Investigation of Characteristics of Maximum Storm Surges in Japanese Coastal Regions Caused by Typhoon Jebi (2018) Based on Typhoon Track Ensemble Simulations

The maximum storm surges caused by Typhoon Jebi (2018) were examined using a storm surge model and using track ensemble simulations based on a meteorological model and a parametric tropical cyclone (TC) model. The storm surge at Osaka Port was estimated more accurately by the meteorological model than the parametric TC model. The differences between both models were due to a “wind setup effect”, where the topography en-Corresponding provide important information enabling effective disaster prevention and risk management.


Introduction
Typhoon Jebi passed over the Shikoku and Kinki districts of Japan in early September 2018. It was accompanied by extreme winds, which caused severe damage to urban centers (e.g., Takemi et al. 2019;Takabatake et al. 2018). Low pressure and strong winds with maximum gusts of 58.1 m s −1 at the meteorological station in Osaka led to a record-breaking maximum storm tide of 3.29 m (storm surges of 2.77 m) 1 for that city. Storm surges caused by Typhoon Jebi resulted in extensive damage to the coastal region. Although storm surges seldom occur, their impact on coastal regions can be devastating (e.g., Kohno et al. 2018;Mori et al. 2019a). Consequently, storm surges from typhoons of varying intensity must be quantitatively estimated throughout coastal regions, including in areas where storm surges have not yet occurred.
Storm surges occur because of the effects of strong onshore winds (wind setup effect) and low pressure (inverse barometer effect). To estimate the severity of storm surges, information regarding the surface winds and sea-level pressure associated with typhoon intensity is needed, as well as a reliable storm surge model. Typically, several methodologies are used to assess storm surge risk. The most popular approach is to simulate a historical event or use a scenario ensemblebased (i.e., storyline approach) framework (e.g., Nino miya et al. 2017). An alternative approach is to use a large number of ensembles, such as "synthetic" typhoons (e.g., Nakajo et al. 2014) or typhoons dynamically simulated by a climate model (e.g., Yasuda et al. 2014;Yang et al. 2020;Mori et al. 2019bMori et al. , 2021.
The key to these approaches is the selection of typhoon characteristics in parameter space, such as the central pressure, maximum wind speed, and radius of the maximum wind speed. However, the influence of topography on the dynamic aspects of a typhoon must also be considered.
The methods used to estimate storm surges in Mori et al. (2020) can be divided into four categories: numerical models of specific typhoons, global climate models, climatological approaches, and statistical approaches. The models used to simulate a typhoon can be subdivided into two groups: parametric tropical cyclone (TC) models and numerical weather prediction models (hereafter, meteorological models). A meteorological model is capable of more accurately simulating a typhoon.
Parametric TC models, except for the Generalized Holland Asymmetric Model (Gao et al. 2014), assume that the sea-level pressure distribution of a typhoon has an axisymmetric structure. Surface winds are estimated by the gradient wind equation with surface friction effect. One advantage of parametric TC models is that they have low computational costs. The axisymmetric sea-level pressure, which varies with typhoon parameters such as strength and size (the radius of the maximum wind), can be easily reproduced by hanced surface winds over Osaka Bay. The typhoon-track ensemble simulations demonstrated that the maximum storm surge was dependent on the perturbation of the track of Typhoon Jebi along the entire coast of the Japanese Islands, including the main island, Kyushu, and Shikoku. Open shallow bays had maximum storm surges exceeding 2.50 m. In coastal areas where larger maximum storm surges were estimated, the longitudinally perturbed "worst-case course" appeared 0.4 -0.8° west or east of the "hit course", indicating that the wind setup effect was an important factor in the maximum storm surge. The distance of the worst-case course from the hit course was almost the same as the radius of the maximum wind of Typhoon Jebi. Although the models had similar worstcase courses for each coastal area, the meteorological model estimated a slightly higher simulated maximum storm surge than the parametric TC model. For the main island, Kyushu, and Shikoku, approximately 6 % of the maximum storm surges exceeded 2.00 m. Although these values may differ for other typhoons and sampling points, it is important to estimate the maximum storm surges and worst-case courses at all coastal areas, including regions where storm surges by typhoons are unknown yet they occurred, because this will provide important information enabling effective disaster prevention and risk management.
Keywords Typhoon Jebi; maximum storm surges; typhoon-track ensemble simulation Citation Otaki, T., H. Fudeyasu, N. Kohno, T. Takemi, N. Mori, and K. Iida, 2022: Investigation of characteristics of maximum storm surges in Japanese coastal regions caused by Typhoon Jebi (2018) based on typhoon track ensemble simulations. J. Meteor. Soc. Japan, 100, 661-676, doi:10.2151/jmsj.2022 a parametric TC model. The typhoon track can be set arbitrarily, so parametric TC models are suitable for ensemble experiments. The sea-level pressure is axisymmetric, but the wind distributions are asymmetric because the effect of movement is taken into account.
The asymmetric characteristics of a typhoon become important when its structure is influenced by complex topography, which is often the case in inland bays. Meteorological models can simulate a typhoon with a structure rendered asymmetric by the influence of topography and mid-latitude environments. Thuy et al. (2014) simulated storm surges due to Typhoon Kalmaegi (2014) at the Hon Dau station in Vietnam using both parametric TC and dynamical models. The dynamical model provided more accurate results, with the difference in storm surges estimated by both models being approximately 1 m.
Storm surges in coastal areas are affected not only by the strength and size of a typhoon but also by the typhoon track. Therefore, ensemble experiments must be able to estimate storm surges for various typhoon tracks. For example, Shibutani et al. (2015) used a coupled model of surge, wave, and tide "SuWAT" for storm surge calculation and a parametric TC model for simulating Typhoon Vera (1959) with various tracks. They showed a variance of approximately 2 m in the maximum storm surges estimated for Nagoya Port, with a longitudinal difference in the typhoon tracks of approximately 50 km. In Toyoda et al. (2020), the maximum storm surge caused by Typhoon Hagibis (2019) in Tokyo Bay was evaluated using a highresolution coupled typhoon-ocean model. The method adopted was almost the same as the Typhoon-Track Ensemble Simulation (T-TES) method, which was developed by Yamasaki et al. (2017). The T-TES method modulates the initial and boundary atmosphere data so that a meteorological model can be used for ensemble experiments with longitudinally perturbed typhoon tracks. If Typhoon Hagibis had passed 100 km west of the port at Tokyo, a storm surge of approximately 3 m might have occurred. Such a storm surge would have exceeded the historical maximum tide level.
The surface winds and sea-level pressure distributions associated with a typhoon simulated by a meteorological model are better suited to estimate the maximum storm surge along coasts. Kowaleski et al. (2020) simulated a storm surge induced by Hurricane Irma (2017) along the southeast coast of the USA using a Weather Research and Forecast (WRF) ensemble model and the advanced circulation (ADCIRC) storm surge model. Colle et al. (2015) studied storm surges by Hurricane Sandy (2012) along the northeast coast of the USA using a subset of WRF ensemble members. Although storm surges in the USA have been investigated via ensemble experiments, no study has estimated the maximum storm surges along the entire coast of Japan via ensemble experiments and a meteorological model. Recently, ensemble experiments have been limited to bays with large storm surges due to the high computational cost (e.g., Toyoda et al. 2020).
This study evaluated maximum storm surges in multiple scenarios, assuming various tracks using the storm surge model and T-TES, which used inputs from both a parametric TC model (hereafter Para-Jebi) and a meteorological model (hereafter WRF-Jebi). The purpose of this paper is to compare the maximum storm surges derived by both models. A further aim of this study is to quantitatively investigate the possible maximum storm surges along the entire coast of Japan, including the main island, Kyushu, and Shikoku. Our ensemble experiments using the T-TES method estimate storm surges and longitudinally perturbed "worst-case courses" assuming various tracks for Typhoon Jebi in each coastal area. The reason for focusing on Typhoon Jebi is that it caused enormous damage from storm surge in recent years and had a typical track of typhoons that affect Japan. By estimating the maximum storm surge along the coast, it was possible to determine the proportion of each region at high risk of storm surges. The remainder of this paper is organized as follows. The following section briefly describes models used in our research. In Section 3, storm surges determined by Para-Jebi and WRF-Jebi are validated and the results of ensemble simulations are discussed. Maximum storm surges and worst-case courses along the entire coast of Japan, including the main island, Kyushu, and Shikoku, are estimated in this section. The main findings are summarized and discussed in the final section.

Methodology
This study used the T-TES method (Yamasaki et al. 2017) to conduct ensemble experiments, in which the tracks of Typhoon Jebi were varied in the longitudinal direction. The T-TES method modulates the initial and boundary atmospheric conditions so that typhoons with different tracks can be analyzed using a meteorological model. It should be noted that the lower boundary conditions (i.e., sea surface temperature) were not modulated in this study. Simulations subsequently performed using the meteorological model can reproduce typhoons passing through the region, assuming a track shifted westward or eastward of the actual track. The T-TES method allows the original speed and direction of travel to be largely maintained throughout the typhoon's tracks. A total of 83 tracks were created for Typhoon Jebi by T-TES, at 0.2° intervals up to 5.0° in the westward direction and 11.4° in the eastward direction. We followed the 0.2° longitudinal shift in the T-TES method to the guideline for "Storm surge inundation are map creation guide" of the Ministry of Land, Infrastructure, Transport and Tourism (2021). The details of how to obtain longitudinal shift are explained in Yamasaki et al. (2017).
The meteorological model (WRF-Jebi) used in this study was the Weather Research and Forecasting Model version 3.6.1 (WRF; Skamarock et al. 2008). A one-way domain was nested within the parent (outer) domain. The outer domain had 220 × 215 grid points with a horizontal resolution of 15 km, whereas the inner domain had 601 × 541 points with a horizontal resolution of 5 km. The outer domain was designed to simulate the large-scale atmospheric environment, including the typhoon structure, whereas the inner domain was designed to use the input for the storm surge simulations. The initial and lateral boundary conditions for the WRF-Jebi were derived from the Japanese 55-year Reanalysis Project datasets (JRA55; Kobayashi et al. 2015; details are available online at https://jra.kishou.go.jp/JRA-55/index_en.html). Only the outer domain was initialized and forced by JRA55, which were modulated using the T-TES method and used as the initial and boundary atmosphere data. The inner domain was initialized and forced based on the outer-domain results every 45 -90 s through 24 h in-tegration as the initial and boundary atmosphere data. The details of the inner-domain initial atmosphere data are provided in Appendix A. The integration period for the outer domain was from 0000 UTC on 2 September to 0000 UTC on 5 September 2018, whereas for the inner domain, it was from 0000 UTC on 3 September to 0000 UTC on 5 September 2018. Table 1 summarizes the WRF settings. A parametric TC model (Para-Jebi) was also used to estimate the surface winds and sea-level pressure distributions associated with 83 tracks of Typhoon Jebi with the same tracks and intensities as those of WRF-Jebi. The axisymmetric sea-level pressure was derived using the formula proposed by Fujita (1952), and the surface wind distribution was calculated from the gradient wind equation. The effect of typhoon movement was determined using the equation proposed by Miyazaki (1961). The surface friction coefficient, which represents the ratio of surface wind to gradient wind, was set to 0.7 in this study. This value was empirically determined based on observed winds and various simulation results. The details regarding the parametric TC model can be found in previous papers (e.g., Meteorological Research Institute of Japan Meteorological Agency (JMA) 2000; Kohno et al. 2001Kohno et al. , 2007Hossain et al. 2017). To allow comparison with the meteorological model, the other computational setting conditions for the Para-Jebi were the same in all simulations, i.e., the typhoon position, central pressure, maximum wind, and radius of the maximum wind used in the Para-Jebi were determined by the results of the WRF-Jebi obtained at 3-h intervals. It is complemented by linear interpolation between the inputs. In this study, the storm surge simulations were conducted using a storm surge model developed by the JMA. Table 2 summarizes the calculation parameters used in the storm surge model. This model has been used in previous research such as Kohno et al. (2007Kohno et al. ( , 2018, Kuroda et al. (2010), and Hossain et al. (2017). The bathymetric conditions used in the model were based on the ETOPO1 bathymetry data (National Oceanic and Atmospheric Administration: https://ngdc. noaa.gov/mgg/global/). Simulations were conducted using a horizontal resolution of 1.7 km. The 1.7-km grid resolution was used in previous operations of JMA (Japan Meteorological Agency 2007) and case studies (e.g., Kohno et al. 2007), and it gave reasonable results. The sea surface stresses are estimated with a constant drag coefficient (Cd = 3.2 × 10 −3 ), which is a typical value in stormy winds. Additionally, typical storm surge mechanisms for the inverse barometer effect and wind setup were considered. Astronomical tides and wave effects were not considered because the focus was to evaluate possible storm surges.

Storm surge simulation at Osaka Port
This study compared the maximum storm surges at Osaka Port derived by observations and both models. Figure 1 shows the track of Typhoon Jebi simulated in the inner domain, an overall map of Japan and enlarged view of major points, and an elevation above sea level around the Kinki district. The positions of WRF-Jebi were determined from the area of minimum sea-level pressure. The track, which was shifted eastward by 0.4° at the initial time using T-TES, was closest to the actual track of Typhoon Jebi from the best track archives (BT) of the Regional Specialized Meteorological Centers-Tokyo Typhoon Center around the Shikoku and Kinki districts. This simulation was defined as the control run (CTL), in which WRF-Jebi made landfall at Osaka Port at approximately 0520 UTC on 4 September 2018, 20 min after the actual time.
The time series of the central pressure and radius of the maximum winds of Typhoon Jebi, derived from the BT and CTL of WRF-Jebi, are shown in Fig. 2 and Table 3. In the BT, Typhoon Jebi had a central pressure of 960 hPa when it was closest to Osaka Port at 0500 UTC on 4 September 2018. The central pressure of WRF-Jebi differed from that derived from BT by the end of 3 September, and this discrepancy was not clear. However, the central pressure simulated at 0530 UTC 4 September 2018 was approximately 966 hPa, which fairly compared with that of the BT. The track and intensity of Typhoon Jebi around the Shikoku and Kinki districts were well reproduced by the WRF-Jebi. Figure 3 shows a time series of the storm surge observed at Osaka Port and that estimated by the storm surge model with both input models. The maximum storm surge of 2.73 m was observed at Osaka Port. The maximum storm surge of 2.49 m estimated using the WRF-Jebi inputs was slightly lower than the observation (difference of 0.24 m). The maximum storm surge estimated using the WRF-Jebi inputs occurred 20 min later than the observation. The maximum storm surge estimated using the Para-Jebi inputs (2.11 m) was lower than the observation by 0.62 m.
There were differences in the storm surges estimated using the WRF-Jebi and Para-Jebi inputs. This discrepancy was not attributed to the inverse barometer effects, but rather to the wind setup effect. The storm surge caused by the inverse barometer effect for WRF-Jebi was 0.46 m, and for Para-Jebi, it was 0.42 m. The wind setup effect for WRF-Jebi was approximately 2.03 m, which was larger than that for Para-Jebi (1.69 m). Figure 4 shows the surface winds around the Kinki district at 0500 UTC and 0520 UTC on 4 September. According to a JMA mesoscale analysis (MSM), a strong local wind area exceeding 35 m s −1 was estimated over Osaka Bay. The WRF-Jebi reproduced a similar local wind area of approximately 30 m s −1 , which agreed well with the MSM. According to data from the Behavior of Hypoxia in Osaka Bay project (http://teiten.pa.kkr.mlit.go.jp/obweb/data/c1/ c1_12.aspx), the maximum wind speed at Osaka Bay during Typhoon Jebi at 0600 UTC on 4 September   2018 was 28.7 m s −1 , which agreed well with the WRF-Jebi. On the other hand, in the Para-Jebi, the surface wind was below 25 m s −1 over Osaka Bay and there was no strong local wind. Unlike the WRF-Jebi, the Para-Jebi was not influenced by topography and was therefore, unlikely to produce realistic winds in areas with a complex topography.

Ensemble experiments at Osaka Port
Our ensemble experiments using the T-TES method estimated storm surges and longitudinally perturbed "worst-case course" assuming various tracks for Typhoon Jebi at Osaka Port. Figure 5 shows the 83 typhoon tracks according to different central pressure and radius of maximum wind values obtained from ensemble experiments using the T-TES method. The tracks were almost parallel to each other. The central pressure of Typhoon Jebi varied among the courses, with the values tending to be the same before making landfall on the Japanese Islands and increasing there after. The temporal changes in the radius of the maximum wind remained small over the ocean and increased after making landfall on the Japanese Islands.
The maximum storm surges at Osaka Port are shown in Fig. 6a for different typhoon tracks using the T-TES method. When Typhoon Jebi passed east of Osaka Port, the maximum storm surge was low; however, it was higher when it passed west of Osaka Port. The "hit course" was 0.2° to the east relative to the CTL, and the maximum storm surges were 2.00 m and 2.11 m in the simulations using the WRF-Jebi and Para-Jebi inputs, respectively. Here, "hit course" stands for the course where the center of a typhoon crosses just over the target location (here, Osaka Port) and is shown as 0.0° in Fig. 6. For all tracks, the CTL had a maximum storm surge of 2.49 m when estimated using the WRF-Jebi input, which represented the worst-case longitudinally perturbed course of Typhoon Jebi for the storm surge at Osaka Port. By contrast, the estimate using the Para-Jebi input had a higher storm surge of 2.30 m when the typhoon track was 0.4° west of the hit course. In Fig. 6a, the maximum storm surge calculated using the Para-Jebi smoothly changed, which gradually increased as tracks approached to the worst-case course from the west and decreased as tracks moved away from the worst course. By contrast, the estimates using the WRF-Jebi inputs abruptly increased as tracks shifted from 0.8° west of the CTL to 0.6° west. Figure  7 shows the surface winds around the Kinki district at the time of the maximum storm surges in cases of 0.6° and 0.8° west of the hit course. Although there was little difference in the wind speeds over Osaka Bay between the two tracks, the wind direction over Osaka Bay changed from south-southwesterly in the 0.6° west course to southerly in the 0.8° west course. The   topography around Osaka Bay was responsible for the changes in the direction of the surface wind over Osaka Bay. Osaka Bay is surrounded by mountains with a relatively high altitude rather than flat lowlands (Fig. 1b). Figure 6a also shows the occurrence time of the maximum storm surge in both models, which was determined by the time elapsed from the time of minimum sea-level pressure at each location (here, Osaka Port) for each track. Note that a negative (positive) time means before (after) the time of the minimum sea-level pressure, i.e., Typhoon Jebi was approaching (leaving). In both models, the occurrence time of the maximum storm surges was approximately 60 min in the worst-case courses. The difference in occurrence time between both models increased with the distance between the hit course and Osaka Port.

Assessment of the worst-case storm surge along
the coastal region Ensemble experiments estimated storm surges and longitudinally perturbed "worst-case courses" assuming various tracks for Typhoon Jebi in other coastal areas. Figure 6b shows the maximum storm surges and their occurrence times at Nagoya Port. The distribution of the maximum storm surges at Nagoya Port was similar to that at Osaka Port. The maximum storm surge estimated using WRF-Jebi inputs was 2.83 m in case of 0.4° west of the hit course, which was much larger than that estimated using the Para-Jebi inputs (1.87 m). Figure 8 shows the surface winds around Ise Bay at 0530 UTC on 4 September when the maximum storm surge occurred, derived by both models in case of 0.4° west of the hit course for Nagoya Port. The large differences in the maximum storm surge between both models were due to the wind setup effect. Figures 6c and 6d show the maximum storm surges and their occurrence times at Unoshima Port and Kushimoto Port, respectively. Unoshima Port is located in Suo Nada and faces north, toward the ocean (Fig.  1a). The maximum storm surges were lower when Typhoon Jebi passed west of Unoshima Port but became higher when Typhoon Jebi passed east of the port. At Kushimoto Port, the maximum storm surges from all ensemble experiments were less than 1.00 m. Because Kushimoto Port faces the deep ocean and has a depth greater than 100 m (not shown), the wind setup effect is not crucial. None of the typhoon tracks caused a large storm surge at Kushimoto Port. However, wave setup can be a key factor in storm surges on coasts such as Kushimoto Port, when high waves hit (e.g., Kohno et al. 2018;Washida et al. 2019). Because wave setup is a local phenomenon that only exerts a crucial effect in specific areas, as revealed by other studies of storm surges, its effects were not considered here. However, its effects are important when assessing the overall risk to the coast, especially including areas that face the open ocean, so it will be the subject of future research. Figure 9 shows maps of the maximum storm surges along the coastline of the main island, Shikoku, and Kyushu. According to simulations using the inputs from both models, maximum storm surges over 2.50 m could occur in the Ariake Sea, Suo Nada, Hiro shima Bay, Osaka Bay, Ise Bay, Mikawa Bay, Tokyo Bay, and Sendai Bay. The regions most at risk were open shallow bays less than 50 m in depth (not shown). At the coastal areas of Seto Inland Sea, maximum storm surges over 2.50 m occurred at some locations in Suo Nada and Hiroshima Bay, and surges over 1.50 m occurred at locations between Yamaguchi and Okayama Prefectures. Specifically, the maximum storm surge occurred in the innermost point of the bay. The largest maximum storm surge in Japan based on WRF-Jebi was 3.22 m in the coastal area of Osaka Bay, whereas that based on Para-Jebi was 2.94 m at Togari Port in the Ariake Sea. Figure 10 shows the maximum storm surges and worst-case courses in the coastal region from Izumo to Choshi (Fig. 1a) according to simulations using both model inputs. There were 1,896 sampling points in this coastal region. The maximum storm surges at most points estimated using WRF-Jebi inputs were generally larger than those estimated using Para-Jebi inputs. The average difference in maximum storm surges between both models was 0.37 m. The maximum difference of 1.19 m occurred at Nagoya Port near Point C (Fig. 1a). The difference between both models increased at points with larger maximum storm surges. In most coastal region, the models pre-dicted almost the same worst-case course. At coastal ports facing north, such as the Unoshima Port near Point I (Figs. 1a, 6a), the worst-case course was on the eastern side of the hit course. The differences in storm surges associated with the west-east courses of TCs depended on the ocean-facing direction of the bay.
This study conducted ensemble experiments to estimate areas at risk of a large storm surge. Table 4 shows the relative proportions of maximum storm surges and worst-case courses in each coastal region as shown in Fig. 1c. Note that 6,285 coastal points were considered at equal space intervals. Coastal areas where the maximum storm surge was produced by a track within 0.2° of the hit courses were defined as hit course areas. Coastal areas where the worst-case course was at least 0.4° east (west) of the hit course were defined as east (west) course areas. The percentage in Table 4 is defined by the ratio of the coastal points where the maximum storm surge exceeds  1.00 m/2.00 m to the number of coastal points in each coastal area.
For the main island, Kyushu, and Shikoku, 5.9 % of the maximum storm surges were more than 2.00 m in extent and 26.8 % were in the range of 1.00 -2.00 m. The proportion of maximum storm surges occurring on the main island was almost the same as the average for all areas. Although the maximum storm surge in Shikoku never exceeded 2.00 m, 45.8 % of the maximum storm surges were in the range of 1.00 -2.00 m, which was larger than the values for the main island and Kyushu. In Kyushu, 7.0 % of the maximum storm surges were more than 2.00 m in extent, which was the largest proportion among all areas studied. Approximately 55 % of the worst-case courses in the main island followed the hit course area, and approximately 35 % (10 %) followed the west (east) course area. In Shikoku, approximately 50 % of all worstcase courses followed the hit course area and the west course area, and there were no cases following the east course area. In Kyushu, approximately 65 % of all worst-case courses followed the hit course area and approximately 30 % (5 %) followed the west (east) course area.
At R1 and R2 on the main island, only 4.7 % of the maximum storm surges were more than 1.00 m in extent, indicating that high storm surges are rare. Notably, maximum storm surges of more than 1.00 m in extent at R6 on the main island were seen in 100 % of cases. The coastline at R6 is at high risk of storm surges, in part due to the presence of an open shallow bay. In Shikoku, there was a large difference between the north side (R7) and south side (R8). At R8, 10.5 % of the maximum storm surges were more than 1.00 m in extent, compared to 73.9 % at R7. In this study, Kyushu was divided into the west side (R9) and east side (R10). At R9, 51.1 % of maximum storm surges were more than 1.00 m, compared to 35.7 % at R10. Storm surges in Kyushu were more dangerous in the west coast than in the east coast. From the entire results, the largest maximum storm surge was in Osaka Bay, which is the R4 region. Osaka Bay would be attributed not only the open shallow bay but also the track of Typhoon Jebi. Because the direction of Typhoon Jebi movement around the Japanese Island is north-northeast and Osaka Bay opened to the southwest in almost parallel to the movement direction, the maximum storm surge became the largest. The values of the maximum storm surges estimated by T-TES for Typhoon Jebi may be slightly different from those for other typhoons, which will research other typhoon cases in the future. The results should provide useful information for disaster risk management.

Discussion and summary
This study estimated the maximum storm surges caused by Typhoon Jebi using the JMA storm surge model in conjunction with simulations based on the meteorological (WRF-Jebi) and parametric TC (Para-Jebi) models. The maximum storm surge at Osaka Port estimated using WRF-Jebi inputs (2.49 m) was closer to the observed storm surge (2.73 m) compared with that estimated using Para-Jebi inputs (2.11 m). We found that the difference between both models was caused by the wind setup effect rather than the inverse barometer effect. In the Typhoon Jebi case, the average difference in maximum storm surges between both models was 0.37 m, and the maximum difference of 1.19 m was found at Nagoya Port near Point C (Fig.  1a). The topography effect might have enhanced the surface winds over Osaka Bay; the parametric TC model could not account for this effect, whereas the meteorological model could, leading to an increase in storm surges due to the wind setup effect according to WRF-Jebi.
Our ensemble experiments based on the T-TES method indicated that the maximum storm surge varied with the longitudinal perturbation of the track of Typhoon Jebi along the entire coast of the Japanese Islands, including the main island, Kyushu, and Shikoku. The difference between both models increased at locations where the maximum storm surge was larger. The worst-case courses for each coastal area were almost the same track for both models.
Our ensemble experiments showed that during the passage of Typhoon Jebi, maximum storm surges of over 2.50 m occurred in coastal areas in the Ariake Sea, Suo Nada, Hiroshima Bay, Osaka Bay, Ise Bay, Mikawa Bay, Tokyo Bay, and Sendai Bay. These regions were typically shallow bays that were less than 50 m deep. In almost all coastal areas with large maximum storm surges, the worst-case course occurred 0.4 -0.8° west of the hit course because the wind setup exerted an important effect on the larger maximum storm surges. The distance between the worse-case and hit courses was consistent with the radius of the maximum wind of Typhoon Jebi.
This study estimated the coastal regions where large storm surges are possible. For the main island, Kyushu, and Shikoku, 5.9 % of the maximum storm surges were more than 2.00 m in extent and 26.8 % were in the range of 1.00 -2.00 m. For the coasts of the main island facing the Japanese sea, 1.5 % of the maximum storm surges were more than 1.00 m in extent; high storm surges were therefore, rare. In the coastal areas of Seto Inland Sea, there was a large risk of high storm surges due to the presence of an open shallow bay. Storm surges in Shikoku (Kyushu) were more dangerous on the north (west) than on the south (east) coast.
The exact reasons how the surrounding topography enhances/suppresses the surface winds over the Osaka Bay and Ise Bay remain unclear. Further, this study was based only on Typhoon Jebi; whether the results generalize to other typhoons are needed. However, our simulations pertained to storm surges and worst-case courses for almost the entire coastline of Japan. Because our simulations cover all areas, including where no significant storm surge by a typhoon is recorded, the results will be surely useful information for disaster risk management, and they provide important information for disaster risk management.

Data Availability Statement
The best-track datasets from Regional Specialized Meteorological Centers-Tokyo Typhoon Center are available at https://www.jma.go.jp/jma/jma-eng/jmacenter/rsmc-hp-pub-eg/RSMC_HP.htm?msclkid=e2335 b79cf5011ec857a83a520a1bbd5. The Japanese 55-year Reanalysis Project datasets are available at https://jra. kishou.go.jp/JRA-55/index_ja.html?msclkid=9147f3c 9cf5111ecb3842c9ffcc2fd49. The datasets from the Behavior of Hypoxia in Osaka Bay project are available at http://teiten.pa.kkr.mlit.go.jp/obweb/index.aspx. pressure differed among the ensemble simulations. The wind direction over the Japanese Islands changes because of the topography of the area. Therefore, the surface winds were simulated considering both the large-scale atmospheric field and topographic effects.