Future changes in typhoons and storm surges along the Pacific coast in Japan: proposal of an empirical pseudo-global-warming downscaling

ABSTRACT

In this study, dynamical pseudo-global-warming downscaling (d-PGWD) was performed with a high-resolution typhoon model for 49 typhoons that made landfall in Japan between 2000 and 2017. It was revealed that the averaged typhoon intensity under future climatic conditions tends to increase at both the peak and landfall times as a result of global warming (averaged central pressures of −45.7 and −5.5 hPa at peak and landfall, respectively). Furthermore, detailed analyses of the time of landfall revealed significant differences in the degree of future changes in typhoon intensity based on both the elapsed time from the genesis to landfall (T_l) and the radius of maximum wind speed (R_ml) at the time of landfall. Considering the relationships of T_l and R_ml between present and future climates, statistical formulas for future changes in the central pressure and R_ml were derived as an empirical PGWD (e-PGWD) method. The validity of this method was confirmed via comparison with d-PGWD results. It is expected that disaster prevention and mitigation measures for future typhoons and coastal disasters in individual regions and ports can be developed by revising storm surge hazard maps using the proposed e-PGWD approach.

KEYWORDS:

• Global warming
• typhoon
• storm surge
• empirical pseudo-global-warming downscaling

1. Introduction

According to the Intergovernmental Panel on Climate Change (IPCC) Fifth Assessment Report (IPCC 2013a), the characteristics (e.g. intensity and frequency) of tropical cyclones (i.e. typhoons, hurricanes, and cyclones) under future climatic conditions are likely to change from those observed in the present climate, and global warming will likely reduce the frequency of tropical cyclones throughout the world. Furthermore, the frequency and intensity of extreme tropical cyclones is likely to increase as a result of increasing air temperature (AT), sea surface temperature (SST), and ocean heat storage (Murakami et al. 2012; Knutson et al. 2015). Based on the numerical simulation performed using the Meteorological Research Institute Atmospheric General Circulation Model (MRI-AGCM), (Murakami et al. 2012) conducted a future-climate experiment for the period of 2075–2099 assuming the Special Report on Emission Scenarios (SRES) A1B scenario in Coupled Model Intercomparison Project, Phase 3 (CMIP3). The MRI-AGCM indicated that the frequency of tropical cyclones will decrease by 13%–25% globally, whereas the intensity will increase by 1%–7%. Knutson et al. (2015) performed dynamical downscaling for tropical cyclones that occurred globally between 1980 and 2008 using the Geophysical Fluid Dynamics Laboratory (GFDL) hurricane model with 13 general circulation models (GCMs) under the Representative Concentration Pathway 4.5 (RCP4.5) scenario in Coupled Model Intercomparison Project, Phase 5 (CMIP5). The GFDL hurricane model predicted that, at the end of the 21st century, the tropical-cyclone frequency will decrease by 16% whereas the intensity will increase, thereby leading to a 24% increase in tropical cyclones reaching category 4 or 5. However, it has been pointed out that the tendency differs depending on ocean basins, and therefore, the impact of global warming on typhoons, hurricanes, or cyclones cannot be discussed together.

The ongoing global warming is already affecting typhoon characteristics (Webster et al. 2005). One of the most prominent examples of recent typhoons affected by global warming is Typhoon Haiyan (2013), which struck Central Philippines in November 2013. Haiyan caused a storm surge exceeding 6 m in Tacloban located behind the Bay of San Pedro, and there was an unprecedented catastrophe that resulted in total damages of approximately 571.1 billion pesos (US $12.9 billion), 6201 deaths, and 1785 missing persons (NDRRMC 2014). Takayabu et al. (2015) employed the Weather Research and Forecasting (WRF) model as a regional climate model (RCM) to evaluate the future change of Haiyan based on dynamical pseudo-global-warming downscaling (d-PGWD) experiments. d-PGWD is well known simplified method for assessing the effects of global warming using monthly mean differences between the future and present climates derived from CO₂-warmed GCMs (Sato, Kimura, and Kitoh 2007). As the SST around the Philippines increased by 0.2–0.8 K because of global warming, Haiyan would be more intense under future climatic conditions, and the associated storm surge increased by approximately 20%. The storm surge disaster caused by Haiyan is now considered one of the worst typhoon disasters affected by global warming. For other typhoon cases, Yoshino et al. (2015) performed d-PGWD experiments for all typhoons that landed in 2004 targeting the end of the 21st century under the SRES A2/A1B/B1 scenarios, and they evaluated the differences in global warming effects on typhoon intensity among different scenarios. They clarified that typhoon intensities at peak times tend to increase on average in proportion to the degree of global warming. For the SRES A2 scenario, in which there is significant warming, “weak” and “strong” typhoons in the present climate are expected to strengthen considerably in the future. In contrast, the intensification of “furious” typhoons in the present climate, which develop to reach the potential intensity (Emanuel 1988), will be suppressed by stratification stabilization in the troposphere. Tsuboki et al. (2015) conducted dynamical direct downscaling (d-DD) experiments on 30 strong typhoons that landed in the period of 1979–1993 assuming the SRES A1B scenario using an RCM. According to those simulations, the lowest central pressure at the end of the 21st century was obtained as 857 hPa. Kanada et al. (2017) also performed d-PGWD experiments on Typhoon Vera (1959), which caused the worst damage (over 5000 deaths and missing persons in total) by a storm surge around Ise Bay, Japan, considering future global warming at the end of the 21st century under the RCP8.5 scenario. They found that future global warming alters the inner-core structure (e.g. eye-wall heights and upward flows) as well as the intensity of the typhoon. The mid-latitude SST near the Japan Islands is expected to rise in the future climate, and typhoons stronger than the present climate will make landfall while maintaining their intensity.

Based on statistics reported by the Japan Meteorological Agency (JMA), Japan is the third-most typhoon-prone country in Asia, with 11.5 typhoons approaching and 2.9 directly impacting the main Island each year (JMA 2018). Therefore, accurate assessments of the impact of global warming on landfalling typhoons and their associated coastal disasters are necessary when establishing disaster prevention and mitigation measures for mid- to long-term expected storm surges. Recently, various studies have been conducted to assess future changes in the typhoon-associated storm surges in Japan under the future climate (Yasuda et al. 2014; Mori and Takemi 2016). Yasuda et al. (2014) investigated future typhoon intensity at the landfall time to Japan using a CO₂-warmed GCM at the end of the 21st century under the SRES A1B scenario. They reported that the possibility of intense typhoons with central atmospheric pressure lower than 930 hPa will increase by a few percentages in the future climate, and the maximum storm surge will increase by 0.5–1.0 m in Ise Bay. Mori and Takemi (2016) evaluated future changes in the typhoon intensity of Typhoon Vera (Takemi, Ito, and Arakawa 2016) using d-PGWD assuming the RCP8.5 scenario at the end of 21st century. In their work, calculations were performed for five cases, with the calculation start time being altered by 12 h intervals. They reported that the averaged central pressure of typhoons in the future climate projections tended to decrease by −14.1 hPa at the peak time, and the resulting storm surge tended to increase by 1.0–1.3 m. Thus, they concluded that strengthened typhoons under the future climate may cause more severe wind-related and coastal disasters such as high waves and storm surges along the Pacific coast in Japan.

In storm surge inundation estimations performed by local governments in Japan, an empirical typhoon model (e.g. Mitsuta and Fujii 1987) is used for calculations of the typhoon meteorological fields because of its low calculation costs (Aichi Prefecture 2014; Tokyo Metropolitan Government 2018). The empirical typhoon model can estimate the internal structure of a typhoon using a small number of typhoon attribute values. Most typhoon attribute values (e.g. center position, central pressure, movement speed, and maximum wind speed radius) required for the empirical typhoon model can be obtained from observations, such as JMA Best Track, or simulations, such as low-resolution GCMs. The accuracy of the meteorological fields in a typhoon can be improved by using observations as input parameters. However, the problem is that the effects of global warming in the future climate never be considered. Although the effects of global warming can be considered using GCMs as input parameters, the intensity and structure of typhoons cannot be well represented because of their low resolution. Therefore, local governments are forced to estimate potential storm surges using observations of the central pressure at landfall (e.g. 911.6 hPa for Typhoon Muroto (1961)) and the maximum wind speed radius at landfall (e.g. 75 km for Typhoon Vera (1959)) of prominent typhoons that have occurred in the past. Although the use of such a historical record incurs a considerably low computational cost, there is a concern that it is unlikely to reflect the effects of global warming in the future climate adequately, thereby leading to the unexpected errors.

In this study, d-PGWD experiments were conducted using a high-resolution typhoon model (HTM) for a total of 49 typhoons that made landfall in Japan between 2000 and 2017. The target typhoons have various characteristics. The HTM is an RCM modified for typhoon intensity predictions (Yoshino et al. 2015). The d-PGWD method employed in this study assumes a future climate at the end of the 21st century (2080–2099) under the RCP 8.5 scenario. This study aims to construct a high-resolution typhoon database under the present and future climates, and to understand the characteristics of future changes in the intensities (central pressures) of typhoons that will make landfall at the Pacific coast in Japan. In addition, another aim of this study was to develop an empirical PGWD (e-PGWD) method for port managers and practitioners, based on a high-resolution typhoon database. e-PGWD is an effective method that enables future climate experiments (at the end of 21st century, under the RCP8.5 scenario) without the need to use items affected by climate change (e.g. AT and SST). Therefore, we aimed to develop a low-cost method that can be used even by non-experts. Many studies relating to climate change impact assessments typically focus on “the end of the 21st century in RCP8.5,” which is of most interest to policy makers discussing future global warming adaptation. Therefore, this study also focuses on the end of the 21st century in RCP8.5. It is inferred that the results would differ for different emission scenarios and different ages. There are two main approaches for calculating future typhoon intensity. The first approach, d-PGWD, is a dynamical method of inputting future parameters affected by climate change, such as SST and AT, and performing calculations with a numerical model. The second approach, e-PGWD, is an empirical method of estimating the future typhoon intensity from the information of values under the present climate.

The remainder of this paper is organized as follows. Section 2 describes the computational method along with the details of the HTM and d-PGWD method used in this study. The characteristics of typhoons under the present and future climates obtained from the high-resolution typhoon database during the period 2000–2017 are discussed in Section 3. In Section 4, e-PGWD on storm surge under the future climate considering future changes in the typhoon intensity at landfall in Japan estimated by the high-resolution typhoon database is proposed. By applying the e-PGWD method to the three typhoons that made landfall in 2018, future storm surges that can occur at Ise Bay and Osaka Bay under the future climate are evaluated and verified via d-PGWD experiments. Finally, in Section 5, we conclude the paper by summarizing the findings obtained through this study.

2. Computational method

2.1. High-resolution typhoon model

The HTM is a type of RCM used in this study for reproducing intensity changes from the genesis to the dissipation stages for target typhoons (Yoshino, Strachan, and Vidale 2012; Yoshino et al. 2015). The HTM is based on the Pennsylvania State University (PSU)/National Center for Atmospheric Research (NCAR) The Fifth-Generation Mesoscale Model (MM5) (Dudhia 1993), and it is a three-dimensional, nonhydrostatic, fully compressible, cloud-resolving atmospheric model used to capture mesoscale and local-scale meteorological phenomena.

The automatic movable nesting technique is introduced to MM5 to reproduce the inner-core structure near the center of a powerful typhoon, from the genesis to dissipation stages, efficiently. In addition, several types of physical parameterizations are incorporated into the original MM5. In particular, the ocean mixed layer (Shade 1999; Emanuel et al. 2004), dissipative heating (Bister et al. 1998; Zhang et al. 1999), and sea-spray processes (Fairall, Kepert, and Holland 1994; Wang, Kepert, and Holland 2001) are implemented to express realistic typhoon intensity and structural characteristics accurately. The ocean mixed layer process is introduced to express the vertical turbulent mixing that occurs in the ocean mixed layer just below the strong winds in typhoons, which can lead to a large SST drop for a slow-moving typhoon. The climatological mean (monthly average) of the ocean mixed layer thickness provided by the National Oceanographic Data Center of the National Oceanic and Atmospheric Administration (NOAA; Monterey et al. 1997) is used as an initial condition of the ocean mixed layer process. Next, the dissipative heating process, which typically occurs in the typhoon boundary layer, is introduced to reflect the effect of diabatic heating caused by viscous dissipation under the strong winds in typhoons. Finally, the sea-spray process that performs the lower atmosphere cooling and wetting because of sea-spray evaporations is introduced in addition to evaporation from the flat ocean surface just below the typhoon.

Previous studies (Yoshino et al. 2015; Toyoda, Yoshino, and Kobayashi 2015, 2016, 2017) have confirmed the high accuracy of HTM ranging from a weak typhoon to a powerful typhoon. Yoshino et al. (2015) performed reproduction experiments for all typhoons in 2004 with a 3 km grid spacing for automatic movable nesting. In their study, the root mean square error of the minimum typhoon pressure was 15.1 hPa, and the correlation coefficient was as high as 0.78. Toyoda, Yoshino, and Kobayashi (2015) conducted a reproduction experiment for Typhoon Haiyan (2013) and its storm surge. The observed minimum central pressure was 895 hPa, whereas the calculation result was 897.1 hPa; they succeeded in reproducing a furious intensity below 900 hPa by adding a 3 km grid spacing in addition to the 9 km grid spacing for automatic movable nesting. In this study, triply nested computational domains D1, D2, and D3 are also employed, and they have horizontal grid spacings of 27, 9, and 3 km, respectively. The movable nesting technique is applied to D2 and D3 to represent the inner-core structure and its intensity from the genesis to the dissipation stage with high accuracy (Figure 1). A four-dimensional data assimilation technique (so-called “analysis nudging”) is utilized for only D1, to assimilate the gridded analyses into the typhoon environmental fields gradually. Typhoon tracks are reproduced with high accuracy by applying analysis nudging; however, analysis nudging is not applied to D2 and D3 because the internal structure and intensity of the typhoon can be expressed only by dynamical and physical processes in the HTM. The analysis nudging, which relaxes the model state toward gridded analysis data, is applied to D1 in the high-resolution typhoon model used in this study. Therefore, the chaotic evolution of initial perturbations seen in ordinary forecasting experiments should be suppressed. Rankine’s vortex, with a wind speed 17.2 m/s, was installed as a TC bogus at the initial position of each typhoon. The other calculation settings are summarized in Table 1.

Table 1. Computational settings of high-resolution typhoon model (HTM)

Domain	Domain 1 (D1)	Domain 2 (D2)	Domain 3 (D3)
Horizontal resolution	27 km	9 km	3 km
Horizontal grids	298 × 349	91 × 91	91 × 91
Time step	90 s	30 s	10 s
Vertical resolution	24 layers (1000–70 hPa)
Initialization	NCEP Final Analyses (1°×1°) each 6 hours	D1 (27 km)	D2 (9 km)
Movable nest	Off	On	On
Typhoon bogus	Rankin’s vortex (wind speed 17.2 m/s)	Off	Off
Nudging (Four dimensional data assimilation)	On	Off	Off
Cumulus convection scheme	Grell (Grell et al .1993)	Off	Off
Cloud microphysics scheme	Reisner graupel (Reisner et al. 1998)
Planetary boundary layer scheme	Mellor – Yamada Level 2.5 Eta PBL (Mellor and Yamada 1982)
Radiation scheme	Cloud radiation (Dudhia 1989)
Land surface scheme	Five layer soil (Dudhia 1996)
Ocean mixed layer scheme	Shade and Emanuel (1999)The climate value of the ocean mixed layer thickness provided by NODCis used as the initial condition
Sea spray scheme	Fairall, Kepert, and Holland (1994)
Dissipative heating scheme	Jin et al. (2007)

Figure 1. Computational domain for the high-resolution typhoon model (HTM) with the sample track of Typhoon CHABA (2004). Black, blue, and green squares are D1, D2, and D3 domains, respectively. The figures on the top right show the horizontal distributions of P_c calculated in D2 and D3 (contoured at every 20 hPa). The figures on the bottom right show the horizontal distributions of wind speed at a 10 m height calculated in D2 and D3 (contoured at every 10 m/s).

2.2. Present-climate and future-climate experiments

In this study, numerical experiments for typhoons under the present and future climates are performed using the above HTM according to the calculation flow shown in Figure 2. First, as present-climate experiments, reproduction experiments are performed to confirm the reproducibility of landfalling typhoons in the present climate. The parameters used for the initial and boundary conditions are temperature, geopotential height, east-west wind velocity, north-south wind velocity, relative humidity, sea-level pressure, and skin temperature obtained from the National Centers for Environmental Prediction (NCEP) Final Analyses (FNL). According to the JMA’s record (JMA 2018), 53 typhoons during the period 2000–2017 that made landfall in Japan are selected as the typhoon cases targeted in this study. However, three typhoons – Typhoons Megi (2004), Dianmu (2010), and Malou (2010) – are eliminated because they show irregular tracks making landfall on the Tohoku region from the Sea of Japan. In addition, Typhoon Chataan (2002) is also removed from the discussion because it is difficult to calculate its entire lifetime using the HTM. Note that Typhoon Chataan is a case where the typhoon structure at the time of genesis could not be properly captured in the NCEP FNL, used as the initial conditions. Therefore, a total of 49 cases are considered in the present-climate experiments (Figure 3). Further, the initial time for all experiments was set at a time when the maximum wind speed exceeded approximately 17.2 m/s, which is defined as “typhoon” in the JMA Best Track data (). The end time for the all experiments is set at a time when the “typhoon” changed to a “tropical depression” or “extratropical cyclone” in the JMA Best Track data.

Figure 2. Computational flow for the pseudo-global warming downscaling experiment (PGWD) method. The left, middle, and right columns show the flows of the present-climate experiment, future-climate experiment, and horizontal resolution at each calculation step, respectively.

Figure 3. Observed tracks of target typhoons from genesis to decay, according to the JMA best track.

Furthermore, future-climate experiments for these 49 cases are performed by applying the d-PGWD method, and future changes are estimated by comparing the results with those of the present-climate experiments. The d-PGWD method is a type of dynamical downscaling technique that considers the expected climate change (Kimura and Kitoh 2007; Sato, Kimura, and Kitoh 2007; Hara et al. 2008; Kawase et al. 2009; Yoshino et al. 2015). The d-PGWD method was first used in Japan, and similar methods have been used in other countries (Knutson et al. 2008). This approach is different from the d-DD method, which uses the data obtained from GCMs directly; instead, the d-PGWD method uses the differences in monthly mean fields between the present and future climates, averaged from GCMs. Hence, the main advantage is that it avoids serious bias errors in GCMs, which are caused by biases in the initial and boundary conditions (Kimura and Kitoh 2007). Biases in GCMs, used to drive an RCM, are one of the largest uncertainties in dynamical downscaling (Sato, Kimura, and Kitoh 2007). As the d-PGWD method is based on specific weather events that occurred in the past, one-to-one relationships between the present and future climate experiments can be drawn easily. Furthermore, as data handling is easier for the d-PGWD method than for the d-DD method, the ensemble average fields obtained from multiple GCMs can be inputted as the monthly mean differences and the variability of the monthly mean differences in each GCM can be reduced (Sato 2010). Hara et al. (2008) verified the d-PGWD approach using historical data, and the precipitation, snow depth, and surface air temperature of the hindcast experiments agreed well with the observation data. Kawase et al. (2009) applied the d-PGWD approach to estimate the change in the Baiu rain band caused by global warming. The results demonstrated that the d-PGWD run has a good similarity to the average of the d-DD runs by seven GCMs. Their results indicated that the d-PGWD can calculate consistent results with d-DD, using the multi selected GCM ensemble mean, thereby reducing the significant computational cost. Next, the limits of d-PGWD are described.

Although the effect of a future change in the mean fields can be assessed by d-PGWD, the effect of a future change in the anomalous fields cannot be assessed because of the limitations of d-PGWD. In this method, an effect of global warming is considered only in a 20-year time-averaged change component, and its anomalous component is assumed to be the same as the present climate. Essentially, it is assumed that the anomalous fields in the future climate will be fixed to the state under the present climate. Therefore, the typhoon tracks in the future-climate experiments, which might be affected by the anomalous fields, are assumed to be the same as those of the present-climate experiments. The calculation end time of the future climate experiments is also set to the same time as the present climate experiments.

The initial, boundary, and assimilation conditions for the future-climate experiments are established by adding the differences of monthly mean fields (∆G) obtained by GCMs to those for the present-climate experiments obtained by NCEP FNL (A_f in Figure 2). The differences of monthly mean fields ∆G are defined as the differences of the 20-year average fields in the 2-D surface (i.e. sea surface temperature and sea level pressure) and the 3-D atmospheric variables (i.e. SST, sea level pressure, temperature, geopotential height, east-west wind speed, north-south wind speed, and relative humidity from 1000 to 70 hPa) between the future climate (2080–2099) and the present climate (2000–2019). The ocean mixing layer depth is fixed at the same conditions as the present climate without considering future changes, because many models lack public access to the data of the mixed layer depth in both climates. Here, the maximum value of relative humidity is fixed at 100%. When the relative humidity exceeds 100% as a result of adding a global warming difference, the specific humidity field is computed by the Clausius–Clapeyron equation in which the “saturated” specific humidity depends on temperature and pressure. In this study, the future climate fields are assumed as the ensemble monthly mean fields averaged during the period of 2080–2099 by eight GCMs for the RCP8.5 scenario in CMIP5. There are systematic errors across the eight GCMs used in this study. Therefore, ensemble-averaged data is used for the reduction of these errors. RCP8.5 is a high-level reference scenario that assumes there is no mitigation policy (Moss et al. 2010). The present climate fields are defined as the ensemble monthly mean fields averaged during the period of 2000–2019 by 15 GCMs for the historical scenario (2000–2005) and eight GCMs for the RCP8.5 (2006–2019), as listed in Table 2. The GCMs are selected because they contain all necessary variables at the necessary altitude during the necessary period. In order to estimate the global warming differences for PGWDs, the future-climate fields are calculated from the monthly mean fields of “8 future-climate (RCP8.5) GCMs for the period 2080–2099.” Then, the present-climate fields are also derived from the monthly mean fields of “8 future-climate (RCP8.5) GCMs for the period 2006–2019.” This means that even the future-climate GCMs cover a major part of the “present-climate” period 2000–2019, which is defined by this study. Unfortunately, the 8 future-climate GCMs in CMIP5 do not include any data for the period 2000–2005. To adjust to the 2000–2019 period for the typhoons targeted in this study, it is reasonable to use the present-climate fields modified by the 15 historical GCMs in CMIP5 for the period 2000–2005 (Figure 4). As the modification is performed by weighted averaging according to the length of the data periods (6 years for historical GCMs and 14 years for RCP8.5 GCMs), the procedure should not significantly change the monthly mean fields obtained by the 8 future-climate GCMs and should not pose major problems. Here, we check the continuity between ensemble mean fields obtained from GCMs under different scenarios. Figure 5a and c show the monthly mean fields of the air temperature at 300 hPa and sea surface temperature for the period 2000–2005, whilst Figure 5b and d show the same information for the period 2006–2019. Figure 5e and f show the differences of the monthly means fields, both of which are not significantly different. Therefore, we did not determine any serious problems in the continuity between the two different data sets. Using the initial, boundary and assimilation conditions added by ∆G, the future-climate experiments are conducted for a total of 49 cases, where the other computational settings are identical to the present climate experiments.

Table 2. List of general circulation models (GCMs) used for pseudo-global warming downscaling experiments (PGWDs). The (a) historical and (b) RCP8.5 data were obtained from 15 and 8 GCMs, respectively, in the Coupled Model Intercomparison Project 5 (CMIP5) (IPCC, 2013). These GCMs shown in table (a) are used for creating that data period of 2000–2005 from the historical of CMIP5, the GCMs in table (b) are used for creating that data period of 2006–2019 and 2080–2099. Global warming difference in present-climate is created by weighted averaging according to the length of the data periods (6 years for historical GCMs and 14 years for future GCMs)

(a)
	Resolution (deg)	Institute
CMCC-CESM	3.4 x 3.8	Centro Euro-Mediterraneo per I Cambiamenti Climatici
CNRM–CM5–2	1.4 x 1.4	Center National de Rcherches Meteorologiques / Center Europeen de Recherche et Formation Avancees en Calcul Scientifique
CNRM-CM5	1.4 x 1.4	Center National de Rcherches Meteorologiques / Center Europeen de Recherche et Formation Avancees en Calcul Scientifique
HadGEM2-CC	1.3 x 1.9	Met Office Hadley Center
HadGEM2-ES	1.3 x 1.9	Met Office Hadley Center
IPSL-CM5A-LR	1.9 x 3.8	Institut Pierre-Simon Laplace
IPSL-CM5A-MR	1.3 x 2.5	Institut Pierre-Simon Laplace
IPSL-CM5B-LR	1.9 x 3.8	Institut Pierre-Simon Laplace
MIROC-ESM-CHEM	2.8 x 2.8	Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies
MIROC-ESM	2.8 x 2.8	Japan Agency for Marine-Earth Science and Technology, Atmosphere and Ocean Research Institute (The University of Tokyo), and National Institute for Environmental Studies
MPI-ESM-LR	1.9 x 1.9	Max Planck Institute for Meteorology (MPI-M)
MRI-CGCM3	1.1 x 1.1	Meteorological Research Institute
NorESM1-M	1.9 x 2.5	Norwegian Climate Center
NorESM1-ME	1.9 x 2.5	Norwegian Climate Center
INM-CM4	1.5 x 2.0	Institute for Numerical Mathematics
(b)
	Resolution (deg)	Institute
CanESM2	2.8 x 2.8	Canadian Center for Climate Modeling and Analysis
CMCC-CESM	3.4 x 3.8	Centro Euro-Mediterraneo per I Cambiamenti Climatici
HadGEM2-CC	1.3 x 1.9	Met Office Hadley Center
IPSL-CM5A-LR	1.9 x 3.8	Institut Pierre-Simon Laplace
IPSL-CM5A-MR	1.3 x 2.5	Institut Pierre-Simon Laplace
IPSL-CM5B-LR	1.9 x 3.8	Institut Pierre-Simon Laplace
NorESM1-M	1.9 x 2.5	Norwegian Climate Center
NorESM1-ME	1.9 x 2.5	Norwegian Climate Center

Figure 4. Differences in the time periods covered by GCMs in CMIP5 and PGWDs in this study, and their respective definitions of “present climate” and “future climate” (modified from Figure SPM. 7 in IPCC AR5 WG1 (Intergovernmental Panel on Climate Change, Fifth Assessment Report, Working Group I. 2013. “Summary for policymakers.” in Climate Change 2013: The Physical Science Basis, edited by the Intergovernmental Panel on Climate Change. https://www.ipcc.ch/report/ar5/wg1/summary-for-policymakers/). This study defines 2000–2019 as the present climate and 2080–2099 as the future climate. As the defined period of this study is different from that of CMIP5 GCMs, it is adjusted when calculating the global warming difference (IPCC, 2013b).

Figure 5. Distribution for continuity among GCM data used in this study. Ensemble mean distribution of temperature at 300 hPa for (a) 2000–2005 and (b) 2006–2019. The contours are drawn every 2.5 K. Ensemble mean distribution of SST for (c) 2000–2005 and (d) 2006–2019. The contours are drawn every 2.5 K. Differences in ensemble mean of (e) air temperature at 300 hPa and (f) SST from 15 GCMs in 2000–2005 and 8 GCMs in 2006–2019. The contours are drawn every 1.0 K (air temperature) and 0.5 K (SST).

2.3. Analysis method for present- and future-climate experiments

Future changes in the intensities of typhoons making landfall in Japan are analyzed using a high-resolution typhoon database (D3 in HTM recorded every 15 min) constructed using the results of the present-climate and future-climate experiments described in the previous section. In particular, the central pressure and maximum wind speed radius are the focus of this analysis. Central pressure is widely used as an index of typhoon intensity. However, the radius of maximum wind speed is a meteorological parameter required for storm surge estimation and is an important parameter for evaluating future changes in typhoon characteristics. The minimum value point of the sea level pressure is defined as the center of a typhoon. The central sea level pressure at the center of the typhoon is determined as P_c, and the distance from the center of typhoon to the point at which the maximum wind speed is at a 10-m altitude is regarded as R_m. However, the output values of P_c and R_m considerably fluctuate over time, and therefore, we analyze the values averaged over a 6-h period, which is the same as the time interval of the JMA Best Track.

To confirm the reproducibility of typhoons in the present climate experiments, the accuracies of P_c and R_m for all 49 typhoons were verified using the available observations. For the verification of P_c, the values obtained from JMA Best Track data are used. In contrast, R_m is not officially recorded in the JMA Best Track data, and therefore, this parameter is indirectly estimated using the Myers equation (Myers 1954)

(1) $P\left(r\right)=P_c+\mathrm{\Delta }Pexp\left(-\frac{R_m}{r}\right)$(1)

where P (r) is the atmospheric pressure (hPa) at a point r km from the typhoon center, P_c is the typhoon central pressure (hPa), ∆P is the pressure drop at the typhoon center from the standard atmospheric pressure (here, this value was calculated as 1013 − P_c), and r is the distance (km) between the typhoon center and the observation points. Equation (1)(1) $P\left(r\right)=P_c+\mathrm{\Delta }Pexp\left(-\frac{R_m}{r}\right)$(1) is rearranged as follows to obtain R_m.

(2) $R_m=\hspace{0.17em}-rln\frac{P_r-P_c}{\mathrm{\Delta }P}$(2)

Thus, R_m can be estimated from the observed P_c and $P_r$ values at the time of typhoon landfall. In Eq. (2(2) $R_m=\hspace{0.17em}-rln\frac{P_r-P_c}{\mathrm{\Delta }P}$(2) ), $P_r$ is the sea-level pressure measured at the JMA observatories located at a distance of r from the typhoon center. As R_m is noted to have relatively large errors (Takagi et al. 2015), the R_m values averaged by the three meteorological stations closest to the center of the typhoon were adopted for validation. The R_m is only evaluated for land because R_m cannot be calculated over the ocean as there are no meteorological stations. In this manuscript, P_c and R_m of the typhoons are also simulated by the meteorological model. However, as the main focus of this manuscript is on the future changes in typhoon intensity, a detailed discussion of the future changes in R_m is omitted.

3. Results and discussion

3.1. Present-climate experiments

First, to confirm the validity of the HTM, the accuracy of the P_c and R_m values reproduced by the present-climate experiments are verified. The lifetime of a “typhoon” differs according to the individual case. For the observations, the P_c values averaged for all cases were 945.1 and 969.6 hPa at peak and landfall times, respectively. For the estimations, the averaged P_c values at the peak and landfall were 946.2 and 965.8 hPa, respectively. Thus, HTM can reproduce the overall intensity changes throughout the lifetimes of typhoons, regardless of their intensity. For example, Typhoon Vongfong (2014), which had the strongest intensity at peak time among the 49 cases, had an observed value of 900 hPa in good agreement with an estimated value of 902.7 hPa. The strongest values at landfall time were observed for Typhoons Songda (2004) and Man-Yi (2007); the observed P_c values were 945 hPa for both cases, and the estimated values were 936.5 and 947.9 hPa, respectively. Thus, the HTM employed in this study has sufficient resolution and performance to reproduce the intensities of the typhoons at their peak and landfall times adequately.

Figures 6a and 6b are scatter diagrams of P_c between the JMA Best Track data (observations) and HTM (estimations) at the peak and landfall times, respectively. At the peak times of the typhoons, the bias error is 1.17 hPa, the root mean square error (RMSE) is 8.72 hPa, and the correlation coefficient is 0.945; thus, the typhoon intensity at the peak times can be reproduced by HTM with high accuracy. At the landfall times, the bias error is −3.80 hPa, the RMSE is 9.67 hPa, and the correlation coefficient is 0.851, which is inferior to the peak value; however, there is no significant bias error in the HTM that should be considered in the analysis in the following sections, and it can be considered that the mean properties of P_c at the landfall times can be accurately reproduced overall. The variance between the estimations and observations is likely to be greater at the landfall times than at the peak times because the typhoon intensity changes drastically at the landfall times.

Figure 6. Scatter diagrams of central pressure (P_c) at (a) peak and (b) landfall times for observed (JMA best track) and estimated (HTM) typhoons.

As R_m can only be estimated at the landfall times for which observations for the sea level pressure are available, we discuss only the results at the landfall times. The mean values of the observed R_m and estimated R_m at the landfall times in all cases were found to be 89.5 and 101.4 km, respectively. The general trend is that R_m decreases from the genesis to the peak and then increases again during the dissipation (extratropical transition); there is a positive (negative) correlation between the change in P_c (V_m) and the change in R_m, indicating that R_m tends to decrease as the typhoon strength increases. This trend is consistent with that observed in the earlier studies (Colon 1963; Shea and Gray 1973). Consequently, one can say that the HTM used in this study can reproduce not only the intensity of a typhoon but also its structure with high accuracy.

Figure 7 is a scatter diagram of R_m between the observations obtained using the Myers equation Eq. (2)(2) $R_m=\hspace{0.17em}-rln\frac{P_r-P_c}{\mathrm{\Delta }P}$(2) and the estimations obtained using the HTM. The estimated R_m at the landfall times has an accuracy with a bias error of 11.9 km, an RMSE of 26.8 km, and a correlation coefficient of 0.780; the RMSE is not negligible. As a previous study confirmed that R_m has a very large temporal variability (Takagi et al. 2015), this can account for the large dispersion (RMSE) between the estimations and observations. However, the systematic bias error is relatively small, which implies that the overall structure in typhoons at the landfall times can be reasonably captured. The abovementioned verification shows that the HTM can accurately representing both the strength and structure of typhoons simulated in the present-climate experiments. Therefore, the results obtained in the future-climate experiments based on the d-PGWD method in the next section can be regarded as reliable to some extent at the peak and landfall times.

Figure 7. Scatter diagram of the radius of the maximum wind speed (R_m) at landfall time for observed (Myers 1954) and estimated (HTM) typhoons.

3.2. Future-climate experiments

As explained in Section 2.2, we assume that the tracks of typhoons in the future climate will not differ from the present climate, because of the limitations of the d-PGWD method.

According to the results of simulations, Typhoon Namtheun (2004) in the future climate is particularly strong, with P_c values of 850.4 and 905.1 hPa at the peak and landfall times, respectively. The differences in the estimated P_c between the present and future climates are −101.1 and −59.9 hPa at the peak and landfall times, respectively. This future change is the largest among all cases, both at the peak and landfall times. The average P_c for all 49 cases at the peak times is 900.5 hPa in the future climate, which is significantly lower than 946.2 hPa obtained from the present climate. For the average P_c at the peak times, there is a significant difference between the present and future climates at a significance level of 5% (one-sided test). With respect to the landfall times, the average P_c for the future climate is 960.3 hPa, which is lower than the average P_c, 965.8 hPa, for the present climate. For the average P_c at the landfall times, there is no significant difference between the present and future climates at a significance level of 5% (one-sided test). The future changes in P_c are reduced by −45.7 hPa at the peak times and by −5.50 hPa at the landfall times, thereby summarizing that the future-climate typhoons tend to be intensified on average.

Figures 8 and 9 show the relative frequency distributions of the central pressure P_c for the observations (dark blue) and for the estimations in the present (blue) and future climates (orange), at peak and landfall, respectively. The frequency distribution of P_c for the peak typhoons in Figure 8 reveals that the majority of the typhoons in the present climate (33%) are concentrated in the 920–940 hPa range, whereas a number of the typhoons in the future climate (27%) corresponds to the 840–860 hPa range. Therefore, the typhoon intensity at the peak times will be intensified by global warming as indicated by the frequency distribution that shifts to the left. Meanwhile, the frequency distribution of P_c at the landfall times in Figure 9 indicates that typhoons (22%) in the present climate are mostly centered in the 960–970 hPa range, although a large number of typhoons (18%) in the future climate have a peak in the 950–960 hPa range. Further, the number of typhoons with a P_c of less than 930 hPa, which never occurs under the present climate, increases to 7 in the future climate. Although there is no clear future change in typhoon intensity, as with the peak times (Figure 8), the number of future-climate typhoons that make landfall while maintaining their strong intensity is likely to increase because of global warming. Figure 10 shows the scatter plot of P_c between present and future climates. In the future climate, the intensity of the typhoon at the peak time will increase in all 49 cases (Figure 10a), and the intensity at the landfall times will increase in 30 cases (Figure 10b).

Figure 8. Relative frequency distributions of central pressure (P_c) at peak times for target typhoons in observations (JMA best track: dark blue bars), present-climate experiments (HTM: blue bars), and future-climate experiments (HTM: Orange bars). Average indicates the average value for each distribution.

Figure 9. Relative frequency distributions of central pressure (P_c) at landfall times for target typhoons in observations (JMA best track: dark blue), present-climate experiments (HTM: blue bars), and future-climate experiments (HTM: Orange bars). Average indicates the average value for each distribution.

Figure 10. Scatter plot of P_c between present and future climates: (a) P_c at the peak times and (b) P_c at the landfall times – the horizontal axis represents the P_c in the present climate, and the vertical axis represents the P_c in the future climate. The blue line shows the regression line and the green line shows y = x.

3.3. Physical factors causing future changes in typhoon intensity at landfall

First, the typhoon intensity at landfall times in each month is discussed. The future changes in P_c in the months from May through August are −1.5, −5.3, −14.9, and −9.7 hPa, respectively, and the future changes in September and October are 0.72 and 1.1 hPa, respectively. Typhoon intensity increased from May through August, but it tended to be almost the same as the present climate in September and October. Typhoons in September and October under the present climate have an average P_c of 961.8 and 953.8 hPa, respectively, and make landfall in a stronger state than in other months. Therefore, the degree of increase in intensity caused by global warming is thought to be smaller than for the period from May to August. In the future climate, even for typhoons in July and August, the P_c at the landfall times intensifies to approximately 960 hPa on average, and the typhoon intensity at the landfall times tends to increase overall.

The Maximum Potential Intensity (MPI) theory (Emanuel 1988; Bister and Emanuel 2002) is widely used as an index of the maximum potential intensity of a typhoon. Here, the distribution of the minimum central pressure calculated by MPI theory is compared between the meteorological fields of the present and future climates. However, the number of typhoons generated monthly should be considered when estimating the average MPI, as this figure is highly variable (e.g. 1 typhoon in May, 15 typhoons in Sep.). Figure 11-upper shows the distribution of minimum central pressure by MPI obtained by the GCMs, weighted by the typhoon number in each month (from May to October) in the Northwestern Pacific Ocean. The MPI value shown here is the minimum central pressure of the typhoon calculated by MPI theory. Compared to the present climate results (Figure 11a-upper left), future climate results (Figure 11b-upper middle) show that the minimum central pressure value is lower and that global warming tends to increase the maximum potential intensity of typhoons (Figure 11c-upper right). Therefore, it can be said that the typhoon intensity is likely to increase in the future climate. The future-climate simulations include super typhoons reaching up to 850–860 hPa, especially in late August and September. The monthly mean MPI for September is also lower than the average for the whole period, as shown in Figure 11b-upper middle. In practice, the future-climate simulations are conducted by the initial and boundary conditions such that an anomalous MPI in the present climate is added to the monthly mean MPI in the future climate. Therefore, some typhoons may develop even stronger intensities. In addition, other previous numerical studies (Tsuboki et al. 2015; Mori et al. 2021) relating to future-climate typhoons also reproduced super typhoons with a central pressure of 850–860 hPa, which is consistent with the results in this study. Further research on the fundamental processes in the development of typhoons is needed to determine whether such typhoons may occur in reality. In addition, future changes in MPI near Japan tend to be greater in the western Japan region. SST is one of the reasons for this trend. Similar changes can be seen in SST, which has a high correlation with MPI. Figure 11-middle shows the SST distributions in the present and future climates, averaged by monthly mean SST, obtained by the GCMs which are weighted by the typhoon number in each month (from May to October) in the Northwestern Pacific Ocean. The typhoon genesis points of all 49 cases, where the maximum sustained wind speed exceeds 17.2 m/s, are also shown (blue dots). Notably differing from the distribution in the present climate (Figure 11a-middle left), the SST in the future climate exceeds 30°C in the southern part of the Japan Islands (Figure 11b-middle), thereby providing a suitable environment to develop a very strong typhoon. Heat and water vapor supply from the warm sea surface is essential for the development of tropical cyclones (Holland 1997), and tropical cyclones are more likely to develop when the sea surface temperature exceeds 26.5°C (Gray 1968). In addition, a tropical cyclone of Category 4 or higher can develop when the SST exceeds 28°C and the temperature below the sea surface is also high (Mainelli et al. 2008; Lin et al. 2008). Therefore, if a tropical cyclone takes time to move over a higher-SST region from the genesis to the landfall, it is expected to be sufficiently developed into a stronger typhoon. According to the future change of the SST distribution (Figure 11c-middle right), the SST near the Japan Islands in the future climate will increase by +3–3.5°C, and is a level identical to that of the present-climate level over the eastern Philippine Sea (Mizuta et al. 2014).

Figure 11. Horizontal distribution of (Upper) minimum central pressure by MPI theory (Bister and Emanuel 2002), (Middle) sea surface temperature (SST), and (Lower) absolute degree of vertical wind shear at 300–850 hPa. (a) is the distribution under the present climate (2000–2019), (b) is the distribution under the future climate (2080–2099), and (c) is the distribution of future changes [(b) future − (a) present]. Each value used the frequency-weighted averaged value obtained using the number of typhoons in each month, derived from CMIP5 GCMs.

Figure 11-lower shows the future change in the vertical wind shear defined as the wind speed difference between 300 and 850 hPa, which is known to be involved in typhoon decay, especially in the mid-latitude baroclinic environment (Yamada, Moteki, and Yoshizaki 2010). In the subtropical high-pressure zone of 20°–30°N latitude, the vertical wind shear is small in both present (Figure 11a-lower left) and future climates (Figure 11b lower middle), which makes it a suitable environment for the development of typhoons. The future changes in the vertical wind shear (Figure 11c-lower right) are within the range of −0.5 to +0.5 m/s near the Japan Islands, which suggests little change in the future. Therefore, the influence of the future change in mean vertical wind shear on the typhoon intensity is also identified to be small.

These findings suggest that the increase in SST has a considerable effect on the future change in typhoon intensity. The high SST area above 29°C is currently confined to the lower latitudes in the eastern part of the Philippines Sea; however, in the future climate, the high SST region is expected to spread to the vicinity of the Japan Islands. It is speculated that many typhoons will be able to develop to a “super-typhoon” level (reaching the maximum sustained 1 min surface wind speeds of at least 67 m/s) even near the Japan Islands. Some of the typhoons are expected to make landfall while maintaining their strong intensity. Although the present-climate typhoons that develop at relatively high latitudes cannot be fully developed throughout their life cycle, they can rapidly develop even at high latitudes in the future climate owing to the high SST exceeding 30°C.

The numerical experiments of the 49 cases in this study show that the future change in typhoon intensity at the landfall times is larger in some cases and smaller in others. Thus, it appears to be a difference in the effects of global warming in different typhoon cases. Therefore, there is a possibility that not only sea surface temperature but also other factors may affect future changes in typhoon intensity.

First, based on the criteria of the JMA Best Track data at the landfall times (Table 3), 24 typhoons in the present climate make landfall with “strong” or “very strong” intensities (V_m of 33 m/s or more), and their mean elapsed time from genesis to landfall is 7.58 days. In addition, their mean R_m for the 24 typhoons is 87.5 km at the landfall times when the average P_c is 953.8 hPa. In contrast, there are 25 out of 49 cases of typhoons in the present climate, which make landfall with a V_m of less than 33 m/s, with a mean elapsed time from genesis to landfall of 6.18 days and with a mean R_m of 114.7 km at the landfall times when the average P_c is 977.3 hPa. In other words, the present-climate strong typhoons at the landfall times tend to have a longer elapsed time from genesis to landfall (1.4 days longer) and a smaller R_m (30 km smaller) than weak typhoons. Thus, the future change in typhoon intensity at the landfall times is greatly influenced by the elapsed time from genesis to landfall (T_l) and the R_m at the landfall times (R_ml) under the present climate. Note that the definition of R_m (all lifetime) and R_ml (only landfall times) are different. Therefore, we further analyze the effect of these two parameters on the future changes in typhoon intensity.

Table 3. Summary of 49 typhoons for V_m, time from genesis to landfall, and R_m by JMA: 24 typhoons make landfall with “strong” or “very strong” intensities (V_m of 33 m/s or more) and 25 typhoons make landfall with a V_m of less than 33 m/s

	The cases of Vm ≥ 33 m/s	The cases of Vm < 33 m/s
The mean elapsed time from genesis to landfall [day]	7.58	6.18
The mean Rm [km]	87.5	114.7
The average Pc at the landfall times [hPa]	953.8	977.3

Figure 12 shows the frequency distribution of T_l for all 49 cases under the present climate. The average T_l value is 6.87 days, which indicates a profile close to a normal distribution. In this study, typhoon cases with T_l below the average are defined as “low T_l” (23 cases), and those with average T_l values or higher are defined as “high T_l” (26 cases). Table 4 summarizes the average P_c and future changes in the present and future climates for the low T_l and high T_l cases. The average future change for the low T_l cases (−13.3 hPa) is larger than that for all cases (−5.50 hPa). In contrast, the average future change for the high T_l cases (+1.40 hPa) is smaller than that for all cases (−5.50 hPa). Furthermore, the difference between the low- and high T_l cases is found to be significant (a significance level of 5%, one-sided test). Therefore, future-climate typhoons will be reinforced in low T_l cases occurring near the Japan Islands because of global warming. Under the present climate, the intensities of the high T_l cases at the landfall times (956.9 hPa) are stronger than those of the low T_l cases at the landfall times (975.8 hPa), and there is a significant difference (a significance level of 5%, one-sided test). However, under future climates, the low T_l cases at landfall times become significantly stronger (962.5 hPa), which obscures the difference in typhoon intensity with the high T_l cases (958.4 hPa). The degree of achievement of the MPI was confirmed for the cases where the strength increased above the average (−5.5 hPa) at the time of landfall (23 cases). As a result, the ratio of the intensity reached to the MPI is 62.0% in the present climate. However, this value is 86.3% in the future climate, and it can be seen that typhoons make landfall after reaching an intensity closer to the MPI. Thus, there is concern that many typhoons in the future climate are likely to mature into a steady-state structure independent of T_l, and they will make landfall on the Japan Islands while maintaining their intensity.

Table 4. Central pressure values (P_c) averaged for all (49 cases), low T_l (23 cases), and high T_l (26 cases) in PGWDs. Typhoons with T_l of less than 6.87 days are referred to as “low T_l” (23 cases), whereas those with T_l of 6.87 days or more as “high T_l” (26 cases). The left, middle, and right columns show the average values for the present-climate experiments, future-climate experiments, and future changes, respectively. The values in parentheses indicate the standard deviation

	Average present-climate Pc [hPa]	Average future-climate Pc [hPa]	Average future change[hPa]
All Typhoons (49 cases)	965.8 (16.9)	960.3 (23.2)	−5.5 (17.6)
low Tl (23 cases)	975.8 (13.8)	962.5 (26.4)	−13.3 (18.1)
high Tl (26 cases)	956.9 (14.3)	958.4 (19.6)	+1.4 (13.9)

Figure 12. Relative frequency distribution of the time period from genesis to landfall (T_l) according to JMA best track. The average values of T_l for all cases is 6.87 days. Cases with values of less than 6.87 days are referred to as low T_l, whereas cases with values equal to or higher than 6.87 days as high T_l.

Figure 13 shows the relationship between T_l and P_c of the landfalling typhoons under the present and future climates. In the present climate (Figure 13a), typhoons with higher T_l tend to yield a lower P_c at the landfall times. In the future climate (Figure 13b), there will be more typhoon cases that can develop to a P_c of less than 940 hPa despite the small T_l. In the future change in P_c (Figure 13c), there is a nonlinear trend, and the regression equation excluding outliers (T_l = 6.79, P_c = −59.9; T_l = 18.25, P_c = −3.8) is shown in the figure. The various combinations of T_l were considered, from linear to quartic, with the cubic form considered to be the best option. Typhoons with shorter T_l make landfall before they are fully developed, under the present climate. Although typhoons will be able to develop even in a short period of time, due to the increase in SST under the future climate, the future change of P_c (∆P_c) will be significant and negative. The extreme minimum value of ∆P_c is around T_l = 3. However, as T_l increases, typhoons can fully develop even under the present climate. Therefore, the difference between the present and future typhoon intensities becomes smaller and, in some cases, positive. Typhoon cases with lifetimes of 7–12 days vary widely. Some typhoons increase significantly in intensity due to increased SST under the future climate, whereas others are suppressed by increased AT in the upper atmosphere (200–300 hPa) or being exposed vertical wind shear for a long time (Frank and Ritchie 2001; Wong and Chan 2004; Takemi, Ito, and Arakawa 2016). Thus, the predominance of SST or other factors depends on the environmental field (season, area etc.) of the typhoon. In contrast, with an approximate border of T_l = 13, ∆P_c again increases in the negative direction. For T_l larger than the second extreme (i.e. T_l > 13), the data are shown as a dotted line. This is because such cases are rare, and we cannot discuss quantitatively which physical factors, either SST or other factors, are dominant. In contrast, it is not important whether these cases are included or not in determining the functional form of ΔP_c.The relationship between the future change in P_c and T_l can be expressed as a cubic equation. The relationships between T_l and P_c in the present (Figure 13a), T_l and P_c in the future (Figure 13b), and T_l, and the future change in P_c (Figure 13c) were tested using the Mann–Kendall test (significance level of 5%, two-sided test), and it was confirmed that these parameters have a clear trend.

Figure 13. Scatter diagrams of the time period from genesis to landfall (T_l) derived from JMA best track and the central pressure (P_c) for target typhoons calculated from: (a) present-climate experiments and (b) future-climate experiments. (c) A scatter diagram of T_l and future changes in P_c between (b) future and (a) present climates. In (a) and (b), the Orange and purple points represent cases where rapid intensification (RI) did and did not occur, respectively. In (c), the Orange and purple points represent cases where RI occurs under only the future climate and all other cases, respectively. Green lines in the diagrams represent regression lines.

Figure 14 shows the frequency distribution of R_ml for all 49 cases under the present climate. The mean value of R_ml is 101.4 km, which shows a normal distribution. Therefore, we set the mean value as the threshold value in the present climate, defining the small R_ml cases (23 typhoons) as those with R_ml below the mean value and the large R_ml cases (26 typhoons) as those with R_ml above the mean value.

Figure 14. Relative frequency distribution of the radius of maximum wind speed at landfall times (R_ml) for target typhoons in the present-climate experiments. The average values of R_ml for all cases is 101.4 km. Cases with values of less than 101.4 km are referred to as small R_ml, whereas cases with values equal to or larger than 101.4 km as large R_ml.

Table 5 summarizes the average P_c and future changes in the present and future climates for the small R_ml and large R_ml cases. The average future change in P_c at the landfall times is negatively larger for the large R_ml case (−11.7 hPa) than for all cases (−5.50 hPa). In contrast, the average future change at the landfall time is larger (change of typhoon intensity to weaken) for the small R_ml cases (+1.60 hPa) than for all cases. In addition, the difference in the average future changes between the large- and small R_ml cases reveal a significant difference (a significance level of 5%, one-sided test). As noted in previous studies (Smith, Montgomery, and Sang 2009), when the typhoon V_m increases, R_m contracts as it satisfies the law of conservation of angular momentum. Therefore, well-developed typhoons tend to have large V_m with small R_m (Colon 1963; Shea and Gray 1973). Therefore, it can be interpreted that typhoons with a large R_ml in the present climate can intensify further because of the future global warming. In contrast, typhoons with small R_ml in the present climate leave little room for their reinforcement at the landfall times.

Table 5. Central pressure values (P_c) averaged for all (49 cases), small R_ml (23 cases), and large R_ml (26 cases) in PGWDs. Typhoons with R_ml of less than 101.4 km are referred to as “small R_ml” (23 cases), whereas those with R_ml of 101.4 km or more as “large R_ml” (26 cases). The left, middle, and right columns show the average values for the present-climate experiments, future-climate experiments, and future changes, respectively. The values in parentheses indicate the standard deviation

	Average present-climate Pc [hPa]	Average future-climate Pc [hPa]	Average future change[hPa]
All Typhoons (49 cases)	965.8 (16.9)	960.3 (23.2)	−5.5 (17.6)
small Rmli (23 cases)	962.1 (17.2)	963.7 (22.9)	+1.6 (16.3)
large Rml (26 cases)	969.1 (15.9)	957.3 (23.0)	−11.7 (16.4)

Figure 15 shows the relationship between R_ml and P_c of the landfalling typhoons under the present (Figure 15a) and future climates (Figure 15b). Generally, the typhoon intensity and R_ml obey the law of conservation of angular momentum. The stronger the typhoon, the smaller the radius tends to be, mainly due to the conservation of angular momentum. However, this may not always be the case, depending on the properties of each individual typhoon. According to the Mann–Kendall test (a significance level of 5%, two-sided test), there is no significant trend between R_ml and P_c in both the present climate (Figure 15a) and the future climate (Figure 15b). In contrast, the large R_ml cases in the present climate have a large and negative future change of P_c, which indicates a moderate negative correlation in Figure 15c (correlation coefficient of −0.49). Typhoons with large R_ml under the present climate tend to increase their intensity due to global warming (∆P_c is large and negative), a clear trend can be found between ∆P_c and R_ml (Figure 15c). This is because the present-climate typhoons with low T_l and large R_ml are often relatively weak at the landfall times, which suggests that there is room for them to develop further with future global warming. The larger the R_ml under the present climate, the larger the negative value of ∆P_c. Therefore, the relationship between the future change in P_c and R_ml under the present climate can be expressed as a linear equation.

Figure 15. Scatter diagrams of the radius of maximum wind speed (R_ml) under present climate and central pressure (P_c) for target typhoons calculated from: (a) present-climate experiments and (b) future-climate experiments. (c) A scatter diagram of R_ml and the future changes in P_c. The points correspond to those provided in .Figure 14

Accordingly, we found that the future change in typhoon intensity at landfall times is strongly constrained by T_l and R_ml in the present climate. In addition, we expect that typhoons in the future climate will develop rapidly and make landfall on the Japan Islands in a short period after their genesis. Hence, we examine the frequency and magnitude of the rapid intensification (RI) of typhoons under present and future climates. In this study, RI is defined as the development of a typhoon exceeding 15.4 m/s (30 kt) in 24 h (Kaplan et al. 2015; Wang et al. 2015; Bhatia et al. 2019).

The scatter plots in Figures 13 and 15 are separated by the presence or absence of RI of the typhoons. The RI is identified for 25 cases under the present climate, i.e. 51.0% of the total (the orange points in Figures 13a and 15a), and for 41 cases under the future climate, i.e. 83.7% of the total. In addition, 16 out of 41 cases, where RI is observed only in future climates, are also detected. In addition, 11 out of 16 cases are detected as low T_l and large R_ml cases. Hence, the low T_l or large R_ml cases are more likely to exhibit RI in the future climate than the high T_l or small R_ml cases.

It is concluded from the results obtained from the high-resolution typhoon database that RI is more likely to occur in more cases under the future climate mainly because of the rise in SST caused by the progressing global warming. In addition, this impact is remarkable in the low T_l and large R_ml typhoons under the present climate.

4. Proposal of an e-PGWD method for storm surges along the Pacific Coast in Japan

In this section, we exploit the statistical and empirical analysis between the present- and future-climate experiments using the d-PGWD method for all 49 typhoon cases described in the previous section, and we propose a simplified method “e-PGWD” for application to the Pacific coast of the Japan Islands. The d-PGWD based on dynamical meteorological models (such as HTM) are extremely computationally demanding, and it is very difficult for policy makers and harbor/port managers to estimate the meteorological fields in a typhoon for storm surge simulations at each port in practice, using it. Therefore, we develop a method for efficiently performing global warming experiments for storm surges. In this approach, the future changes ∆P_c and ∆R_ml are calculated based on empirical equations for the input of the observed T_l and R_ml. The results are fed into a two-dimensional empirical typhoon model that operates with a low computational cost to obtain the meteorological field under the future climate. In this study, Typhoons Jongdari (2018), Jebi (2018), and Trami (2018), all of which made landfall to Japan in 2018, are selected as case studies (none of these cases were included in the 49 cases considered in the previous section). The storm surge estimations given by both d-PGWD and e-PGWD are compared to validate the proposed method.

4.1. 4.1. e-PGWD methodology

4.1.1. A calculation flow for e-PGWD

The proposed e-PGWD method is implemented according to the flowchart shown in Figure 16. First, attribute values of a target typhoon, i.e. typhoon position (TY_pos), and P_c and R_m1 (P_cp and R_mlp, respectively) are obtained from the JMA best track as a present-climate typhoon. As R_mlp cannot be observed directly, the value is indirectly assessed using the estimation formula given in Eq. (2)(2) $R_m=\hspace{0.17em}-rln\frac{P_r-P_c}{\mathrm{\Delta }P}$(2) . An empirical typhoon model (ETM) (Mitsuta and Fujii 1987) is used to express typhoon meteorological fields under the present climate. The obtained typhoon meteorological field, including wind vectors and sea surface pressure, is fed into a storm surge model (SSM) to evaluate the storm surge associated with the target typhoon in the present climate.

Figure 16. Computational flow of the e-PGWD method. The left, middle, and right columns show the flows of the present-climate experiment, future-climate experiment (e-PGWD), and horizontal resolution at each calculation step, respectively.

The e-PGWD method we propose has three constraints. The first is that future changes can be calculated using easily available data, the second is that the data to be input are values for the present climate and, lastly, the data to be input have to correlate with future changes in typhoon intensity. To achieve the goal that propose future-climate experiment method that can be conducted by practitioners considering these constraints, we devised an equation to estimate future changes in typhoon intensity from present-climate information without using future-climate parameters. In Section 3.3, the correlation between many parameters available in the present climate – peak intensity, the intensity at landfall times, typhoon location, season (including temperature and SST), the region of landfall, radius, and time – and ∆P_c and ∆R_ml were used to select the explanatory variables for the estimation formula. As a result, it was revealed that T_l and R_ml are correlated most strongly with ∆P_c and ∆R_ml. In addition, the relationship between ∆P_c and T_l is expressed in cubic form (Figure 13c), and the relationship between ∆P_c and R_ml is expressed linearly (Figure 15c). Therefore, an estimation formula was devised for ∆P_c and ∆R_ml with nonlinear regression analysis, using the generalized additive model and the results from Section 3.3. The future changes, ∆P_c and ∆R_ml, calculated by the estimation formula, can be added to the attribute values, P_cp and R_mlp, of a present-climate typhoon to express the intensity and structure, P_cf and R_mlf, of a future-climate typhoon.

Compared to the flowchart of the d-PGWD method shown in Figure 2, it is not necessary for the e-PGWD method to use a three-dimensional dynamical meteorological model with high computational cost, and therefore, this approach is characterized by high computational efficiency. However, it should be noted that, like the d-PGWD, the e-PGWD is calculated under the assumption that the typhoon track and its translation speed in the future climate will not differ from those in the present climate. Typhoon intensity is determined by the influence of many parameters (e.g. SST) that vary as a result of climate change. However, the method of inputting the physical quantities of the future climate and evaluating the future change of the typhoon must use a meteorological model that requires specialized knowledge. This deviates from the purpose of this research that propose method that can be conducted by practitioners. Therefore, the e-PGWD proposed in this paper uses only data that are readily available in the present climate. By using the nine combinations of P_ce and R_mle considering the uncertainties of the empirical formula of ∆P_c and ∆R_ml, typhoon meteorological fields under the future climate are calculated using the ETM in the same way as for the present climate. In addition, e-PGWD calculations for storm surges associated with a future-climate typhoon can be conducted by the SSM based on the created typhoon meteorological fields. It should be emphasized that the proposed e-PGWD is a method for people who are not experts in climate change, such as port managers, to carry out pseudo-global warming experiments easily.

4.1.2. Regression equations for typhoon future changes

According to the discussion in the Section 3.3, the future changes in the intensity and structure of landfalling typhoons will be significantly affected by T_l and R_ml at landfall times in the Japan Islands, and there are significant differences between low T_l cases (small R_ml cases) and high T_l cases (large R_ml cases). Therefore, the following estimated equation is obtained using a nonlinear regression analysis with future change in P_c (∆P_c) as the objective variable and T_l and R_ml as explanatory variables, based on the high-resolution typhoon database. The form of estimation formula is based on the discussion in Section 3.3, and the best combination is cubic for T_l and linear for R_ml. We construct the formula by nonlinear regression analysis using the results of Section 3.3 and the generalized additive model. The magnitude of estimation errors (bias, root mean square), the correlation coefficient (CORR), the degree of freedom-adjusted coefficient of determination (CD), and Akaike’s Information Criterion (AIC) are used to determine whether the estimated equation is the best combination of T_l and R_ml. The estimation formula Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) has the highest CORR (0.64), CD (0.42), and lowest AIC in all combinations we considered.

(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3)

This formula (regression coefficients) is constructed using robust estimation to remove outliers or noise in the data and bound the calibration signal within narrow limits. In addition, a constant indicates residuals. This equation consists of an additive model of a cubic equation for T_l and a linear equation for R_ml. Both T_l and R_ml are parameters under the present climate. T_l obtained from the JMA best track data and R_ml estimated using Eq. (2)(2) $R_m=\hspace{0.17em}-rln\frac{P_r-P_c}{\mathrm{\Delta }P}$(2) are input into Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) . P_c in the future climate is calculated as the sum of the observed P_c in the present climate and the future change ∆P_c obtained by Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) . Figure 17 provides a comparison of the estimated P_c for e-PGWDs performed using Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) with d-PGWDs performed using HTM. The obtained correlation coefficient is as high as 0.71, and thus, it is possible to capture the overall trend with a simple method. However, there are bias errors in describing typhoons weaker than 980 hPa as stronger, and typhoons stronger than 940 hPa as weaker; further, there must exist non-negligible uncertainty with a root mean square error (RMSE) of 17.3 hPa. Overall, although it is necessary to consider the uncertainty of the estimation formula, the simple formula expressed in Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) is expected to estimate P_c under the future climate with a certain degree of accuracy. The standard errors of Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) is 9.87 hPa. The correlation coefficient between T_l and R_ml is −0.1, which suggests that multicollinearity in the equation is not a problem.

Figure 17. Scatter diagram of future-climate central pressure (P_c) in PGWDs (horizontal axis) and e-PGWDs (vertical axis).

In addition to the typhoon intensity P_c, the typhoon structure R_ml may change under the future climate conditions as a result of global warming. Therefore, it is necessary to consider future changes in both P_c and R_ml in the storm surge estimation for the future climate. As R_ml is a required input condition for the ETM, the following estimation formula is also obtained using a nonlinear regression analysis with the future change in the typhoon structure ∆R_ml as the objective variable, and T_l and R_ml as explanatory variables, in the same process to ∆P_c.

(4) $\mathrm{\Delta }{R}_{ml}=-0.0263T_l^3+0.338T_l^2+2.40T_l-0.648R_{ml}+31.2$(4)

where ∆R_ml is the future change in the radius of the maximum wind speed (km), and both T_l and R_ml are taken as explanatory variables under the present climate. In addition, the constant means residuals. For ∆P_c, the regression equation is constructed by performing a robust estimation. This equation also consists of an additive model of a cubic equation for T_l and a linear equation for R_ml. The relationship between ∆R_ml and T_l (Figure 18a), and between ∆R_ml and R_ml under the present climate (Figure 18b) show that the data are fitted using cubic and linear functions, respectively. The obtained correlation coefficient is 0.49, and the RMSE is 25.4 km. Although the correlation of the R_ml formula is inferior to that of the P_c formula, it is considered that the tendency of future changes is captured. Equations (4)(4) $\mathrm{\Delta }{R}_{ml}=-0.0263T_l^3+0.338T_l^2+2.40T_l-0.648R_{ml}+31.2$(4) has a standard error of 23.1 km.

Figure 18. (a) The scatter plot between T_l and future change of R_ml (ΔR_ml.). The scatter plot between R_ml under present climate and future change of R_ml (ΔR_ml). The correlation coefficient is 0.24 for T_l and ΔR_ml and −0.61 for R_ml under the present climate and ΔR_ml.

In this study, we estimate the typhoon meteorological field under the future climate using P_c + ∆P_c and R_ml + ∆R_ml as input parameters for the empirical typhoon model, as evaluated by Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) and (4(4) $\mathrm{\Delta }{R}_{ml}=-0.0263T_l^3+0.338T_l^2+2.40T_l-0.648R_{ml}+31.2$(4) ). To account for the uncertainties in Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) and (4(4) $\mathrm{\Delta }{R}_{ml}=-0.0263T_l^3+0.338T_l^2+2.40T_l-0.648R_{ml}+31.2$(4) ), in addition to the case where the estimates of ∆P_c and ∆R_ml are used directly, we compute the cases of adding and subtracting the standard errors (P_ce = 9.87 hPa and R_mle = 23.1 km) to them (P_c + ∆P_c and R_ml + ∆R_ml). (The above results are presented as ±P_ce and ±R_mle in the tables). In other words, we perform computations for a total of nine cases combining three patterns of P_ce (−9.87, 0.0, and +9.87 hPa) and three patterns of R_mle (−23.1, 0.0, and +23.1 km). Thus, we conduct the future-climate experiments for storm surges using the ETM and SSM, while considering the uncertainties of the e-PGWD method.

4.1.3. Two-dimensional empirical typhoon model

In the e-PGWD approach, an ETM is used to construct meteorological fields for storm surge calculations (Mitsuta and Fujii 1987) because of its low computational costs. The ETM is composed of the Myers equation (Myers 1954) for the atmospheric surface pressure field given as

(5) $P\left(r\right)=P_c+\mathrm{\Delta }Pexp\left(-\frac{R_m}{r}\right)$(5)

and the gradient wind equation for the wind velocity vector field, which is expressed as

(6) $\frac{V^2}{r_t}+fV=\frac{1}{\rho }\frac{\mathrm{\partial }P}{\mathrm{\partial }}$(6)

where $V$ is the wind velocity based on the atmospheric pressure gradient; $r_t$ is the radius of curvature of the air-particle trajectory, which coincides with the distance $r$ from the typhoon center; $f$ is the Coriolis parameter; and $\rho$ is the air density. In addition to the above two equations, Blaton’s equation is adopted to incorporate the effect of the typhoon movement on the wind velocity vector field.

(7) $\frac{1}{r_t}=\frac{1}{r}+\frac{1}{V}\frac{\mathrm{\partial }\psi }{\mathrm{\partial }t}$(7)

where $t$ is time and $\psi$ is the particle movement direction. Through the introduction of Eq. (7(7) $\frac{1}{r_t}=\frac{1}{r}+\frac{1}{V}\frac{\mathrm{\partial }\psi }{\mathrm{\partial }t}$(7) ), stronger winds blow in the dangerous semicircle on the right side of the typhoon’s traveling direction and weaker winds exist in the navigable semicircle on the left side of the traveling direction. Furthermore, the ground surface friction effect is assumed using Goldberg–Mohn friction, and the blowing angle of the wind velocity vector is set to 10°, which is typical on the sea.

4.1.4. Storm surge model

An SSM is employed in this study to estimate the storm surge height (sea-level anomaly) at the target bay. The SSM is composed of a one-layer nonlinear long-wave equation, and the ocean tide level and ocean current vector are used as the prognostic variables. The typhoon meteorological fields obtained by the ETM are input into the SSM every 15 min as the sea-surface boundary condition. The horizontal grid spacing of the SSM is 1 km. For the sea bathymetric data, 1 min mesh global ground relief data (ETOPO1) provided by NOAA is used (Amante and Eakins 2009). For the coastline data, 30 s mesh global land-use data provided by the United States Geological Survey is utilized (Danielson and Gesch 2011). The target bays for the storm surge calculations are Ise Bay and Mikawa Bay (Figure 19a) for Typhoons Jongdari and Trami, and Osaka Bay for Typhoon Jebi (Figure 19b).

Figure 19. Horizontal distributions of the ocean topography around (a) Ise Bay and Mikawa Bay, and around (b) Osaka Bay are used in the storm surge model (SSM). The red points indicate the target ports for validating and discussing the SSM results: (a) Port of Nagoya and (b) Port of Osaka.

4.2. Verification and assessment of e-PGWD

We considered the three cases of Typhoons Jongdari (2018), Jebi (2018), and Trami (2018) to verify the e-PGWD validity and calculated their future changes (Table 6). The T_l and R_ml values for those typhoons are obtained from the JMA best track as input parameters described in Table 6. Then, future changes ∆P_c and ∆R_ml are estimated by Eq. (3)(3) $\mathrm{\Delta }{P}_c=-0.0304T_l^3+0.759T_l^2-3.73T_l-0.243R_{ml}+8.60$(3) and (4(4) $\mathrm{\Delta }{R}_{ml}=-0.0263T_l^3+0.338T_l^2+2.40T_l-0.648R_{ml}+31.2$(4) ) for each case. Typhoon Jongdari with small T_l and large R_ml tends to have a large negative future change in P_c. In the case of Typhoon Jebi, where T_l is large and R_ml is small, the future change in typhoon intensity is not considerably large. When calculating ETM, the standard errors of the estimation formula for ∆P_c and ∆R_ml are also added and subtracted to consider uncertainties for the future change. A total of nine calculations for each typhoon cases are performed considering the combination of three patterns of ∆P_ce (−9.87, 0.0, and +9.87 hPa) and three patterns of ∆R_mle (−23.1, 0.0, and +23.1 km). For the storm surge calculation, the d-PGWDs and e-PGWDs obtained from the SSM are evaluated and validated at the Port of Nagoya for Typhoons Jongdari and Trami, and the Port of Osaka for Typhoon Jebi (the red points in Figure 19a and b). The total calculation period is 24 h from before to after the landfalls of the typhoons. Our ensemble calculations for the validation of e-PGWD assume that the typhoon track is not changed from the present climate, and 9 types of input values are prepared for the typhoon intensity and typhoon radius. In addition, 9 types of storm surge results can be obtained in each typhoon case.

Table 6. Attribute variables for three target typhoons in 2018 (Typhoons Jongdari, Jebi, and Trami) used for e-PGWD: input typhoon parameters (P_c,T_l, and R_ml), estimated future typhoon changes (∆P_c and ∆R_ml), and estimated future typhoon parameters (P_cf and R_mlf) with standard errors of estimation (P_ce and R_mle)

		Typhoon Jongdari (2018)	Typhoon Jebi (2018)	Typhoon Trami (2018)
Target Bay (target point)		Ise Bay (the Port of Nagoya)	Osaka Bay (the Port of Osaka)	Ise Bay (the Port of Nagoya)
Input	Pc [hPa]	970	950	960
	Tl [day]	4.17	7.13	9.21
	Rml [km]	113.9	77.9	109.3
Future change	ΔPc [hPa]	−23.6	−9.36	−11.7
	ΔRml [km]	−28.6	+5.47	−9.40
Estimated values	Pcf [hPa]	946.4	940.6	948.3
	±Pce	±9.87
	Rmlf [km]	85.3	83.4	99.9
	±Rmle	±23.1

First, we confirm the reproducibility of the present-climate experiments for Typhoons Jongdari, Jebi, and Trami (Figure 20). The maximum sea-level anomalies reproduced by HTM are 0.75, 2.85, and 1.33 m for Typhoons Jongdari, Jebi, and Trami, respectively. The maximum sea level anomalies observed by JMA are 0.73, 2.77, and 1.44 m, respectively, which indicate that the comparison between the HTM and observations accurately reproduces the sea level anomalies that occurred. Next, the results obtained using the e-PGWD method are validated by comparing them with the results obtained using the d-PGWD approach. The maximum sea-level anomalies given by the d-PGWD are 1.22, 3.08, and 2.05 m for Typhoons Jongdari, Jebi, and Trami, respectively. Considering the uncertainties in the ∆P_c and ∆R_ml estimates by the e-PGWD, the ensemble means (standard deviations) of the maximum sea level anomalies are found to be 1.74 m (0.52 m), 3.26 m (0.74 m), and 1.98 m (0.54 m) for Typhoons Jongdari, Jebi, and Trami, respectively. The maximum sea-level anomalies estimated using d-PGWD are distributed within the standard deviation of using e-PGWD in all cases (the whiskers in the boxplot). Therefore, the e-PGWD method based on the constructed estimation formula yields reasonable results. Thus, although this is not a sufficient verification because there were only three cases, it was confirmed that it is possible to estimate the future changes of storm surges of different scales using e-PGWD.

Figure 20. Box-and-whisker plot of the maximum sea-level anomalies at the target port simulated by e-PGWDs for nine cases. The blue, dark-blue, Orange, and red points indicate the maximum sea-level anomalies at the target port simulated by the observations (black points), present-climate experiments (green points), PGWD results by HTM (Orange points), and average values for all e-PGWDs. The box height includes the half of all the ensemble sizes in e-PGWDs, and the black line in the box represents the median value.

5. Concluding remarks

In this study, the d-PGWD experiments were performed using a three-dimensional HTM for 49 typhoons that made landfall in Japan from 2000 to 2017. Under the future climate, typhoon intensity was found to show a tendency to increase at both peak and landfall (average values of −45.7 and −5.5 hPa at the peak and landfall times). The tendency of future changes in typhoon intensity obtained in this study are consistent with the tendency of increasing typhoon intensity identified in previous studies. Detailed analyses of the characteristics of typhoons making landfall in the Pacific coast in Japan showed that there is a significant difference in the future changes in the typhoon intensity based on T_l and R_ml. In particular, a typhoon that has a lower T_l than the average (6.87 day) and a larger R_ml than the average (101.4 km) in the present climate has the potential to exhibit greater future changes in terms of typhoon intensity, which indicates that the future-climate typhoons will become more intense. Therefore, weak typhoons characterized by low T_l and large R_ml under the present climate are likely to develop into the maximum intensity characterized by high T_l and small R_ml under the future climate. To the best of our knowledge, there is currently no published literature that has used a RCM to explore the characteristics of a number of typhoons similar to the 49 events used in this study. Therefore, the scale of this investigation presents a significant contribution to the existing knowledge-base on future changes in typhoon intensity.

Furthermore, regression equations underlying the e-PGWD approach were constructed based on the trend of future changes in the typhoon intensity derived using the high-resolution typhoon database with d-PGWD. The typhoon intensity under the future climate was determined by the sum of the present-climate value P_c with the future change (∆P_c), and the typhoon meteorological fields were calculated by an ETM with a low computational cost. The storm surge heights at the target bays were estimated using an SSM. e-PGWD experiments were conducted for three typhoons that made landfall in Japan in 2018 (Typhoon Jongdari, Typhoon Jebi, and Typhoon Trami), and the results were compared with those of the d-PGWD experiments for validation. The storm surge heights obtained by d-PGWD fell within the error range of those obtained using e-PGWD, and thus, the validity of the e-PGWD method proposed in this study had a certain level of accuracy. The e-PGWD differs from the original d-PGWD method in that it uses a low-cost model that allows policymakers and port managers to implement it easily. Thus, the proposed e-PGWD method can contribute to disaster prevention and mitigation measures for coastal disasters caused by the future climate (in 2080–2099, and under the RCP8.5 scenario), such as storm surges caused by typhoons, for major ports along the Pacific coast in Japan. However, compared to the d-PGWD method, the e-PGWD method has a tendency to underestimate typhoon intensities with a central pressure of less than 920 hPa at landfall under the present climate, thus, it requires further development. In addition, the future changes in typhoon intensity evaluated by the e-PGWD capture the average characteristics of typhoons that have landed on the Japanese archipelago, meaning that it is not possible to discuss regional differences in future changes. Therefore, it is necessary to accumulate more cases after reducing the uncertainty to discuss regional differences with this method. A challenge for future research is to construct an advanced e-PGWD method that has a higher accuracy by assessing more typhoon cases with the d-PGWD method. Once a sufficient number of cases are available, it is expected that more robust estimation equations can be devised by classifying the data into training and test data, and validating each with independent numbers (T_l,R_ml,∆P_c). In addition, though harbor or port managers do not need to calculate additional d-PGWD, it is necessary to update the e-PGWD provided by us when the new calculation results are obtained. The present form of the e-PGWD approach has the limitations of restricted applicability and high uncertainty. Our future work will concentrate on refining e-PGWD to reduce uncertainty in the scenarios, time periods, and models because there are social demands to evaluate different time periods (e.g. 2030s and 2050s) and different scenarios (e.g. RCP4.5 and RCP2.6).

Acknowledgments

This research was supported by a JSPS Research Fellow Grant (No. 17J04771) and a JSPS KAKENHI Grant (No. 18H01542). This manuscript incorporates material from the first authors’ PhD dissertation, Graduate School of Engineering, Gifu University.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Additional information

Funding

This work was supported by the Japan Society for the Promotion of Science [17J04771]; Japan Society for the Promotion of Science [18H01542].