 JST 20 August, could be the most accurate, possibly because the corresponding forecasts quantitatively reproduce the senjo-kousuitai. This means that the analysis for the occurrence time of the event is uncertain in this case. The FLWV values for the initial times of 1900 JST and 2000 JST are larger than those for the initial time of 1800 JST although the maximum precipitation amounts are reduced by about half. Larger inflow amounts of low-level water vapor result in wider rainfall areas with moderate precipitation amounts of 10–30 mm (Figs. 11a, b). On the other hand, intense rainfall predicted for the initial time of 1800 JST is more concentrated in a band-shaped area extending from southwest to northeast. This concentration could be strongly influenced by the low-level wind field. After the initial time of 2100 JST 19 August, the FLWV is more reduced as it becomes closer to the occurrence time of the senjo-kousuitai (Figs. 12e, f), which causes the further reduction of precipitation amounts (Figs. 11c, d).
(a) Distribution of water vapor flux amounts (g m−2 s−1) and horizontal winds at a height of 500 m, depicted from JMA mesoscale analysis at 0300 JST 20 August 2014. Same as (a), but for 2kmNHM forecasts for the initial times of (b) 1800 JST, (c) 1900 JST, (d) 2000 JST, (e) 2100 JST, and (f) 2200 JST 19 August.
The influence of the low-level wind field on the concentration of intense rainfall areas is examined by comparing the forecasts for the initial times between 1800 JST and 1900 JST 19 August. Figures 13a and 13d show the distributions of relative humidity (RH) and horizontal winds at a height of 500 m predicted by 2kmNHM for the initial times of 1800 JST and 1900 JST, respectively. In both forecasts, the inflow of relatively dry air is found from the southwest on the southwestern side of the senjo-kousuitai, and its speed is faster in the forecast for the initial time of 1900 JST. This faster inflow causes large temperature drops exceeding 1°C around the occurrence areas of multi-cell clusters due to the evaporation of raindrops at a height of 500 m between 0100 JST and 0300 JST 20 August. These areas, marked by the blue circles in Figs. 13e and 13f, correspond to cold pools that were not found in the observations. These cold pools shift the occurrence areas of new multi-cell clusters eastward, and consequently rainfall areas with moderate precipitation amounts become more widespread (Fig. 11a). In contrast, in the forecast for the initial time of 1800 JST, notable temperature drops are not found around the occurrence areas of multi-cell clusters until 0300 JST 20 August (Figs. 13b, c). This indicates that slight differences in the low-level wind field significantly influence the concentration of intense rainfall areas, resulting in the formation of senjo-kousuitai.
(a) Distributions of relative humidity (%) and horizontal winds at a height of 500 m at 0200 JST 20 August 2014, simulated by 2kmNHM for the initial time of 1800 JST 19 August. The bold blue arrow and curve represent the relatively dry inflow from the western side and the head of the senjo-kousuitai, respectively. Same as (a), but for the temperature (°C) at (b) 0100 JST and (c) 0300 JST 20 August. FT denotes the forecast hour. (d), (e), and (f) same as (a), (b), and (c) but for forecasts for the initial time of 1900 JST 19 August.
The previous section shows that numerical models with a horizontal resolution of 2 km can successfully reproduce senjo-kousuitai events, such as the Hiroshima event. However, senjo-kousuitai events often fails to be forecasted even if high resolution models can be used. This is because initial conditions and physical processes adopted in numerical models are not complete, as well as the horizontal resolution being inadequate for the perfect expression of individual convective cells. Moreover, the forecast period of high resolution models used in operational weather centers is short; for example, it is only 10 hours for the JMA local model with a horizontal resolution of 2 km (Japan Meteorological Agency 2019). Therefore, the method of diagnostically forecasting senjo-kousuitai events that may occur even after 10 hours is constructed using the formation factors of senjo-kousuitai that are judged from environmental atmospheric fields.
In previous studies, several factors that favor the occurrence of senjo-kousuitai events have been examined, including in several of the case studies described in the Introduction. It can easily be understood that favorable conditions for the formation and development of convective cells include a low LFC and a high EL (e.g., Kato and Goda 2001; Kato 2006). Both these conditions are often produced by the inflow of low-level air with high equivalent potential temperature. Moreover, large inflow amounts of water vapor below a height of 1 km can produce large amounts of precipitation. Since the existence of midlevel dry air inhibits the further development of convective cells, it is also necessary for the air at this level to be humid (e.g., Kato 2006).
Furthermore, Section 3 explains that senjo-kousuitai events are caused by the stagnation of band-shaped precipitation systems that are organized by several multi-cell clusters consisting of convective cells. In previous studies (e.g., Fovell and Ogura 1988; Yoshizaki and Seko 1994), two-dimensional ideal numerical experiments were performed and showed that the low-level wind shear is significant for the formation of multi-cell clusters. Moreover, Seko (2010) showed that the difference in wind direction between low and middle levels influences the back building type formation of band-shaped precipitation systems; however no index for quantitatively estimating three-dimensional vertical shear for the organization of convective cells has ever been proposed.
Storm relative environmental helicity (SREH), often used in occurrence conditions of supercell storms, is examined as an index to quantitatively estimate three-dimensional vertical shear. Figure 14 displays the meaning of SREH on a hodograph. The values of SREH used in this study are calculated by integrating the inner products of storm relative horizontal wind vectors and horizontal vorticity vectors associated with the vertical shear between the near surface and a height of 3 km, where the storm relative horizontal wind vectors describe the difference between the environmental wind vectors and the traveling vectors of storms. The SREH values correspond to the double of gray area shown in Fig. 14. The positive values in Fig. 14 mean that the vertical change in the wind direction veers to the upward direction, which indicates low-level warm air advection, and the larger values correspond to larger inflow amounts of water vapor on the right-hand side of the traveling directions of storms. The methods of Maddox (1976) and Bunkers et al. (2000) are used to estimate the traveling vectors of storms, and the maximum of the SREH values calculated using these two vectors is adopted for the diagnostic forecasting of senjo-kousuitai events.
Meaning of storm relative environmental helicity. The black curve is the hodograph between the near surface and a height of 6 km. 
is the environmental wind vector, 
is the traveling vector of the storm, and 
is the horizontal vorticity vector associated with vertical shear. The subscript on denotes the height (km) of the environmental wind vector, and 
is the unit vector in the vertical direction. u and ν are the zonal and meridional winds, respectively.
The traveling vectors of storms estimated by Maddox (1976) and Bunkers et al. (2000) are compared with the orientations of the band-shaped rainfall areas observed in the senjo-kousuitai events introduced in Section 2. In a typical case of the broken line type (Fig. 6a), the Izu-Oshima event (Fig. 1a), the orientations of the three arrows shown Fig. 15a are significantly different, which could be caused by the formation of senjo-kousuitai extending along a stationary local front. The vertical shear is, however, very strong (SREH > 500 m2 s−2 around the local front). In typical cases of the back building type (Fig. 6b), the events of Hiroshima and Kumamoto (Figs. 1b, 2), the orientations of the senjo-kousuitai are sandwiched between two estimated traveling vectors of storms, and the orientations of the three arrows almost match (Figs. 15b, c). These patterns for the broken line and back building types were also found in other senjokousuitai events (not shown).
Hodographs at about 50 km low-level upstream point of the senjo-kousuitai in the cases of (a) Fig. 1a (0300 JST 16 October 2013), (b) Fig. 1b (0000 JST 19 August 2014), and (c) Fig. 2 (2100 JST 20 June 2016, depicted from JMA mesoscale analysis. Blue numbers denote pressure levels (hPa); pink and green arrows denote the traveling vectors of storms estimated from environmental conditions by using the methods of Maddox (1976) and Bunkers et al. (2000), respectively. Black arrows show only the direction of the observed senjo-kousuitai (their lengths do not denote the traveling velocity of storms).
It should be noted that some problems remain in using SREH to diagnostically forecast senjo-kousuitai events. The orientations of band-shaped rainfall areas could roughly correspond with the traveling direction of convective cells in cases of the back building type, as shown in Fig. 7a; however, the traveling velocity of convective cells also significantly affects the calculation of SREH. Furthermore, in cases of the broken line type, the meaning of SREH should be discussed in relation to the discrepancy between the traveling direction of convective cells and the orientation of band-shaped rainfall areas. After taking these considerations of the validity of SREH calculations into account, the estimation of the traveling vectors of storms should be examined to improve the forecasts of senjo-kousuitai events, as well as the geographical points used for calculating SREH.
5.2 Six favorable occurrence conditions of senjo-kousuitaiA set of conditions that favor the occurrence of senjo-kousuitai events are proposed here based on the results of previous studies discussed in this paper. Large inflow amounts of water vapor is necessary for the occurrence of heavy rainfall. This occurrence condition is estimated by FLWV at heights of 500-m height data, produced from data at 500 m above sea level for grid points with model terrain heights ≤ 300 m and from data at model terrain heights plus 200 m for those points with model terrain heights > 300 m (Kato 2018). The low-level water vapor field is judged based on 500-m height data, because heights of 500-m height data are representative heights to examine the initiation of moist convection that cause heavy rainfall in East Asia. It should be noted that 500-m height data is used because a height of 500 m is below the ground level in mountainous areas. The formation of convective cells is judged by the distance between heights of 500-m height data and LFC estimated from the heights (dLFC). The humid condition at middle levels is checked by RH at pressure levels of 700 hPa and 500 hPa. In addition to these three conditions, SREH are statistically investigated to estimate the vertical shear in environmental fields of previous 24 senjo-kousuitai events, listed in Table 2. It should be noted that the condition related to the equivalent potential temperature is not considered because its values are heavily dependent on the region and season, and it is indirectly related to the conditions of FLWV and dLFC.
The environmental conditions of the previous senjokousuitai events that occurred in various places in Japan shown in Fig. 16 are estimated from the JRA-55 reanalysis (Kobayashi et al. 2015) for the events before 2004 and from the JMA mesoscale analysis (Japan Meteorological Agency 2013) for the events after then. The values of FLWV and dLFC are taken from points around 50 km upstream from the areas where the senjo-kousuitai occurred, and those for SREH and RH are roughly averaged over an area lying within about 50 km of the event. For most of the events, the FLWV is greater than 200 g m−2 s−1, dLFC is less than 500 m, the RH is higher than 80 %, and the SREH is over 150 m2 s−2 (see Table 2). Since the adoption of these values for forecasts may lead to some senjo-kousuitai events being missed, threshold values that would meet the conditions for all the examined events, which are shown in Table 3, are set to describe the conditions that favor the occurrence of senjo-kousuitai events. In addition to these four conditions, synoptic-scale ascending areas and the development of convective cells are considered to construct six favorable occurrence conditions of senjo-kousuitai events that have been used operationally by the JMA since 2016. It should be noted that these two additional conditions are introduced to reduce missing forecasts of senjo-kousuitai events. The synoptic-scale ascending areas are judged by the upward velocity field at 700 hPa (W700), averaged horizontally by about 400 km. The development of convective cells is judged by the condition that EL estimated from heights of 500-m height data is higher than 3000 m, because warm air advection frequently appears at 700–850 hPa, suppressing the further development of convective cells (Kato et al. 2007).
Occurrence locations of the senjo-kousuitai events listed in Table 2. Prefecture names are in italics.
An example of diagnostic forecasts of senjo-kousuitai events with six favorable occurrence conditions is also given here. Figure 17 shows the forecasts of the Hiroshima event by the JMA mesoscale model (MSM) (Japan Meteorological Agency 2013) for the initial time of 1200 JST 19 August 2014 (Fig. 1b). The MSM fails to directly forecast the Hiroshima event (Fig. 17c) because, as discussed in subsection 4.2, senjo-kousuitai events are difficult to be reproduced by numerical models with a horizontal resolution of 5 km, such as MSM. On the other hand, the six conditions estimated from the MSM forecasts can successfully capture the occurrence areas of the Hiroshima event (solid ellipses in Fig. 17b). The six conditions can also diagnostically forecast the occurrence of senjo-kousuitai events in the northwestern part of Kyushu Island (broken red ellipses); however, a false alarm is made for the occurrence of senjo-kousuitai events on the western side of Shikoku Island (blue ellipses).
(a) Distributions of hourly accumulated R-A precipitation amounts at 0300 JST and 0400 JST 20 August 2014. (b) Distributions of areas that satisfy the six favorable occurrence conditions of senjo-kousuitai events at 0200 JST and 0300 JST 20 August, judged from the forecasts of JMA mesoscale model for the initial time of 1200 JST 19 August. FT denotes the forecast hour. (c) Same as (a), but for the forecasts of JMA mesoscale model for the initial time of 1200 JST 19 August.
In Japan, band-shaped precipitation systems sometimes stagnate for a few hours, causing localized heavy rainfall with accumulated three-hour precipitation amounts larger than 200 mm. In such events, the distributions of accumulated precipitation amounts also show band-shaped, and these band-shaped areas have recently been given the name “senjo-kousuitai” in Japanese, based on the characteristics of their shape. The formation processes of senjo-kousuitai events are categorized mainly into two types: One is the broken line type, in which convective cells simultaneously form on a quasi-stationary local front by the inflow of warm and humid air, and the other is the back building type, in which new convective cells successively forming on the upstream side of low-level winds linearly organize with pre-existing cells. In some events with the typical back building type, a hierarchical structure consisting of convective cells, multi-cell clusters, and quasi-stationary band-shaped precipitation systems can be observed. Based on their shape and formation characteristics, the senjo-kousuitai is defined as a band-shaped heavy rainfall area with a length of 50–300 km and a width of 20–50 km, produced by successively formed and developed convective cells, lining up to organize multi-cell clusters, and passing or stagnating at almost the same place for a few hours. A purely quantitative definition of senjo-kousuitai is not possible because the senjo-kousuitai may be directly linked to severe disasters such as landslides and floods. In other words, if some quantitative values (e.g., hourly accumulated precipitation amounts) were to be adopted for the definition, senjo-kousuitai events that do not satisfy the definition may occur to cause severe disasters and their forecasts may be missed. A similar definition has been used operationally by JMA since 2016.
The numerical reproduction of senjo-kousuitai events requires a horizontal model resolution of at least 2 km, which can roughly resolve the formation and development processes of senjo-kousuitai; however, a resolution of 250–500 m is necessary to accurately reproduce the inner-core structures of senjo-kousuitai, which agrees with the results of Bryan et al. (2003). Moreover, the accuracy of initial conditions strongly influences the reproduction of senjo-kousuitai events. In the Hiroshima event on 20 August 2014, NHM2km quantitatively reproduced the senjo-kousuitai when the initial conditions from 10 hours before the event were used, although the predicted maximum accumulated precipitation amounts were considerably reduced as the initial time became closer to the occurrence time of the senjo-kousuitai. This indicates that the successful reproduction of the Hiroshima event introduced in this study is fortunate, and in the present circumstances, it is usually difficult to accurately forecast senjo-kousuitai events even if a 2-km horizontal resolution model can be used. The improvement of the forecasts of senjo-kousuitai events requires a further higher horizontal model resolution and more precise physical processes, as well as more accurate initial conditions especially in low-level water vapor and wind fields. In addition, ensemble forecasts with mesoscale models are useful for reducing the uncertainty in numerical predictions and for providing probability information related to the occurrence of senjo-kousuitai events (e.g., Otani et al. 2019).
Another attempt to diagnostically forecast the occurrence of senjo-kousuitai events was conducted in this study, and the following six favorable occurrence conditions were constructed based on the statistical examination of environmental atmospheric fields in previous 24 senjo-kousuitai events; 1) FLWV > 150 g m−2 s−1, 2) dLFC < 1000 m, 3) RH at 500 hPa and 700 hPa > 60 %, 4) SREH > 100 m2 s−2, 5) W700 > 0, and 6) EL > 3000 m. This diagnostic method can successfully capture occurrence areas of senjo-kousuitai events that are not forecasted by MSM, such as the Hiroshima event. On the other hand, this method often makes a false alarm for the occurrence of senjo-kousuitai events because threshold values for these six conditions are set so as not to miss any events. Therefore, other conditions should be additionally used to reduce false alarm cases, for examples these could include the high equivalent potential temperature and the existence of low-level wind convergence.
Statistical verification of the six favorable occurrence conditions of senjo-kousuitai events is required to ascertain their effectiveness for the operational use by the JMA. For the verification, it is necessary to strictly determine the scale of senjo-kousuitai and accumulated precipitation amounts in the events, and also to automatically extract the events. An attempt to do this was recently conducted by Hirockawa et al. (2020). Moreover, the definitions of the favorable occurrence conditions should be improved based on an examination of further number of senjo-kousuitai events, and the estimation of the traveling vectors of storms used in calculating SREH should be examined so that they improve the forecasts of the events. These are our future issues.
The author thanks anonymous reviewers and Prof. M. Kawashima of Hokkaido University for their helpful comments, which improved the original manuscript. He is also grateful to Mr. Hiroshige Tsuguti for extracting heavy rainfall events and producing the R-A precipitation data with a horizontal resolution of 5 km in 1989–2015 warm seasons (April–November) by using the method of TK14.
