The Role of Typhoon Kilo (T1517) in the Kanto-Tohoku Heavy Rainfall Event in Japan in September 2015

From 9 to 11 September 2015, the Kanto and Tohoku regions of Japan experienced an extremely heavy rain fall event. The synoptic-scale field was characterized by two typhoons, Etau (T1518) and Kilo (T1517). After Etau made landfall in the Tokai region and transformed into an extra-tropical cyclone over the Sea of Japan, meridionally-oriented rain bands persisted over the Kanto region for about 12 hours and caused heavy rainfall, particularly over the Tochigi prefecture. During this time, Kilo approached the eastern ocean of the Kanto region. In this study, we examine the role of Kilo in this event by conducting numerical experiments using a stretched version of the Nonhydrostatic Icosahedral Atmospheric Model, configured with a minimum grid interval of about 5.6 km. The control experiment reproduced intense rain bands around the same period and place as the observed event, although they were not reproduced in an experiment with a longer lead time. Sensitivity experiments were conducted, in which Kilo was weakened by removing moisture in its central region with a longer lead time. In contrast to the expectation that reduced moisture would lead to a weaker typhoon, and hence weaker rain, the sensitivity experiment reproduced the rain band with a realistic location, but 5 % less precipitation than the control experiment. Furthermore, this experiment indicated that precipitation over the outer band of Etau, which covers the Kanto region, increased by 10 % compared to the control experiment. We found that a southeasterly wind, induced by a high pressure ridge between Kilo and the Kanto region, played a greater role in supplying moisture to the Kanto region than the strong easterly wind produced by the pressure gradient between Kilo and the Okhotsk high. In this case, weaker Kilo resulted in an enhanced northwestward moisture flux associated with the ridge, thereby inducing heavier rainfall over the Kanto region.


Introduction
From 9 to 11 September 2015, an extremely heavy rainfall event occurred in the Tochigi prefecture, 120 km north of Tokyo, Japan, causing flooding the Kinu River. During this event, a linear rain band (LRB) formed over the Tochigi prefecture (LRB1; Fig. 1a), in the northern part of the Kanto region (Fig. 1b). In the Tochigi prefecture, a local observation site measured total precipitation exceeding 600 mm in 48 hours. In the later stages of this event, another LRB formed in the northern part of the Tohoku region (LRB2; Figs. 1a, b) causing heavy rainfall there. Both LRBs extended hundreds of kilometers. According to the Japan Meteorological Agency (JMA), the observed precipitation of LRB1 was 59.5 mm hr −1 at its maximum at 15UTC 9 September, and that of LRB2 peaked at 62.0 mm hr −1 at 14UTC 10 September. This heavy rainfall event is now referred to as the September 2015 Kanto-Tohoku heavy rainfall event.
Two typhoons were associated with this event. Typhoon Kilo (T1517) initially formed in the eastern Pacific and moved westward, approaching Japan on 9 September. Typhoon Etau (T1518) was generated to the south of Japan on 7 September, made landfall in the Tokai region on 9 September, and subsequently transformed into an extra-tropical cyclone on the same day in the Sea of Japan (Fig. 1b). LRBs in this heavy rainfall event were formed in the outer band region of Etau.
Previous studies have investigated heavy rainfall events associated with LRBs. Kato and Goda (2001) argued that a heavy rainfall event in Niigata, Japan, on 4 August 1998 was caused by strong convective instability due to lower-tropospheric humid air and mid-tropospheric dry air. In contrast, the importance of mid-tropospheric moisture was acknowledged by Hirota et al. (2016), who showed that a large transient filamentary region of water vapor reaching the midtroposphere (known as an atmospheric river) played an essential role in supplying moisture to an extreme rainfall event in Hiroshima, Japan, on 19 August 2014. Wang et al. (2009) suggested that Typhoon Songda (2004) had a dominant role in supplying moisture to a heavy rainfall event that occurred over southern central Japan from 2 to 4 September 2004, despite being 1200 km away over the western North Pacific.
For the September 2015 Kanto-Tohoku heavy rainfall event, several studies have proposed LRB formation mechanisms. Kunii et al. (2016) conducted ensemble data assimilation using the atmospheric motion vectors from Himawari-8 rapid scan data (Bessho et al. 2016). They suggested that the LRBs were strongly correlated with low-to mid-level moisture to the south of the Kanto region and were intensified by low-level moisture convergence. Section 3 of Fujitani et al. (2016), led by H. Tsuguti, used backward trajectory analysis to argue that moist warm air was advected from the region around Kilo. Section 5 of Fujitani et al. (2016), led by K. Tsuboki, pointed out the contributions of huge southerly moisture transport and a high pressure low-level ridge at the southeast of the Kanto region (cf. Fig. 1b). He suggested that anti-cyclonic flow around this ridge converged to the south of the Kanto region, with westerly and southwesterly winds related to the uppertropospheric trough. Kitabatake et al. (2017) analyzed these synoptic-scale characteristics in further detail. They found that the upper-tropospheric trough ad-vected dry air over the Kanto region, resulting in a convectively unstable condition. In addition, since a high potential vorticity disturbance was approaching, synoptic upward motion was diagnosed by Trenberth forcing (Trenberth 1978). This is an approximation to a forcing term of the quasi-geostrophic omega equation, which is calculated from horizontal advection of absolute vorticity by thermal wind. Adding to these favorable conditions for heavy rainfall, latent heat release by precipitation to the north of Etau caused negative potential vorticity advection, prohibiting eastward propagation of the trough. It also generated an upper-tropospheric ridge to the east of the trough, which intensified the low-level Okhotsk high pressure. Kitabatake et al. (2017) argued that the Okhotsk high pressure supplied moisture to the Kanto region via the associated easterly or southeasterly wind.
The typhoons Etau and Kilo are notable characteristics of the September 2015 Kanto-Tohoku heavy rainfall event. Kitabatake et al. (2017) argued that Kilo and the Okhotsk high cooperatively enhanced the meridional pressure gradient and intensified the easterly wind. Fujitani et al. (2016) showed that this low-level easterly wind brought moist air, while moisture was also advected from the south, and argued for the importance of the southeasterly wind related to the low-level ridge. In summary, it is unclear whether the easterly associated with Kilo was an essential contributor to the heavy rainfall event or only a secondary contributor. It is also unclear how Kilo affected the low-level ridge and associated southeasterly wind.
In this paper, we investigate the role of Kilo in the September 2015 Kanto-Tohoku heavy rainfall event using non-hydrostatic numerical simulations. We carry out sensitivity experiments to study the effect of Kilo on synoptic conditions, especially the low-level easterly and southeasterly winds and the low-level ridge to the east of the Kanto region.

Model description
We used a global stretched version of the Nonhydrostatic Icosahedral Atmospheric Model (NICAM; Tomita and Satoh 2004;Satoh et al. 2008; that enhanced the horizontal resolution around a selected target location (Tomita 2008a;Uchida et al. 2016). Grid intervals were about 5.6 km at the target location (36.7°N, 139.7°E in the Kanto region) and 560 km at the opposite side of the globe. The model had 40 layers, with a top height of about 38 km. Moist processes were explicitly calculated using a six-category single-moment bulk cloud microphysics scheme (Tomita 2008b), without any cumulus parameterization. Unrealistic representation of clouds in the remote regions with coarse grids has minimal influence on the results near the target region in this study because of the short integration period of up to three days. More detailed characteristics of the stretched-grid NICAM simulations are described in Uchida et al. (2016). A modified Mellor-Yamada level 2 scheme (Mellor and Yamada 1982;Niino 2004, 2009;Noda et al. 2010) was applied for boundary layer turbulent processes. Radiation processes were calculated with K-distribution model (MSTRNX; Sekiguchi and Nakajima 2008). The sea surface temperature was predicted using a slab ocean model with 15 m depth and was nudged toward reference data interpolated from the National Center for Environmental Prediction final analysis dataset at a relaxation time of five days. The MATSIRO scheme (Takata et al. 2003) was applied for land surface boundary conditions.
We used the Grid Point Value (GPV) dataset, which is used as initial data of JMA Global Spectrum Model (GSM); the Japanese 55-year Reanalysis dataset (JRA55) produced by JMA (Kobayashi et al. 2015); and the best track data of Kilo and Etau produced by JMA for verification. We also used Global Satellite Mapping of Precipitation (GSMaP; Kubota et al. 2007).

Experimental set-up
After examining preliminary experiments starting at various initial conditions, we defined a control experiment with an initiation date of 00UTC 9 September, hereafter referred to as CTL. CTL reproduced realistic heavy rainfall in terms of location and precipitation amount (see Section 3.1).
We conducted two sensitivity experiments. The first was identical to CTL except that it was initialized at 00UTC 8 September (hereafter referred to as EARLY). As shown in detail in Section 3.1, EARLY reproduced Kilo with lower central pressure and a westward-shifted track compared to observations but did not reproduce the rainfall over the Kanto region.
The second sensitivity experiment was a moistureremoved experiment initialized at 00UTC 8 September (hereafter referred to as RH0). In RH0, the relative humidity derived from the GPV dataset for the Kilo region, where the wind exceeds 30 kt defined by best track data of JMA, was set to 0 % at all levels on the initiation date (cf. Fig. 1c). When the moisture was removed near the center of Kilo, its central pressure became higher, and its cyclonic circulation became weaker. Our intention was to investigate the effect of Kilo on the easterly or southeasterly winds and the contributions of these winds to the moisture supply to the Kanto region. We chose to initialize RH0 at 00UTC 8 September because the moisture effect of Kilo is inseparable from that of the environment after 9 September.
We defined the center of the simulated typhoon as the location of minimum sea level pressure in a 0.5° square around the previous center (one hour before).

Simulated profiles of precipitation and typhoons
CTL reproduced a LRB between 00UTC 9 September and 00UTC 11 September over the Kanto region (Fig. 2a), and its location corresponded well to the observed LRB1 (Fig. 1a). CTL also reproduced a rain band in close proximity to the observed LRB2 but with less precipitation. By contrast, EARLY did not reproduce the rain bands (Fig. 2b). CTL reproduced characteristics of Kilo well in terms of surface pressure distribution and the track compared with the best track data provided by JMA. In EARLY, Kilo had about 5 hPa lower central pressure, and its track was shifted about 100 km to the west (Figs. 3a, b).
The moisture-removed experiment (RH0) was intended to simulate a weakened Kilo. The sea level pressure at the center of Kilo was more than 1000 hPa at 18UTC 9 September (Fig. 3b), which is weaker than the 975 hPa in CTL at the same time. Wind speeds near the center of Kilo in RH0 were much weaker than those of CTL and EARLY (details are described in later section). Kilo was located up to 200 km further west than the best track data. As for the Kanto region, a LRB was reproduced, despite the early initiation date, as in the observed LRB1 but with 5 % less rain in the Kanto region than in CTL (Fig. 2c). However, over the outer band of Etau, the amount of precipitation increased by 10 %. This was unexpected because previous studies (cf. Section 3 of Fujitani et al. 2016) suggested that Kilo played a role in intensifying the rain band corresponding to LRB1. Conversely, this result suggests that Kilo weakened the heavy rainfall. The rain band corresponding to LRB2 was not reproduced clearly in RH0 (Figs. 2b, c).
The simulated Etau is stronger than best track data in all three experiments, and the central pressure is about 10 hPa lower. The central pressure variability among the three experiments is within 5 hPa. Its tracks almost overlapped in the three experiments except in the later period (Figs. 3a, c).

Synoptic conditions
All three experiments simulated the evolution of the upper-tropospheric fields similarly (Fig. 4): the upper trough was over the west side of Japan and stagnated, and the upper ridge was over the northeast of Japan. This behavior is consistent with the analysis by Kitabatake et al. (2017).
The role of Kilo on synoptic scale conditions is more clearly visible by examining the low-level struc-ture. The lower-tropospheric fields were significantly different (Fig. 5), particularly around the Kanto region and the eastern sea. The low-level ridge, represented by a 1500 m isobar around 35°N 145°E, extended more southward in CTL and RH0, and the southeasterly wind prevailed near the Kanto region, associated with this low-level ridge. By contrast, in EARLY, the low-level ridge was not clear, and the easterly wind prevailed. Comparing the precipitating (CTL and     RH0) and non-precipitating (EARLY) cases showed that southeasterly wind can be important. The timelongitudinal section of geopotential height around  shows that two typhoons formed this low-level ridge, which is located at 140°E to 145°E. This ridge is the relatively high pressure region between these two typhoons. In EARLY, the westward-shifted Kilo with lower pressure resulted in a lower height of the low-level ridge. By contrast, in RH0, the weakened Kilo maintained the high geopotential height region. This indicates that the lowlevel ridge extension was sensitive to the strength and track of Kilo. The zonal cross-section, averaged over the mature stage of the low-level ridge (Fig. 6e), shows that the zonal height gradient is positive (nearly zero) in CTL and RH0 (in EARLY). This suggests that the westward-shifted Kilo, with lower pressure, disturbed the southerly component of wind at 140°E to 145°E. Previous studies argued that the pressure gradient between Kilo and the Okhotsk high enhanced the easterly and southeasterly winds, and these winds supplied moisture. Although the three experiments described here showed the Okhotsk high with similar intensity, the simulated precipitation is different. The low-level ridge was independent of the easterly wind, which differs from the arguments made by Kitabatake et al. (2017).

Moisture flux variabilities
The amount of low-level moisture flux was larger in CTL than in EARLY (Fig. 7). The southeasterly wind of the low-level ridge supplied moisture more effectively around the Kanto region, resulting in a more moisture-rich condition in the lower troposphere in the Kanto region. This result shows that the southward extension of the ridge supported the moisture supply. Cross-section of (a) to (d), averaged from 12UTC 9 September to 00UTC 10 September. This is consistent with the role of the low-level ridge, which is pointed out in Section 5 of Fujitani et al. (2016). The differences between EARLY and CTL (Figs.  8a, b) show that a stronger and westward-shifted Kilo prevented extension of the ridge and reduced the southerly component of horizontal wind significantly. Around Kilo, the moisture flux was larger due to strong wind. However, over the Kanto region, the moisture flux decreased. The differences between RH0 and CTL (Figs. 8c, d) show that the southeasterly flow was maintained, and the amount of moisture flux increased around the Kanto region. This structure was favorable for moisture supply.
Our results suggest that the southeasterly moisture advection, at the lower troposphere by the low-level ridge, was of primary importance. When only easterly winds existed, and the low-level ridge was weaker (EARLY), the low-level moisture supply to the Kanto region was underestimated, although air parcels of heavy rainfall originated near Kilo (Section 3 of Fujitani et al. 2016). To understand the role of the moisture supply to the Kanto region, it is necessary to separately evaluate the easterly wind, which is enhanced by the pressure gradient between the Okhotsk high pressure and Kilo, and the southeasterly wind, associated with the low-level ridge.
The southerly moisture-rich air along 140°E lon- gitude is another characteristic condition ( Fig. 5 and cf. Section 5 of Fujitani et al. 2016). All three experiments similarly reproduced this situation over the Kanto region and its southern sea along 140°E. This southerly moisture-rich air has high relative humidity in the mid-level atmosphere and is convectively unstable (Fig. 9). This suggests that Etau and its outer band led to conditions that were favorable for convection, and the low-level moisture transport may have triggered the rainfall. In EARLY, this unstable region was relatively unclear, and the middle-level atmosphere was drier. The stronger cyclonic circulation of Kilo in EARLY may oppose the favorable conditions for heavy rainfall in the outer band of Etau. On the southern sea of the Kanto region, middle tropospheric humidity was increased in RH0. This corresponds well to a large area of precipitation (Fig. 2c), although we cannot determine which is the cause and which is the result.

Summary and discussion
The present study examines the cause of the heavy rainfall event that occurred between 9 and 11 Sep-tember 2015 over the Kanto region, Japan. Numerical experiments, using the stretched version of NICAM, were designed to investigate the effects of Kilo on this event. We reproduced a LRB over the Kanto region, which had a realistic location and precipitation amount, in a control experiment initialized at 00UTC 9 September (CTL), while only a weak rain band was produced over the Kanto region in an experiment initialized at 00UTC 8 September (EARLY). These two experiments show the same favorable conditions for heavy rainfall, such as the same upper-tropospheric conditions and middle tropospheric moisture, and similar convectively unstable conditions. The only exception was the low-level moisture flux, which increased in CTL to the southeast of the Kanto region (Fig. 7). Our results suggest that the southeasterly moisture supply associated with the low-level ridge, pointed out by Fujitani et al. (2016), was essential in supporting the heavy rainfall, while that of the easterly moisture supply, pointed out by Kitabatake et al. (2017), was of secondary importance.
In contrast to the expectation that a stronger typhoon would cause heavier rainfall, rainfall over the Kanto region was enhanced in the sensitivity experiment, in which Kilo was weakened (RH0). Our result suggests that, in this particular event, Kilo had a suppressing effect on the low-level ridge formation, which provided the substantial southeasterly moisture flux and supported the heavy rainfall. Figure 10 schematically illustrates the relationship between Kilo, the ridge, and the rain bands. In RH0, despite the reduced easterly wind, moisture supply was plentiful due to the unsuppressed low-level ridge (yellow solid line in Fig. 10) and the associated southeasterly flow. The relative anti-cyclonic circulation of the weakened Kilo supported the low-level ridge and maintained the southeasterly wind (Fig. 9). In other words, the cyclonic circulation of Kilo disrupted the southward extension of the low-level ridge.
In this study, EARLY and RH0 did not reproduced a rain band corresponding to LRB2. The initiation date of these two experiments, 00UTC 8 September, is too early to reproduce and discuss LRB2, which peaked at 14UTC 10 September. The importance of the lowlevel ridge is suitable for LRB1. However, in the period when LRB2 developed, Kilo and Etau moved, and the low-level ridge is vanished. Thus, we need more study to discuss the role of Kilo in LRB2. In EARLY, Kilo was reproduced slightly stronger, and its track was shifted westward (Fig. 3), resulting in a weaker ridge and reducing the moisture supply (Fig. 6). We note that accurate forecasting of typhoon track and intensity is important for predicting relatively remote heavy rainfall events.
Although this study clarifies the effect of Kilo on the ridge and the southeasterly water vapor transport between Kilo and Etau, the simulated track of Kilo strongly depends on the lead time. A sufficiently large ensemble experiment would enable us to extract more robust information for this event.