Tropical Cyclones Affecting Tokyo and Changing Storm Surge Hazard since 1980

This study investigated tidal records and landfall tropical cyclone (TC) best tracks in Japan from 1980 to 2019 to determine changes in storm surge heights in coastal regions of eastern Japan, including Tokyo. The results indicate that annual mean storm surge heights have increased in the last 20 years (2000–2019) compared to those in 1980–1999, and that these changes are noteworthy, particularly in Tokyo Bay. The storm surge hazard potential index (SSHPI), proposed by Islam et al. (2021), is positively correlated with surge height. The temporal change analysis of SSHPI suggests that TC wind intensity and size during landfall time frame have become stronger and larger, respectively, corresponding to increasing storm surge magnitudes from 1980 to 2019. The increased occurrence frequency of TCs with more northeastward tracks is another factor that may have contributed to the increased surge hazards around Tokyo. Tokyo area is likely to experience increasing numbers of extreme storm surge events in the future, if, the current increasing tendency continues.


Introduction
Tropical cyclone (TC)-generated storm surge is one of the most deadly and costly global hazards. In the last two decades, Hurricane Katrina (2005), Cyclone Sidr (2007), Cyclone Nargis (2008), and Typhoon Haiyan (2013) have each caused more than 1,000 fatalities, many of which were the direct result of storm surges (Esteban et al. 2015).
Although extreme storm surges do not occur often, the related damage and impacts can be enormous in densely populated urban areas. For example, Hurricane Sandy (2012) inflicted a catastrophic storm surge of ~2.7 m and resulted in $50 billion losses, particularly from storm surge along the mid-Atlantic coast, including the New York City metropolitan area (Blake et al. 2013). Typhoon Jebi in 2018 caused large-scale inundation in Osaka Bay, Japan, and the Kansai-Osaka International Airport was entirely inundated, which resulted in USD 2 billion in damages (Anh et al. 2019). Recent studies have suggested that the number of residents in coastal urban areas and the number of individuals exposed to flooding from once in 100-year storm surge events will continue to increase in the future (Nicholls and Cazenave 2010;Brown et al. 2013;Hallegatte et al. 2013;Neumann et al. 2015). Therefore, it is of great importance to understand whether any changes in storm surge hazards affecting large cities have occurred in recent years, despite the relatively infrequent occurrence of major storm surge events.
Analyzing the changes in storm surge trends in a certain location helps to assess whether coastal flooding risk reduction efforts are effective. Therefore, previous studies have analyzed changes in historical surge magnitudes in a specific location or globally. Several of these studies have concluded that surge levels are increasing over time. For example, Woodworth and Blackman (2004) and Menéndez and Woodworth (2010) analyzed storminduced abnormal sea-level changes by using a global tide gauge dataset. They concluded that many of the gauges showed significant increases in abnormal sea levels since 1970. Bromirski et al. (2003) found a significant increasing trend in extratropical surges in San Francisco since 1950. Church et al. (2006) observed an increase in the frequency of highwater levels (storm surges plus mean sea-level rise) after 1950 by using data from two tide gauges on the eastern and western Australian coasts. Although these studies have confirmed that mean sea-level rise, interannual, and decadal variations are the main factors contributing to the increase in maximum surge height, changes in TC characteristics such as intensity, track, size, and translation speed are also capable of modulating the surge hazard potential (Irish et al. 2008;Sebastian et al. 2014;Takagi 2020a, 2021).
Various indexes have been used to characterize TC activity, including annual storm counts, intensity, categories (Simpson 1974;Webster et al. 2005), and the accumulated cyclone energy (ACE) index (Bell et al. 2000). Some of the index-based analyses have exhibited excellent performances in explaining variations in TC activity trends on global and regional scales. For example, Emanuel (2005) showed that the TC power dissipation index (PDI; destructive potential of TCs) is positively correlated with the increase in sea surface temperature since the 1970s, where PDI is an aggregate of storm intensity, frequency, and duration. Emanuel (2005) also argued that PDI is a better indicator of TC threat than storm frequency or intensity alone. Utilizing the ACE index, similar to the PDI, Camargo and Sobel (2005) showed that there is a tendency in El Niño years toward TCs that are both more intense and longer-lived than in La Niña years in the Western North Pacific (WNP) basin from 1950 to 2002.
To improve the prediction accuracy of previous studies, Islam et al. (2021) statistically developed a storm surge hazard potential index (SSHPI) for estimating TCinduced maximum surge levels at a coast. SSHPI can provide an instantaneous measure to quantify the relative magnitude of a peak surge potential and advance TC measures (i.e., early warning, evacuation decision), akin in philosophy to the Saffir-Simpson Hurricane Wind Scale (SSHWS) (Saffir 1973;Simpson 1974). Although numerical surge modelling is the most sophisticated, advanced, and reliable method to forecast surge catastrophe, constraints related to TC forecast track information (i.e., landfall location and timing, approach angle, intensity, size) remain a major obstacle to a reliable operational surge forecast (Japan Meteorological Agency 2019) (hereafter Japan Meteorological Agency is abbreviated as JMA). As the degree of unusual characteristics of a TC increases, numerical surge forecast tends to underestimate the severity of a surge event (World Meteorological Organization 2011). Nevertheless, several practical solutions such as analytical, graphical (nomograms), empirical, and statistical approaches (i.e., SSHPI) can supplement numerical surge models (World Meteorological Organization 2011).
A retrospective analysis of SSHPI using data from 1978-2019 in Japan suggests that this index captures historical events reasonably well. The SSHPI has two major advantages: it explains the surge hazard potential for a TC that vary in size and speed and decreases numerical computational requirements. It also infers that a surge estimation derived from using only intensity information provides less information on the overall surge hazard than the full form of SSHPI (see Section 2 for the definition of SSHPI). Therefore, assessing whether other factors influence storm surge hazards over time at a specific location is also necessary. Specifically, this allows us to apply a surge index, i.e., SSHPI to explain temporal variations of surge hazards.
The WNP basin is the most active area on Earth for TCs, as approximately one-third of the world's TCs develop in the region (Elsner and Liu 2003). Tokyo is one of the largest and busiest cities in this region (United Nation 2018) where TC-induced storm surge poses a considerable threat to millions of coastal inhabitants (Islam et al. 2018). Swiss Re (2013) ranked 616 cities worldwide in terms of their potential risks for natural disasters, including TCs and storm surges, and reported that the Greater Tokyo Area is subject to the highest risks. Although Tokyo is prone to the effects of TCs, the TC landfall frequency around Tokyo is not very high compared to that of southwestern Japan (Islam and Takagi 2020a).
Nevertheless, strong TCs sometimes reach around Tokyo, causing billions of dollars in damage to the city's infrastructure (JMA 2020a). Among the major TCs that have directly affected Tokyo over the past century, a TC in 1917 (known locally as Typhoon Taisho) was the most devastating. The storm surge induced by this typhoon claimed 1,301 lives, destroyed 43,083 houses, and swept away 8,220 marine vessels (Omori 1918).
In modern times, although, most of the coastal areas of Japan including those of Tokyo are protected from storm surges by coastal dikes, the return period of their design high tide level has not been evaluated accurately because of short period of tidal observations (Torii and Kato 2004). In addition, it remains unclear whether overtopping of the dikes would lead to coastal flooding because current dikes did not perfectly resist overtopping against TC-induced high waves (Takagi and Takahashi 2021). Thus, there is an uncertainty regarding whether urban waterfronts of Tokyo and its neighboring cities are sufficiently resilient against future powerful TCs (Islam et al. 2018). Concerns about TCs and storm surge impacts have increased since Typhoon Faxai and Typhoon Hagibis hit the Greater Tokyo Area in 2019. These two typhoons caused severe coastal damage and were classified as the strongest typhoons since the JMA typhoon database was created in 1951 (Shimozono et al. 2020;Takagi et al. 2020). Yamaguchi and Maeda (2020a) showed that the number of TCs, including stronger storms approaching the southern coast of Japan, including Tokyo, has a decadal increasing trend since 1980. They also determined that the translation speeds of TCs have decreased from 1980 to 2019, which indicates that Tokyo and its surroundings experienced TC influences of a longer duration over time (Yamaguchi and Maeda 2020b). These recent studies and TC events raise questions as to whether the magnitude of the storm surge that has affected Tokyo and its neighboring cities has increased recently and whether changes in the storm surge hazard potential are larger than previously determined. Currently, there is no study that has examined the influence of surge hazard potential and long-term storm surge observations around Tokyo.
This study examines changes in storm surge hazards around Tokyo (see Section 2 for the definition of target area) since 1980 in terms of temporal variations in storm surge heights and analyzes the SSHPI to investigate factors that influence surge magnitudes. We also inspected changes in the TC tracks and their potential influence on storm surge variations. We used tidal records from ten tide stations and best-track information for TCs that made landfalls around Tokyo.

Data and Methods
In this study, we considered a period spanning from 1980 to 2019 and obtained TC best track data from the JMA archives, including data pertaining to TCs that originated in the WNP and made landfalls around Tokyo (JMA 2020a). The best track data acquired during the pre-satellite era (i.e., before 1980) contain heterogeneities and large uncertainties in data quality (Chan 2019;Moon et al. 2019) and therefore were ignored. Although the time period processed was relatively short, the period of 1980-2019 was the longest period covered by the JMA best track dataset that has uniform data quality. Data collection was limited to TCs that made landfalls within 300 km of Tokyo (latitude: 34°-37°N, longitude: 137°-141°E; Fig. 1) with a maximum sustained wind speed (Vmax) of greater than 17 m s -1 (33-kt). Although the primary focus of this study is to deal with storm surges in Tokyo, the justification for adopting a distance threshold is twofold. First, the JMA operationally considers any TC with a tropical storm intensity or stronger to be approaching Japan when the central position of the TC is within 300 km of any of the JMA meteorological offices (JMA 2020b). Second, the number of TCs (Vmax > 33-kt) that made landfall in Tokyo Bay (Fig. 2) from 1980 to 2019 is just eight (JMA 2020a), which is insufficient for conducting temporal change analysis of storm surge. Nonetheless, six of the top ten storm surges in Tokyo Bay were not due to direct TC landfalls in Tokyo Bay, but those that made landfall more than 50 km southwest of Tokyo Bay (Islam and Takagi 2020a). Therefore, the analyses presented in this study apply to Tokyo and its neighboring coastal areas (i.e., Hamamatsu, Shizuoka, Shimoda, Katsuura, and Choshi; Fig. 3). TCs are not always powerful (i.e., tropical depressions) and may not produce noticeable storm surge; thus, it would be more meaningful to focus on tropical storms (34-kt ≤ Vmax ≤63-kt) or stronger TCs (i.e., category 1, 2 in SSHWS) for disaster risk management. Based on these three criteria, 36 TCs were selected for analysis (Fig. 1). These TCs were further divided into two categories based on TCs impacted more than one tidal station located on the open coastlines and bay areas, respectively. As a result, there were 72 and 90 available storm surge cases for open coasts and bay areas, respectively, for a total of 162 cases. We analyzed best track TC central positions, intensities, sizes, and translation speeds at the landfall time frame (T) because storm surge tends to be amplified at the time of landfall. However, TC characteristics during the landfall time frame are not necessarily the most adverse condition to cause the largest storm surge on coasts. For example, storm surges would take the largest value when the TC track is closer to the bay. In this study, we considered TC landfall time rather than the time closest to each tide station located at open coasts and bays as a representative condition for causing peak surge because, (a) TC characteristics (i.e., intensity, size, translation speed), after landfall, differ from those over the ocean. Thus, the TC information over land is considered less reliable (Huang et al. 2021).
(b) Although the recent TC best track contains more detailed information about when a TC approaches the land, the JMA best track historically has provided TC information at every six-hour intervals. It is difficult to identify the time of closest approach to a station with the 6-hour interval data (c) In the current dataset, 26 of 36 TCs affected both open coasts and bays, and storm surge information was recorded for more than one tidal station for those TCs. Thus, a unique characteristic (i.e., landfalling TC intensity) for each storm can provide a simple basis for statistical analysis.   set-up is often the dominant driver for generating storm surge in some selected tidal stations (i.e., Omaezaki, Irouzaki, Mera); however, this influence was not considered, to keep the analysis simple. Sea surface anomalies were defined as the storm surge magnitude (in cm in this study) and were estimated by deducting the predicted astronomical tide from the observed storm tide.
The SSHPIs at T for all TCs were calculated to demonstrate how the factors influencing storm surge hazard potentials have changed over the last 40 years. Background information and the effectiveness of the SSHPI for predicting peak surge hazard potential were discussed in Islam et al. (2021). TC best track data were used to calculate the SSHPI for each storm. Table 1 shows the predictors used in the SSHPI. The SSHPI incorporates meteorological information, including TC intensity, size, and translation speed, along with coastal geometry (open coasts and bays) and regional scale bathymetry into a single measure of the expected surge hazard potential along the coast (Islam et al. 2021)  This suggests that large (i.e., R50 > 95 nm) and fast-moving (i.e., S > 35 km h -1 ) intense (i.e., Vmax > 50 kt) TCs will generate a larger storm surge along open coasts, whereas bays are more susceptible to large, slowly moving (i.e., S < 35 km h -1 ) intense TCs. Each parameter has a dimension, but the SSHPI is the product of a dimensionless quantity.

Observational evidence
The time series of annual mean storm surge heights is shown in Fig. 4a. A highly significant increase in storm surge height of +41% occurred per decade from 1980 to 2019.
Annual mean storm surges of 24 cm and 55 cm occurred near Tokyo in P1 and P2, respectively. The difference between the periods is statistically significant at the 95% level.
A similar increasing tendency (+40% per decade from 1980 to 2019) was also observed for the dataset comprising annual peak storm surge heights (Fig. 4b). At least one storm surge event with a peak surge height of ~100 cm occurred annually near Tokyo since 2011. In terms of the criteria considered in this study (see Section 2 for the definition of the TC landfall), P1 contained 11 years when no identifiable storm surge occurred near Tokyo (1980, 1982, 1984, 1986, 1987, 1988, 1992, 1994, 1995, 1996, and 1999). In contrast, P2 contained only four years without a storm surge occurrence (2003, 2006, 2008, and 2010).
Such differences in the TC landfall frequency appear to contribute to a larger storm surge tendency in P2 than in P1. However, a significant change was also evident in the years with storm surge events. To test the robustness of the statistics presented herein, particularly the outliers, the results were reaffirmed using the L 1 norm (not shown) in place of the ordinary least-squares (L 2 ) norm. None of the significant changes from the L 2 regression was insignificant. The distribution of storm surge events exhibited a clear shift towards greater heights during P2 and the differences were significant throughout most of the distribution (Fig. 5).
For example, there was a significantly higher probability of storm surges >95 cm during P2, and a significantly higher probability of storm surges <65 cm during P1. In other words, the occurrence frequency of significant storm surges (i.e., >95 cm) in the study area increased in P2 (eight times) than in P1 (two times).  (1980-1999) and P2 (2000-2019).
Error bars show two-sided 95% confidence intervals. Figure 6 shows the spatial distribution of storm surge statistics between P1 and P2 at the ten tidal stations. While an overall increase in storm surge height was evident at all stations, a statistically increasing tendency was more pronounced at four stations (Harumi, Chiba, Yokohama, and Mera) in Tokyo Bay. The magnitude of the change indicates that storm surge heights in Tokyo Bay increased by an average of +45% per decade, which is 4% larger than the overall increase (+41%) observed along the coasts of Tokyo and its'

Changes in TC parameters influence storm surge
This observation indicated that the number of landfalling TCs that fulfilled the criteria used in this study (n = 36; see Data and Methods for the definition of TC selection). The numbers of occurrence frequencies of TCs with Vmax > 60-kt during P1 and P2 were 0.15 and 0.75, respectively (Fig. 7a). Thus, there were five times more chances for a TC to affect Tokyo during P2 than during P1; thus, the frequency of strong TCs has increased over time. This finding is consistent with those of Yamaguchi and Maeda (2020a), who found that the frequency of strong TCs with a central pressure <980 hPa that approached the Tokyo area increased by 2.5 times. Figure 7b shows that the numbers of occurrence frequencies during P1 and P2 in the R50 ≤ 40 nm bin were 0.3 and 0.35, respectively. In contrast, TCs with R50 > 70 nm had five times more chance of affecting the Tokyo area during P2 compared with P1. No clear changes were observed in the occurrence frequencies of faster or slower TCs during the last 40 years (Fig. 7c). It needs to be noted that the statistics presented in Fig. 7 are based on low-frequency TC events (lower than one per year); therefore, the likelihood of overestimation or underestimation estimation is not negligible.

Changes in storm surge hazard potential
The purpose of this study was to determine whether there have been significant changes in the storm surge hazard around Tokyo in the last 40 years in terms of the storm surge magnitude and TC parameters that can influence the surge hazard. In particular, it is important to quantify the contribution of TC parameters on the increasing surge magnitude as shown in Section 3. Therefore, we applied SSHPI to demonstrate whether a similar increasing tendency is reproduced with the index and whether it can explain the possible contribution of each TC parameter in changing surge hazards. suggesting that SSHPI is strongly correlated with storm surge magnitude. This result supports the analysis by Islam et al. (2021), in which SSHPI was strongly correlated with actual surge heights observed throughout Japan. The SSHPI time series exhibited an evident increase of +63% per decade from 1980 to 2019. Annual mean SSHPIs of 0.21 and 0.71 were obtained for P1 and P2, respectively. The difference between these SSHPIs is statistically significant at the 95% level, indicating that the increase in SSHPI during P2 (more than three times) is unusual. A similar tendency was observed for both open coastlines and semi-enclosed bays (not shown). These findings are consistent with the observed storm surge statistics, as described in Results.   Figure 4. We assumed that only one of the SSHPI parameters changed between P1 and P2, and maintained the other parameters' value at their averages from P1 (Table 2). Changes in TC intensity and size contributed 35% and 32%, respectively, to the overall increase in surge height. Furthermore, their joint contribution may be responsible for as much as 79% of surge variance. Figure 9 shows similar findings by comparing surge height with the time series diagram of the surge index that reduces the predictors from the SSHPI into three patterns: (a) Vmax; (b) Vmax and R50; and (c) Vmax, R50, and S. Table 3 and Fig. 9 indicate that changes in TC intensity (Vmax) may have played the most significant role in the increase of storm surge magnitudes observed over the last 40 years, followed by changes in TC size (R50). This finding is consistent with the findings of Takagi (2020a, 2020c), who numerically showed that, at a constant TC wind intensity, storm surge heights in Tokyo Bay increased linearly as TC size increased.
It is possible because a large swath of strong winds can affect a larger ocean area for a longer period of time, inducing larger storm surges.

Changes in TC track
We also inferred the reasons for the increase in storm surge magnitude by analyzing the differences in the TC tracks between P1 and P2. Six-hourly TC best track positions, along with intensity information of landfalling TCs, were analyzed at each grid point from 34° to 37°N and from 137° to 141°E,at 1° intervals (Figs. 10a,10b). The ratios of the changes in landfalling TC frequency for each grid point (i.e., the ratio of the number of TC central positions in P2 to P1) are shown in Fig. 10c. An overall increase in TC activity is evident in the grids covering Tokyo and its surroundings. One remarkable feature is a statistically significant increase in TC activity close to Tokyo Bay (near 34°-36°N, 139°-140°E) during P2 than in P1. In other words, the occurrence frequency of more northeastward TC tracks that impacted Tokyo's shallow coastal areas (Fig. 3b) increased during P2 than in P1 (Fig.   10c). A significant number of TCs with greater intensities (i.e., Vmax ≥ 64 kt) evidently impacted Tokyo Bay during P2, and such TCs occurred at low-frequencies during P1 (Fig.   10a, b). Although the storm surge increasing tendency is independent of TC frequency (discussed in Section 3), changes in the occurrence frequency of more northeastward TC tracks can be regarded as the third most influential factor after Vmax and R50. Therefore, the large increase in surge magnitude, particularly inside Tokyo Bay (Fig. 6) over the past 40 years, might be attributed to the combined influence of TC intensity, size, and track change.

Summary and Conclusions
Based on observational data, the present study demonstrates that storm surge heights in eastern Japan, including those in Tokyo, have increased significantly from 1980 to 2019. The SSHPI from 1980 to 2019 increased more than three times from P1 to P2. It also suggests that the changes in both TC intensity and size at the time of landfall played a significant role in increasing storm surge magnitudes. This finding is consistent with the observation where it is shown that TCs have become stronger and larger, respectively during P2 than in P1. The increased occurrence frequency of TCs with more northeastward tracks contributed to the increased surge hazard around Tokyo.
This study suggests that the Tokyo area is likely to experience increasing numbers of extreme storm surge events in the future, if the current increasing tendency continues.
The coastal flooding of Tokyo could have affected the Japanese economy substantially because much of the Japanese GDP is concentrated in the Tokyo area. Thus, our findings are valuable for disaster risk managers, coastal engineers, climate change scientists, and governmental bodies attempting to mitigate storm surge risk in coastal urban areas. The SSHPI can help explain variations in surge events on regional and local scales, such as Tokyo area targeted in this study.
The analyses presented in this study do not constitute the detection or contributions of global warming and climate change impacts on an increasing storm surge tendency.
Although the difference between P1 and P2 could be regarded as a trend associated with global warming, several recent studies have projected with high confidence that sea-level rises that accompany warming will lead to higher storm inundation levels (i.e., Knutson et al. 2019;Knutson et al. 2020;Mori et al. 2021). However, Yamaguchi and Maeda (2020a) reported that this difference may not be distinguishable in the interdecadal variability of such a short period (40 years Particularly, we have discussed the possible contribution of TC wind speed, size, and track in changing surge hazards around Tokyo; nonetheless, other factors such as air pressure, wave set-up can also modulate surge hazards. The physical meaning of these influential factors remains unknown. The present analyses can be improved further by improving the current data quality, in time and spatial coverage.

Acknowledgments
The first author is thankful to the Ministry of Education, Culture, Sports, Science, and Technology (MEXT) of Japan for the provided scholarship to conduct research in the field of disaster risk reduction. This study was conducted while the first author was participating in the internship at Atmosphere and Ocean Research Institute, The University of Tokyo.
Authors are grateful to Dr. Munehiko Yamaguchi, Japan Meteorological Agency, for providing the approached TC number information for Tokyo area .