Heatstroke Risk Projection in Japan under Current and Near Future Climates

概要

August
Journal 2022
of the Meteorological Society of Japan, 100(4),
S. NAKAMURA
597−615, 2022. 
et al. doi:10.2151/jmsj.2022-030

597

Heatstroke Risk Projection in Japan under Current and Near Future Climates
Shingo NAKAMURA
Graduate School of Life and Environmental Sciences, University of Tsukuba, Tsukuba, Japan

Hiroyuki KUSAKA
Center for Computational Sciences, University of Tsukuba, Tsukuba, Japan

Ryogo SATO1
Graduate School of Life and Environmental Sciences, University of Tsukuba, Tsukuba, Japan

and
Takuto SATO
Center for Computational Sciences, University of Tsukuba, Tsukuba, Japan
(Manuscript received 17 November 2021, in final form 11 March 2022)

Abstract
This study assesses heatstroke risk in the near future (2031 – 2050) under RCP8.5 scenario. The developed model
is based on a generalized linear model with the number of ambulance transport due to heatstroke (hereafter the
patients with heatstroke) as the explained variable and the daily maximum temperature or wet bulb globe temperature (WBGT) as the explanatory variable. With the model based on the daily maximum temperature, we
performed the projection of the patients with heatstroke in case of considering only climate change (Case 1);
climate change and population dynamics (Case 2); and climate change, population dynamics, and long-term heat
acclimatization (Case 3). In Case 2, the number of patients with heatstroke in the near future will be 2.3 times
higher than that in the baseline period (1981 – 2000) on average nationwide. The number of future patients with
heatstroke in Case 2 is about 10 % larger than that in Case 1 on average nationwide despite population decline.
This is due to the increase in the number of elderly people from the baseline period to the near future. However,
in 20 prefectures, the number of patients in Case 2 is smaller compared to Case 1. Comparing the results from
Cases 1 and 3 reveals that the number of patients with heatstroke could be reduced by about 60 % nationwide
by acquiring heat tolerance and changing lifestyles. Notably, given the lifestyle changes represented by the widespread use of air conditioners, the number of patients with heatstroke in the near future will be lower than that
of the baseline period in some areas. In other words, lifestyle changes can be an important adaptation to the risk
of heatstroke emergency. All of the above results were also confirmed in the prediction model with WBGT as the
explanatory variable.

Corresponding author: Hiroyuki Kusaka, Center for Computational Sciences, University of Tsukuba, 1-1-1, Tennodai,
Tsukuba-shi, Ibaraki 305-8577, Japan
E-mail: kusaka@ccs.tsukuba.ac.jp
1 Present affiliation: Sompo Risk Management Inc., Tokyo,
Japan
J-stage Advance Published Date: 8 April 2022
©The Author(s) 2022. This is an open access article published by the Meteorological Society of Japan under
a Creative Commons Attribution 4.0 International (CC BY 4.0) license (https://creativecommons.org/licenses/by/4.0).

598

Journal of the Meteorological Society of Japan

Vol. 100, No. 4

Keywords  number of patients with heatstroke; near future projection; heat acclimatization; climate change adaptation; generalized linear model
Citation  Nakamura, S., H. Kusaka, R. Sato, and T. Sato, 2022: Heatstroke risk projection in Japan under current
and near future climates. J. Meteor. Soc. Japan, 100, 597–615, doi:10.2151/jmsj.2022-030.

1. Introduction
In recent years, the incidence of heatstroke in
Japan has increased due to climate change, and this is
becom­ing a major social issue (e.g., Ando et al. 2004;
Fujibe 2013). For example, from May to September
2018, which was an abnormally hot summer across
the country, the number of emergency patients with
heatstroke was 95,137 nationwide, of which 32,496
were hospitalized and 160 died (Fire and Disaster
Management Agency 2019, https://www.fdma.go.jp/
disaster/heatstroke/item/heatstroke003_houdou01.
pdf). In 2018, the number of deaths due to heatstroke
was 1,581. This number of deaths is far greater than
the number of deaths caused by other weather-related
disasters, such as floods and landslides (the number of
deaths from the 2018 Japan floods, which were one of
the most torrential in decades, was 225). Residents are
concerned that heatstroke will become increasingly
serious as climate change progresses. Therefore, it is
important to assess all the risks associated with heatstroke in a future climate.
Extensive studies on the increase in heat-related
excess mortality or deaths associated with future climate change have been conducted mainly in Europe,
the United States, Japan, and China (e.g., Hayhoe
et al. 2004; Knowlton et al. 2007; Doyon et al. 2008;
Gosling et al. 2009; Jackson et al. 2010; Li et al.
2013; Honda et al. 2014). Li et al. (2013) predicted
that future heat-related excess deaths in New York,
USA, under the Special Report on Emissions Scenarios (SRES) A2 scenario, would increase by +22.2 %
(2020s), +49.4 % (2050s), and +91.0 % (2080s),
compared to levels in the 1980s. Doyon et al. (2008)
predicted a 10 % increase in summer heat-related
mortality in Montreal, Canada, in 2080, compared
to that in 1981 – 1999 under the SRES A2 scenario.
Similar studies have continued to be conducted after
the release of the future climate projection datasets for
the Representative Concentration Pathway (RCP) scenarios (Chen et al. 2017; Huber et al. 2020). In recent
years, projections have also been conducted in developing countries, including those in Southeast Asia.
Gasparrini et al. (2017) projected heat-related excess
mortality rates of more than 5 % in Southeast Asia,

Central and Southern Europe, and Latin America in
the 2090s under the RCP8.5 scenario. Guo et al. (2018)
predicted that heat-related deaths would increase by
more than 700 % in some Southeast Asian and South
American countries during the period of 2031 – 2080
under the RCP8.5 scenario compared to the 1971 – 
2020 period. Thus, future projections of heatstroke
risk have been dominated by studies that use heatrelated excess mortality or deaths as indicators. In
these studies, it is necessary to consider not only
climate change but also social change. Social changes
include demographic changes and long-term heat
acclimatization over a span of several decades due to
lifestyle changes. Among the previous studies, those
that consider demographic changes include Gosling
et al. (2009), Jackson et al. (2010), Honda et al. (2014),
Chen et al. (2017), and Guo et al. (2018). Studies considering long-term heat acclimation include Hayhoe
et al. (2004), Knowlton et al. (2007), Gosling et al.
(2009), Li et al. (2013), and Guo et al. (2018).
Therefore, the main purpose of this study is to
develop a statistical model and predict heatstroke
risk (the number of ambulance transport due to heatstroke) in the near future (2031 – 2050) under RCP2.6
and RCP8.5 scenarios all over Japan by prefecture.
This statistical model is based on the generalized
linear model, which uses maximum temperature or
WBGT as explanatory variable and daily number of
ambulance transport due to heatstroke as a predictor
variable. When predicting the number of ambulance
transport due to heatstroke by statistical model, it
is known that there is a problem of underestimation
in early summer and overestimation in late summer
(Fuse et al. 2014; Sato et al. 2020; Ikeda and Kusaka
2021). This error is due to short-term heat acclimatization (Ono 2013; Fujibe et al. 2018b). Therefore,
our model takes this effect into account. The near
future heatstroke risk is determined by three types of
experiments, namely, (i) future projection considering
only climate change, (ii) future projection considering
climate change and population, and (iii) future projection considering climate change, population, and longterm acclimatization. Section 3 describes the detailed
information of experiments.

August 2022

S. NAKAMURA et al.

2. Data
2.1  Number of heatstroke emergency patients
This study used a dataset on the number of ambulance transport due to heatstroke for 2010 – 2018 published by the Fire and Disaster Management Agency
of the Ministry of Internal Affairs and Communications, Japan.
Heatstroke is defined as “a general term for any disorder that results from an imbalance of water and salt
(e.g., sodium) in the body due to a breakdown in the
body’s ability to regulate the temperature in a hightemperature environment” and includes sunstroke,
heat cramps, and heat exhaustion (Fire and Disaster
Management Agency 2021). Based on the above definition, a medical doctor determines whether the patient brought to the emergency room has a heatstroke.
This study used the data on the number of emergency
patients with heatstroke by a medical doctor’s initial
diagnosis. There are three types of age-related data
in this dataset: the number of heatstroke emergency
patients per day by prefecture in all age groups, aged
65 years and older, and under 64 years old (newborn
babies, infants, juveniles, and adults combined). The
number of ambulance transport due to heatstroke is
simply called “the number of patients with heatstroke”
and is used as an indicator of heatstroke risk in this
study.
2.2  Current climate data
The temperature data were taken from hourly observations made by the Automated Meteorological Data
Acquisition System (AMeDAS) operated by the Japan
Meteorological Agency (JMA). AMeDAS stations
are located at a density of approximately 20 km. We
used the spatial average of all stations’ values within a
prefecture to improve the spatial representativeness of
the temperature value used for each prefecture. However, because the Tokyo’s climate differs markedly
between the mainland and the islands, spatial averages
of Tokyo are calculated by excluding data from observation stations on the islands (these islands account
to 0.2 % of Tokyo’s total population). The daily maximum temperatures were determined from the hourly
temperature values obtained from these averages.
WBGT was calculated using the formula of Yaglou
and Minard (1957). The black globe temperatures
there that are not measured by JMA were estimated by
the method of Okada and Kusaka (2013). The daily
maximum WBGT was calculated from the hourly
values of WBGT. Supplement 1 describes the detailed
methods for estimating the WBGT.

599

2.3  Climate scenario data
As the climate scenario data, we used the 1-km
mesh statistical downscaling (DS) dataset provided by
Institute for Agro-Environmental Sciences, National
Agriculture and Food Research Organization (NARO)
(Nishimori et al. 2019). This DS dataset were created
from four GCMs outputs, i.e., MIROC5, MRI-CGCM3,
GFDL-CM3, and HadGEM2-ES. These GCMs were
carefully selected by SI-CAT, project for climate
change adaptation in Japan. For the period of climate
scenarios used in this study, the baseline period is set
to 1981 – 2000, and the near future is set to 2031 – 2050.
Unfortunately, the NARO dataset stores only data
for daily (mean, maximum, and minimum) and monthly mean values and not hourly values. Due to this
limitation, it is impossible to calculate the daily
maximum WBGT with only this dataset. In addition,
it should be noted that the reliability of each meteorological variable differs. In fact, it is reported that
the reliability of air temperature and solar radiation is
relatively high, while that of humidity and wind speed
is relatively low (Nishimori et al. 2019).
In this study, a similar idea as the pseudo-global
warming approach (Kimura and Kitoh 2007; Sato
et al. 2007) was applied to estimate the future WBGT
to overcome these problems. First, a time series of
daily maximum temperature from June 1 to September
30 is generated using the baseline period data from
NARO’s dataset. Second, this time series is averaged
over 15 days and then averaged over 10 years. Third,
similar time series data is generated using the future
climate scenario data of NARO’s dataset. From the
difference between these two-time series, the climate
change component data (ΔT ) was obtained. This ΔT is
daily data of the amount of temperature increase from
the present to the future, containing a gentle seasonal
change. The pseudo-future dry bulb temperature is
estimated from the actual temperature of the present
climate T plus future temperature increase ΔT. The
pseudo-future WBGT is estimated using pseudo-future
dry bulb temperature (Td = T + ΔT ), wet bulb temperature (Tw), and globe temperature (Tg). Here, the
future Tw should be calculated from the future relative
humidity and the pseudo-future temperature (T + ΔT ).
However, in this study, the pseudo-future Tw is calculated from the current relative humidity and the pseudofuture temperature, considering the result of the previous study indicating that the relative humidity does not
change significantly in Japan in the near future (Byrne
and O’Gorman 2016). Similarly, the pseudo-future
Tg is calculated from the current solar radiation, wind
speed, and the pseudo-future temperature.

600

Journal of the Meteorological Society of Japan

2.4  Population data
As the current (baseline) population data by prefecture, we used the data from the 1990 Population
Census. As the future population data by prefecture,
we used the “Future Population Estimates by Region
for Japan” provided by the National Institute of Population and Social Security Research (National Institute
of Population and Social Security Research 2018).
This dataset is a statistical future projection of the
population by prefecture and municipality. This data
is suitable for the purpose of this study because it is
estimated by age group (0 – 14 years, 15 – 64 years, 65
years and older, and 75 years and older).
Here, the population data is the nighttime population for both base and near future values. If the
population of a prefecture is expressed using nighttime population, there will be an error in the risk of
heatstroke if a person suffers from heatstroke during
the daytime in a prefecture other than his or her home.
However, this error is expected to have only a little
effect on the predictions of this study for the following
two reasons. The first reason is that the difference
between the daytime and nighttime populations is
small except in a few prefectures. According to the
2005 census, the difference between the daytime and
nighttime populations is about 20 % even in Tokyo,
where the daytime population is much larger than the
nighttime population, and about 12 % even in Saitama,
where the daytime population is much smaller than
the nighttime population. In other prefectures, the
difference between the daytime and nighttime populations was less than 10 %. The second reason is that
most people suffering from heatstroke are young children and the elderly. Since the difference between the
daytime and nighttime populations occurs mainly in
the age group that commutes to work or school, these
are different age groups from the young children and
elderly.
3. Method
3.1  Model overview
In this study, the six models presented in Table 1
were created and compared for accuracy. The characteristics of the proposed models for the number of
patients with heatstroke prediction are as follows:
(i) The model is based on generalized linear models
(GLM, Nelder and Wedderburn 1972).
(ii) The predictor variable is the number of heatstroke emergency patients.
(iii) The default explanatory variable is the daily
maximum temperature (but we can also use
WBGT instead).

Vol. 100, No. 4

Table 1.  List of models that were compared for accuracy.
Fitted Data
Tokyo
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6

○
○

Each
Prefecture

○
○
○
○

Period
Division

Age
Group
○

○
○

○
○

(iv) Differences in regional, seasonal (short-term heat
acclimatization), and age of heatstroke risk were
considered when identifying the model parameters.
Regarding (i), the GLM equation is expressed as
follows:
log ( y) = α + β x ,

(1)

where x is the explanatory variable, y is the objective
variable, and α and β are partial regression coefficients (parameters). Each parameter was identified by
the maximum likelihood method, assuming a Poisson
distribution. First, as a default model, we created a
model with the estimated parameters using data from
Tokyo and adapted the model to the entire country.
Regarding (ii), the results of this model will provide
useful information for examining the requirements
of the emergency medical system, considering the
increase in the number of patients with heatstroke due
to future climate change.
Regarding (iii), it is expected that the use of the
daily maximum temperature leads to a high practicality in making future predictions. This is because
the humidity, wind speed, and solar radiation used
in the WBGT estimation have a tendency with lower
availability and robustness of future climate scenario
data, compared with temperature. On the other hand,
WBGT is possibly more suitable for explanatory
variables under current climate than temperature.
These pros/cons are trade-off relationship for future
projection; thus, we compare the accuracies between
the two models: one uses temperature as the explanatory variable, and the other uses the WBGT. We then
individually predict future heatstroke risk using the
two models. The comparison of such models might
be an important attempt to understand the uncertainty
among prediction models.
Regarding (iv), it is expected that the proposed
model will improve the accuracy of the future pro-

August 2022

S. NAKAMURA et al.

601

Fig. 1.  An example of period division used in this study.

jection of the number of emergency transport due to
heatstroke by considering the factors not limited to the
meteorological field. Sections 3.2 – 3.4 describe these
factors in detail.
3.2 Consideration of regional dependency in the
model
The degree of heat tolerance of people is known to
vary among regions (Keatinge et al. 2000; Curriero
2002; Gosling et al. 2007; Fujibe et al. 2018a). For example, when exposed to the same temperature, people
in the cooler regions of northern Japan have a higher
risk of heatstroke that people in warmer regions (Fujibe
et al. 2018a). To account for these regional differences
in heat tolerance, a parameter estimation for each
prefecture individually was performed.
3.3 Consideration of short-term heat acclimatization
in the model
The predictions calculated from Eq. (1) are problematic in that they underestimate the predictions in
the early summer and overestimate the predictions in
the late summer. This is because the effect of shortterm acclimatization is not included when using a
single equation as described before. Like Ikeda and
Kusaka (2021), using an actual number of patients
with heatstroke 1 day before and the cumulative days
from the start of summer season as explanatory vari-

ables is an example of ways to consider the short-term
acclimatization effect. However, the actual number
of patients with heatstroke cannot be used under the
future climate projection. Cumulative days might be a
useful idea in the future projection because it indicates
the number of hot days experienced in one summer.
However, it cannot be applied to the model in this
study because the timing of midsummer may change
in the long term; in that case, simple cumulative days
may not be able to represent this change.
In this study, we propose the method to divide the
predicted period from June to September into three
subperiods, i.e., early summer, midsummer, and late
summer, based on the time series of daily maximum
temperature (Fig. 1). The equations are respectively
constructed for early summer and late summer using
data in these subperiods (Eqs. 2, 3) to consider the
effect of short-term acclimatization. These equations
are respectively used in early summer and late
summer instead of Eq. (1).
log ( yp1) = α p1 + β p1 x ,
log ( yp3) = α p3 + β p3 x .

(2)
(3)

As mentioned above, if Eq. (1) is used for the entire
summer, it will underestimate the number of emergency cases in early summer and overestimate the number
of emergency cases in late summer. In this study, in
order to mitigate these errors, we divided the period

602

Journal of the Meteorological Society of Japan

into three parts, focusing on the temperature increase
from early summer to midsummer and the temperature
decrease from midsummer to late summer. The period
division was carried out using the values of posterior
5-day mean minus previous 5-day mean (hereafter
referred to as the “5-day mean difference”). This fiveday mean difference represents the trend of temperature change in about 10 days. When temperature rises
over a span of about 10 days, the 5-day mean difference shows a positive value. The method of period
division is presented. Figure 1 shows an example of
this method.
• Start date of the early summer period: June 1.
• End date of the early summer period: 7 days after
the last day when the value of the 5-day mean
difference exceeded the threshold. This end date is
selected in the period from June 1 to August 9. The
thresholds are 50 – 95th percentile of the 5-day mean
difference and set by prefectures. For example, at
Fukuoka in 2018, the end date of the early summer
period is set to August 9 (the end of the period
shown in orange in Fig.1). If the date selected is on
or after August 10, the end date of the early summer
period is uniformly set to August 9. This is because
the tendency to underestimate the prediction values
generally finishes by early August in any year.
• Start date of the late summer period: The date when
the value of the 5-day mean difference falls below
the threshold for the first time during the period
from August 10 to September 30. The thresholds
are 5 – 50th percentile of the 5-day mean difference
and set by prefectures. For example, at Fukuoka in
2018, the start date of the late summer period is set
to August 14 (the start of the period shown in blue
in Fig. 1).
• End date of the late summer period: September 30.
• Midsummer period: From the day after the end of
the early summer period to the day before the start
of the late summer period (the period shown in
green in Fig. 1). In midsummer period, the error in
the predictions based on the non-division model is
enough small, and there is no need to revise them.
3.4 Consideration of differences in patient’s age in
the model
It is well known that the risk of heatstroke is higher
in the elderly than in the young (Nakai et al. 1999;
Smoyer et al. 2000a; McGeehin and Mirabelli 2001;
Basu and Samet 2002; Flynn et al. 2005; Hajat et al.
2007; Anderson and Bell 2009). Therefore, to account
for these differences in heatstroke risk by age, we
separately predicted the number of patients with heat-

Vol. 100, No. 4

stroke 65 and older and under 64 years of age (Fig. 2).
3.5  Factors not considered in the model
The following factors related to the heatstroke risk
are not used in the prediction model: (i) sex (Semenza
et al. 1996; Whitman et al. 1997; Havenith 2005;
Vaidyanathan et al. 2020), (ii) use of air conditioners
or air conditioner penetration rate (Semenza et al.
1996; Basu and Samet 2002; Anderson and Bell 2009),
(iii) socioeconomic status (Anderson and Bell 2009;
Hondula et al. 2015; Fujibe et al. 2020), (iv) whether
they are living in a nursing home or not (Kovats and
Hajat 2008), (v) clinical or pathophysiological factors,
(vi) urban heat islands (Kovats and Hajat 2008), and
(vii) air pollution levels (Piver et al. 1999).
(i) In this study, sex could not be considered because
the dataset on the number of heatstroke emergency patients did not distinguish between men and
women.
(ii) In most prefectures, the penetration rate of air
conditioners is around 90 %. The presence or absence of air conditioner use may have something
to do with the presence or absence of heatstroke
occurrence, but it is difficult to obtain such data
at the national level. For this reason, this factor is
not used in the prediction model.
As for (iii) and (iv), in Japan there is almost no
gap between the rich and the poor, and social security
and medical insurance are almost well provided for
all citizens. Thus, air conditioners are considered to
be sufficiently widespread for nursing care facilities.
Regarding (v), predicting what will happen to the
number of people with diseases related to heatstroke
risk in the future (whether it will increase or decrease)
is highly uncertain and unrealistic. Regarding (vi),
Japan’s cities are already mature, and it is unlikely
that further urbanization will enhance the heat island
effect (Adachi et al. 2012; Kusaka et al. 2016).
Regarding (vii), the effect of air pollutants on heatstroke is smaller than the effect of temperature (e.g.,
Shumway et al. 1988; Smoyer et al. 2000b; Rainham
and Smoyer-Tomic 2003). The impact of air pollutants
on heatstroke in Toronto in 1980 – 1996 was small
(Rainham and Smoyer-Tomic 2003). During that
period, the NO2 concentration in Toronto was 0.0238
ppm, while the NO2 concentration in Tokyo in 2018
was 0.015 ppm. In addition, air pollutants in Tokyo
have been decreasing in recent years and are expected
to continue to decrease in the future (Morikawa et al.
2021). Therefore, air pollutants are not considered in
this study.
In addition, this study did not consider the geospa-

August 2022

S. NAKAMURA et al.

603

Fig. 2.  Scatterplot showing the relationship between the daily maximum temperature and the number of patients in
Fukuoka Prefecture in 2018. Red, green, and blue plots indicate early summer, midsummer, and late summer periods, respectively. The lines denote prediction equations fitted from the data indicated by the plots. The scatterplot (a)
shows the number of patients who are under 65 years of age. The scatterplot (b) shows the number of patients who
are 65 years of age or older. The scatterplot (c) shows the number of patients who are all ages.

tial population density pattern within a prefecture.
However, if it is considered, the risk of heatstroke can
be assessed in more spatial detail. This will be useful
information for the optimal allocation of medical
facilities.

human models. In this study, we used the daily maximum WBGT as explanatory variable as well as daily
maximum temperature and investigated the effect of
different explanatory variables on the prediction accuracy.

3.6  Changing explanatory variables in the model
The thermal indices, WBGT (Yaglou and Minard
1957) and Universal Thermal Climate Index (UTCI;
Fiala et al. 2012), are widely used to measure heatstroke risk in the world. In Japan, WBGT is the most
widely used and recognized as an effective guideline
for work and exercise environments. Moreover,
WBGT has been standardized internationally by the
International Organization for Standardization. The
UTCI is often used worldwide, but its application to
Japanese people is considered questionable as it is
based on the physiological responses of Caucasian

3.7  Verification of model accuracy
Cross-validation was performed with any 1 year of
data from 2010 to 2018 as test data and the remaining
8 years as training data. The predictive accuracy of the
models was assessed by mean absolute error (MAE)
and root mean square error (RMSE). Models with
small values of each of these parameters were considered to have higher predictive accuracy.
3.8  Design of baseline and near future projection
First, we will estimate the number of patients with
heatstroke in the baseline period (1981 – 2000) using

604

Journal of the Meteorological Society of Japan

Table 2.  List of future projection experiments and featured
factor.
Climate
Change Population
Scenarios
Case 1
Case 2
Case 3a
Case 3b

RCP 8.5
RCP 8.5
RCP 8.5
RCP 8.5

1990
2040
2040
2040

Long-term
Acclimatization
―
―
Late summer equation
Climate analog

statistical models developed in Chapter 3 by prefecture. Second, we will perform the future projection of
heatstroke risk in Japan by prefecture. In this study,
heatstroke risk means the number of patients with
heatstroke, as described in Section 1. We use Model
6 in Table 1 for future projection of the number of
patients with heatstroke. We perform two sensitivity
experiments (Cases 2, 3) in addition to control experiment (Case 1) to discuss the uncertainty of future
projection results. Table 2 summarizes the future pro­
jection experiments.
• Case 1: Future projection considering neither near
future demographics nor long-term acclimatization
into account.
• Case 2: Future prediction considering only the near
future demography.
• Case 3: Future prediction considering both near
future demography and long-term acclimatization.
Case 1 is an experiment to evaluate the increase in
the risk of heatstroke due solely to the increase in temperature caused by climate change. In this experiment,
the number of patients with heatstroke in the entire
region is used as the risk indicator, but it is assumed
that the demographics will not change between now
and the future. In other words, the increase in risk in
this experiment is the same as the increase in the risk
of heatstroke for each individual resident.
Case 2 is an experiment to evaluate the variation in
the risk of heatstroke by considering the temperature
increase due to climate change and demographic
change from the baseline period to the near future. In
this experiment, we can obtain the projected number
of patients with heatstroke for the entire region at
each time point in the baseline period and near future.
Thus, this future projection can assess the risks related
to the burden on the emergency medical system associated with an increase in the number of patients with
heatstroke. The burden on the emergency medical
system refers specifically to the shortage of emergency
transport systems and inpatient beds, as indicated in
Chapter 1. Therefore, the results of this future projec-

Vol. 100, No. 4

tion are expected to be very useful for the government
to formulate adaptation measures to climate change.
Heat acclimatization is known to occur over a long
period of time, apart from short-term acclimatization
throughout the single summer. Petkova et al. (2014)
noted that the excess mortality observed between 1973
and 2006 was much lower than that observed between
1900 and 1948, indicating that people have become
acclimatized to heat during this period. They concluded that this acclimatization is due to the improvement
of the living environment and the widespread use of
air conditioners. Therefore, in this study, experiments
(a) and (b) are conducted to evaluate long-term heat
acclimatization from the baseline to the near future.
In both Cases 3a and 3b, population dynamics were
considered.
(a) An experiment in which individuals are assumed
to have heat tolerance equivalent to late summer
throughout one summer season (Case 3a).
(b) An experiment using a climate analog to account
for lifestyle changes in a cold region with particularly low air-conditioning penetration (Case 3b).
In the prediction experiment of Case 3a, we particularly examine the effect of long-term acclimatization
due to the acquisition of heat tolerance. Equation (3)
for late summer, described in Section 3.3, is used to
predict the number of patients with heatstroke in near
future over the entire summer period, including early
and midsummer. This is based on the assumption
that the government and individuals will have heat
tolerance equivalent to that of late summer throughout
the entire summer period by taking all kinds of heat
countermeasures.
In the prediction experiment of Case 3b, we examined the effects of long-term acclimatization due to
the acquisition of heat tolerance and lifestyle changes.
In this experiment, the target areas are Hokkaido,
Aomori, Iwate, Miyagi, Akita, Yamagata, Fukushima,
Nagano, and Yamanashi. These areas have low percentages of households with air-conditioning during
the baseline period. We first looked for three prefectures with a current daily maximum temperature that
is close to the near future daily maximum temperature
of a target prefecture. Using the prediction models of
the selected three prefectures, the near future projections were then made for the target prefecture. This
procedure was finally conducted for nine target prefectures with low air conditioner penetration rate today.
This method is a kind of climate analog approach (e.g.,
Ishizaki et al. 2012). This near future prediction is
based on the assumption that the inhabitants of the regions with low air conditioner penetration rates in the

August 2022

S. NAKAMURA et al.

605

baseline period will acquire the same heat tolerance or
change their lifestyles as those of other regions with
similar climates in the near future.
The targeted nine prefectures (Hokkaido, Aomori,
Iwate, Miyagi, Akita, Yamagata, Fukushima, Nagano,
and Yamanashi) had particularly low air conditioner
penetration rates in 1999 (9.3 % in Hokkaido, 30.2 %
in Aomori, 35.6 % in Iwate, 59.1 % in Miyagi, 56.7 %
in Akita, 67.8 % in Yamagata, 58.4 % in Fukushima,
44.8 % in Nagano, and 72.0 % in Yamanashi). The
air conditioner penetration rates in the other prefectures are all above 80 % (based on the 1999 National
Survey of Actual Consumption, https://www.e-stat.
go.jp/dbview?sid=0000111013).
The future projections are carried out using daily
maximum WBGT instead of daily maximum temperature as an explanatory variable. Section 2.3 and
Supplement 1 describe the method of calculating the
daily maximum WBGT in baseline and near future.
4. Accuracy of the proposed statistical models
under the current climate
4.1 Improvement in model accuracy by considering
regional and short-term heat acclimatization and
age
First, we developed a model to predict the number
of heatstroke emergency patients using the daily
maximum temperature data for Tokyo and conducted
prediction experiments and accuracy verification
(cross-validation) for each prefecture (Model 1). The
prediction errors of the Model 1 were 5.5 (MAE) and
10.6 (RMSE), on average, across the country.
Second, we performed prediction with Model 3
and compared the results between Models 1 and 3. As
a result, it was confirmed that the MAE could be reduced by about −19 % (−46 % to −3 % in each prefecture) and the RMSE by about −25 % (−48 % to −0 %
in each prefecture) on average, across the country by
considering regional characteristics (Fig. 3).
Third, we performed prediction with Model 5 and
compared the results between Models 3 and 5. From
the results, we found that considering the short-term
heat acclimatization (i.e., effect of Model 5) reduced
the MAE by about 12 % (−22 % to −3 % in each prefecture) and the RMSE by about 12 % (−20 % to −4 %
in each prefecture) on average, across the country.
Last, we compared errors between the odd-numbered
model group (Models 1, 3, and 5) with the evennumbered model group (Models 2, 4, and 6), indicating that the prediction accuracy on average, across the
country, remained almost unchanged when differences
in risk by age were considered.

Fig. 3.  (a) MAE and (b) RMSE of the number of
patients in 2018 predicted using each model. Box
whiskers represent the range in values obtained
for 46 regions. To remove the effect of population
size, MAE and RMSE were plotted as normalized values per 10,000 people.

We explicitly show the effect of improving the accuracy after considering the period division (i.e., Model
5 effect) using data for 2018 Fukuoka Prefecture (one
of the major prefectures in Japan) as an example from
the cross-validation results. In 2018, a severe heat
wave was experienced across Japan. Thus, predicting
the number of patients with heatstroke in 2018 using
climate data from 2010 to 2017 is a good example
for a prediction experiment for a warmer future using
standard summer data. The results showed that the
early summer period is characterized by having a
relatively high number of patients with heatstroke,
and the late summer period is characterized as having
relatively fewer patients (Fig. 4). The same was also
confirmed in many prefectures other than Fukuoka.
Figure 4 shows the time series of daytime predictions
obtained from the model with and without period division and benchmark model (i.e., Models 1, 3, vs. 5).
It can be seen that the model without period division
(Model 3) significantly underestimates the peak in the
number of patients from early July to early August.
It also tends to overestimate the peak in mid to late
August. On the other hand, these tendencies of under-

606

Journal of the Meteorological Society of Japan

Vol. 100, No. 4

Fig. 4.  Time series of the daily maximum temperature and actual and predicted number of patients in Fukuoka Prefecture in 2018. The black line is the daily maximum temperature, the gray bar is the observed number of patients,
the blue line is the number of patients predicted by the benchmark model (Model 1), the green line is the number
of patients predicted by the model that fitted with data for each prefecture (Model 3), and the orange line is the
number of patients predicted by the model that considered short-term heat acclimatization (Model 5).

estimation and overestimation are greatly improved
in the model with period division (Model 5) (32 %
reduction in MAE and 29 % reduction in RMSE).
4.2 Effect of different explanatory variables on
prediction accuracy
The explanatory variables with the highest prediction accuracy for each region were investigated for
the predictions obtained using Model 6. From the perspective of MAE (Fig. 5), the daily maximum WBGT
would be selected as the best explanatory variable in
27 of the 46 regions. From the perspective of RMSE
(Fig. 6), the daily maximum WBGT would be selected as the best explanatory variable in 31 of the 46
regions. These results suggest that WBGT is a better
explanatory variable than daily maximum temperature
in predicting the number of patients with heatstroke.
This is consistent with studies that have shown that
humidity is an important explanatory variable for
heatstroke risk (Zhang et al. 2014; Sherwood 2018).
However, in the majority of prefectures, the difference
in the error between the temperature models and
WBGT models was less than 10 %, with a maximum
of 20 % (MAE) and 25 % (RMSE).
5. Future projection of the number of patients
with heatstroke
5.1 Baseline
Figure 7 shows the estimated total number of
patients with heatstroke per summer (averaged for 20
years × 4 GCMs) for the baseline period. The figure
shows that the average total number of patients with
heatstroke in all prefectures is 3.8/10,000 per summer,
with a spread from a maximum of 6.3/10,000 (Kago­
shima) to a minimum of 1.6/10,000 (Hokkaido) by

prefecture. This spread reflects the regionality of both
the temperature spread and tolerance to the heat.
5.2 Result of near future projection-only effect of
climate change: Case 1
Figure 8a shows a map of future changes in the risk
of heatstroke (for Case 1). The figure indicates that
the average total number of patients with heatstroke
in all prefectures is 8.9/10,000 per summer, with a
large spread from the maximum value of 18.6/10,000
(Kago­shima) to the minimum value of 5.2/10,000
(Tokyo) by prefecture.
Figure 9 shows the rate of increase in the number
of patients with heatstroke from the baseline period
(1981 – 2000) to the near future (for Case 1) on average nationwide. This figure indicates that the number
of patients with heatstroke in the near future will be
1.2 – 2.9 times (2.1 times in the ensemble average of
4 GCMs) in the case of RCP2.6 scenario and 1.4 – 3.3
times (2.2 times in the ensemble average of 4 GCMs)
in the case of RCP8.5 compared to the baseline
period. This range of values is due to the uncertainty
of the GCMs. Since no significant difference in the
prediction results is found between the RCP2.6 and
RCP8.5 scenarios due to near future projection, we
will only discuss the prediction results for RCP8.5
from now on. The regions with the highest increase
in the heatstroke risk from the baseline period to the
near future are Hokkaido, northern Tohoku, southern
Kanto, Tokai, and Kyushu (Fig. 10a) (see Fig. S1 in
Supplement 2 for the names of Japanese prefectures
and regional categories). The prefecture with the highest rate of increase was Hokkaido, with 313.6 %. One
reason may be that Hokkaido has experienced a larger
increase in temperature due to climate change (about

August 2022

607

S. NAKAMURA et al.

Fig. 5.  Better explanatory variables (daily maximum temperature or daily maximum WBGT) for prediction. MAE
is used as an evaluation criterion for prediction accuracy. Model 6 was used. Green: Prefectures where the daily
maximum temperature model produces higher prediction accuracy. Blue: Prefectures where the daily maximum
WBGT model produces higher prediction accuracy. White: Prefectures where the difference in the prediction
between the daily maximum temperature model and the daily maximum WBGT model is 4 % or less. The color
shading represents [1-(MAE of the model with high accuracy)/(MAE of the model with low accuracy)*100 (%)].

Fig. 6.  Better explanatory variables (daily maximum temperature or daily maximum WBGT) for prediction. RMSE
is used as an evaluation criterion for prediction accuracy. Model 6 was used. Green: Prefectures where the daily
maximum temperature model produces higher prediction accuracy. Blue: Prefectures where the daily maximum
WBGT model produces higher prediction accuracy. White: Prefectures where the difference in the prediction
between the daily maximum temperature model and the daily maximum WBGT model is 4 % or less. The color
shading represents [1-(RMSE of the model with high accuracy)/(RMSE of the model with low accuracy)*100 (%)].

Fig. 7.  The number of patients with heatstroke per 10,000 people (average per
summer) during the baseline period
(1981 – 2000) estimated by the prediction
model.

608

Journal of the Meteorological Society of Japan

Vol. 100, No. 4

Fig. 8.  Predicted number of patients with heatstroke (per 10,000 population) under the RCP8.5 scenario of the near
future climate, using daily maximum temperature as the explanatory variable. (a) Prediction without population
dynamics (Case 1), (b) prediction with population dynamics (Case 2), (c) prediction using the late summer equation (Case 3a), and (d) prediction using the climate analog (Case 3b). The areas shaded by gray color are outside of
analysis target.

Fig. 9.  The rate of increase in the number of patients with heatstroke in Japan from the baseline
to the near future. The relative value when the
number of patients with heatstroke during the
baseline period is set to 1.

2.2°C increase) than other regions (see Figs. S2a, b in
Supplement 3).
5.3 Result of future projection with population
dynamics: Case 2
Figure 8b shows the risk map of patients with

heat­stroke in the near future (2031 – 2050) obtained
from the future prediction experiment of Case 2. The
figure indicates that the total number of patients with
heatstroke nationwide is 9.6/10,000 per summer,
with a large spread from a maximum of 20.4/10,000
(Kagoshima) to a minimum of 5.7/10,000 (Tokyo) by
prefecture.
Figure 10b shows a map of the increase rate in the
number of patients with heatstroke from baseline to
the near future (under RCP8.5 scenario) for each prefecture of Case 2. On average nationwide, the increase
rate in the number of patients with heatstroke from
baseline period to the near future obtained from Case
2 is 234.4 % in the ensemble mean of four GCMs.
This increase rate on the average nationwide is about
10 % larger than that in Case 1. The reason must come
from the differences between Cases 1 and 2, i.e., (i)
the increase in total population from the baseline to
the near future, (ii) the increase in the elderly population, or (iii) both. Let us now consider which of these
three factors was dominant. The population of Japan
in the baseline (1990) is about 120 million, while the
population in the near future (2040) will be about 110
million. Therefore, if the experiment only considers
the increase or decrease in population, the number of
patients with heatstroke in Case 2 should be smaller
than in Case 1. This means that the reason for the

August 2022

S. NAKAMURA et al.

609

Fig. 10.  The rate of increase in the patients with heatstroke from the baseline period to the near future (RCP8.5
scenario) using daily maximum temperature as the explanatory variable. (a) Prediction without population dynamics (Case 1), (b) prediction with population dynamics (Case 2), (c) prediction using the late summer equation (Case
3a), and (d) prediction using the climate analog (Case 3b). The areas shaded by gray color are outside of analysis
target.

increase in the number of patients with heatstroke
is the increase in the elderly population. In fact, the
proportion of elderly people in the total population has
almost tripled from 12.0 % to 35.3 % from baseline
to near future. In all prefectures, the increase rate was
higher than 100 %. We can see that the increase rate
is high in prefectures with large population, including
the Tokyo metropolitan area and other major urban
areas. Among these prefectures, the difference in the
prediction between Case 1 and Case 2 is largest in
Tokyo, where the rate of future increase is 360.0 %
in Case 2 but 239.3 % in Case 1. The population of
Tokyo as a whole increases by 16.6 % from baseline
to the near future, and the aging rate also increases by
18.6 % from the baseline to the near future. In other
words, in Tokyo, the risk of heatstroke in Case 2 was
particularly high compared to Case 1 due to two effects, i.e., total population increase and increase in the
aging rate from the baseline period to the near future,
in addition to climate change.
The demographic changes from the baseline to the
near future can be classified into the following four
patterns for each prefecture.
(1) The population of the prefecture increases, and the
proportion of elderly people in the total population
also increases (Tokyo type).
(2) The population of the prefecture increases, but the

proportion of elderly people in the total population
decreases.
(3) The population of the prefecture decreases, but the
proportion of elderly people in the total population
increases.
(4) The population of the prefecture decreases, and the
proportion of elderly people in the total population
decreases.
In type (1), the number of patients with heatstroke
is definitely higher in Case 2 than in Case 1 where
only the temperature increases considering climate
change. However, in the case of type (3), the results of
future projections will depend on whether the decline
in population or the increase in the aging rate is dominant. No prefectures corresponded to types (2) and
(4) (i.e., prefectures where the population aging rate
decreases from the baseline to the near future).
As a result of comparing Cases 2 and 1, we found
that the number of patients with heatstroke was
higher in Case 2 in 26 out of 46 prefectures. Of the 26
prefectures, 6 prefectures, including Tokyo, were classified as type 1 (Tokyo type). In these prefectures, the
number of patients with heatstroke will increase due
to the following three factors: (1) climate change, (2)
population growth, and (3) increase in the aging population. The remaining 20 prefectures were classified
as type 3. In these prefectures, the number of patients

610

Journal of the Meteorological Society of Japan

Table 3.  Patterns of change in population and increase/
decrease in risk of heatstroke emergencies from the baseline period to the near future.
The proportion of elderly people
in the total population
Increase

Population

Decrease

Increase

Prefectures at
increased risk: 6

―

Decrease

Prefectures at
increased risk: 20
Prefectures at
decreased risk: 20

―

with heatstroke will increase due to climate change
and an increase in the aging population. Among
these 20 prefectures, Fukuoka will have the highest
increase rate. In Fukuoka Prefecture, the increase in
the number of patients with heatstroke from the baseline to the near future in Case 2 was estimated to be
289.8 % (compared to 236.5 % in Case 1).
In contrast to the prefectures belonging to type 1 or
type3 (e.g., Tokyo and Fukuoka), 20 of the 46 prefectures had a lower number of patients with heatstroke
in Case 2 than in Case 1. The largest difference in the
prediction between Cases 1 and 2 was observed in
Akita Prefecture, where the increase in Case 2 was
only 174.8 % but 235.9 % in Case 1. In other words,
the risk in Case 2 is 61.1 % lower than in Case 1.
Focusing on demographic changes in Akita, the total
population will decrease by 45.2 % from the baseline
period to the near future, while the population aging
rate will increase by 31.9 %. This situation has both a
restraining effect on the number of patients with heatstroke (population decline) and an increasing effect on
the number of patients with heatstroke (aging of the
population). In the case of Akita, this restraining effect
was dominant, which may have resulted in a lower
number of patients with heatstroke in Case 2 than in
Case 1. Thus, demographic changes have the effect of
increasing or decreasing the number of patients with
heatstroke, which is an important consideration for
future projections (Table 3).
5.4 Result of near future projection with consideration
of long-term acclimatization: Case 3
Figure 8c shows the map of the near future projection for Case 3a. The figure shows that the average
total number of patients with heatstroke for all prefectures is 7.3 per summer, with a wide range from a
maximum of 14.7 per 10,000 people (Kagoshima) to a
minimum of 3.9 per 10,000 people (Tokyo) by prefec-

Vol. 100, No. 4

ture.
Figure 10c shows a map of the average increase
rate in the number of patients with heatstroke in each
prefecture in Case 3a. The average increase rate on
average nationwide is 164.5 %. This is about 60 %
smaller than Case 1, where considers only the effect of
temperature increase due to climate change. In Hokkai­
do, where the increase in the number of patients with
heatstroke from the baseline to the near future was the
highest in Case 1, the value in Case 3a was reduced by
about 100 % compared to Case 1.
Figure 8d shows the map of the near future projection for Case 3b. The figure shows that the average
total number of patients with heatstroke in the nine
prefectures is 5.3 people per summer, with a spread
from a maximum of 10.1 people/10,000 people (Yama­
nashi) to a minimum of 1.4 people/10,000 people
(Hokkaido) by prefecture. Figure 10d shows a map of
the increase rate in the number of patients with heatstroke from the baseline period to the near future for
Case 3b. The average value for the nine prefectures is
119.7 %. In four of the nine prefectures, the number
of emergency heatstroke cases decreased compared
to the current climate (Hokkaido, 66.0 %; Miyagi,
85.3 %; Yamagata, 77.0 %; and Fukushima, 92.6 %,
assuming the value of baseline to be 100 %).
5.5 Near future projections with explanatory
variables changed to daily maximum WBGT
(with population dynamics)
Figure 11 shows the map of the number of patients
with heatstroke when the same assumptions as in
Cases 1, 2, 3a, and 3b are made, and the explanatory
variable is changed to the daily maximum WBGT to
predict the number of patients with heatstroke in the
near future. Taking Case 2 (experiment considering
demographics) as an example, the total number of
patients with heatstroke is 10.4/10,000 per summer
nationwide, with a large spread from the maximum
value of 18.2/10,000 (Saga) to the minimum value of
5.1/10,000 (Hokkaido). The difference in the prediction between the model with daily maximum WBGT
and the model with daily maximum temperature is
only about 9 %. This result suggests that there is no
significant difference in the prediction results of the
two models when we focus on the number of patients
with heatstroke nationwide. However, looking at
each prefecture, there are some prefectures where the
results of near future prediction between the daily
maximum temperature model and the daily maximum
WBGT model are largely different (Tables S1a, b in
Supplement 4).

August 2022

S. NAKAMURA et al.

611

Fig. 11.  Predicted number of patients with heatstroke (per 10,000 population) under the RCP8.5 scenario of near future
climate with daily maximum WBGT as explanatory variable. (a) Prediction without population dynamics (Case 1),
(b) prediction with population dynamics (Case 2), (c) prediction using the late summer equation (Case 3a), and (d)
prediction using the climate analog (Case 3b). The areas shaded by gray color are outside of analysis target.

6. Conclusions
The main aim of this study was to estimate the
number of ambulance transport due to heatstroke under
the current and near future climates with a newly developed statistical model. The model proposed in this
study has the following three characteristics:
(1) The dependent variable (predictor) was set as the
number of heatstroke emergency patients. Directly
predicting the number of emergency patients
allows us to assess not only the risk of heatstroke
incidence among people but also the burden on the
emergency medical system.
(2) The daily maximum temperature, which is readily
available from future climate prediction datasets,
was selected as an explanatory variable.
(3) The seasonality of heatstroke risk (short-term heat
acclimatization) was considered by dividing the
summer period into three subperiods, namely, early
summer, midsummer, and late summer, with para­
meter identification appropriate for each period.
The proposed model considers not only temperature
but also three main factors, i.e., region, short-term
heat acclimatization, and age, which that are considered to affect the prediction accuracy. The results of
cross-validation showed that the prediction error was
reduced by about 22 % and 12 %, respectively, due
to considering regional characteristics and short-term

heat acclimatization. On the other hand, age did not
contribute much to the model accuracy.
In order to confirm the practicality and validity
of the proposed model, we compared its accuracy
with models in which the explanatory variables were
changed from the maximum temperature to WBGT.
The model with WBGT was the most accurate in the
majority of prefectures. However, the difference in the
prediction error between the model with temperature
and the model with WBGT was less than 10 % in the
majority of prefectures. Therefore, we conclude that
models using maximum temperatures instead of the
WBGT as the explanatory variable can be used in
practical situations by considering regional differences
and short-term heat acclimatization.
With the statistical model developed, three near
future projections of the heatstroke risk were made:
one considering only temperature increase due to
climate change (Case 1), one considering temperature
increase due to climate change and demographic
change (Case 2), and one considering temperature
increase due to climate change, demographic change,
lifestyle change, and long-term heat acclimatization
(Cases 3a, b). In Case 1, the risk of heatstroke from
the perspective of residents increases by about 2.2
times from the baseline to the near future on average
nationwide (the ensemble means of four GCMs under
the RCP8.5 scenario). The increase in risk was par-

612

Journal of the Meteorological Society of Japan

ticularly pronounced in Hokkaido, where the risk of
heatstroke increase was greater than three times. The
risk of heatstroke from the perspective of the government in Case 2 increased by a factor of 2.3 from
the baseline to the near future on average nationwide.
This result suggests that the burden of heatstroke
emergency cases on the emergency medical system in
the near future cannot be ignored. The heatstroke risk
in the near future in Case 2 is greater than that in Case
1 on average nationwide. However, in some prefectures, such as Akita, the effect of population decline
on risk reduction is more dominant than the climate
change on risk increase. Whether demographic change
increases or decreases risk is not uniquely determined.
From the prediction of Case 3a, it is found that the
risk of emergency heatstroke can be reduced by about
30 % on average nationwide by acquiring heat tolerance and changing lifestyles.
Lifestyle changes mean various changes for the
adaptation to the worse thermal environment, as represented by the widespread use of air conditioners (see
Section 3.8 for details). Case 3b shows that the risk of
emergency heatstroke in the near future is lower than
that in the baseline in some regions, such as Hokkaido.
In other words, the results suggest that there is much
room for risk control in cold regions by promoting the
acquisition of heat tolerance and lifestyle changes.
Finally, in order to confirm the uncertainty of the
explanatory variables, a comparison experiment was
conducted using the daily maximum WBGT as an explanatory variable. As a result, the difference between
the prediction result of the number of patients with
heatstroke by the daily maximum temperature model
and that by the daily maximum temperature WBGT
model was about 9 % on average nationwide.
Data Availability Statement
• The number of ambulance transport datasets analyzed in this study are available at [https://www.
fdma.go.jp/disaster/heatstroke/post3.html].
• The current climate data (AMeDAS) analyzed in
this study are available at [https://www.data.jma.go.
jp/gmd/risk/obsdl].
• The statistical downscaling datasets (Nishimori
et al. 2019) analyzed in this study are available at
[doi:10.20783/DIAS.568].
• The population datasets analyzed in this study are
available at [Baseline (1990); https://www.e-stat.
go.jp/dbview?sid=0000031399] and [Near future
(2040); https://www.ipss.go.jp/pp-shicyoson/j/shi
cyoson18/t-page.asp].

Vol. 100, No. 4

Supplements
Supplement 1: How to calculate the maximum daily
WBGT
In this study, the following equation was used to
calculate WBGT (Yaglou and Minard 1957). Day and
night were discriminated based on the value of horizontal-plane insolation; a positive horizontal-plane
insolation value was judged to be daytime and zero
was judged to be nighttime.
WBGT = 0.7Tw + 0.2Tg + 0.1Td (daytime),
WBGT = 0.7Tw + 0.3Td (nighttime).
The dry-bulb and wet-bulb temperatures were
based on the aforementioned values. The black-bulb
temperature (Tg) was estimated using the equation by
Okada and Kusaka (2013). When using this equation,
the values of wind speed and solar radiation are also
required. The wind speed was the spatial average of
AMeDAS observations, as well as the temperature.
Solar radiation was measured by the meteorological
observatory. However, some meteorological observatories do not observe insolation. In such cases, the
values were estimated from the time series of sunshine
duration using the equation by Kondo (1994) and
Kondo and Xu (1997). The daily maximum WBGT
was obtained from the hourly values of WBGT obtained using this method.
Supplement 2: Regional Classification of Japan
Figure S1: Regional classifications and names of
major prefectures in Japan. Based on the forecast categories used in the JMA’s regional seasonal forecasts.
Note that this classification is slightly different from
the standard classification by the government.
Supplement 3: Increase in daily maximum temperature
and daily maximum WBGT from the baseline period
to the near future period
Figure S2: Increase in (a) daily minimum temperature and (b) daily maximum WBGT (°C) from the
baseline period to the near future period for each
prefecture. Daily maximum temperature and daily
maximum WBGT were ensemble averages from four
GCMs, GFDL-CM3, HadGEM2-ES, MIROC5, and
MRI-CGCM3 (RCP8.5).
Supplement 4: The number of people transported to
emergency rooms for heat stroke in each experiment
(Beseline, Cases1, 2, 3a, 3b) and the days with high
risk of heat stroke
Table S1: The number of heatstroke emergency pa-

August 2022

S. NAKAMURA et al.

tients in summer and days with high risk of heat stroke
in each prefecture predicted in this study. (a) explanatory variable is daily maximum temperature, (b) daily
maximum temperature is daily maximum WBGT. The
days with high risk of heat stroke are (a) extremely
hot days (daily maximum temperature ³ 35℃) and (b)
dangerous days (daily maximum WBGT ³ 31℃).
Acknowledgments
This work was supported by the Social Implementation Program on Climate Change Adaptation Technology (SI-CAT) Grant Number JPMXD0715667165
from the Ministry of Education, Culture, Sports, Science and Technology (MEXT), Japan. This research
was performed by the Environment Research and
Technology Development Fund JPMEERF20192005
of the Environmental Restoration and Conservation
Agency of Japan.
References
Adachi, S. A., F. Kimura, H. Kusaka, T. Inoue, and H. Ueda,
2012: Comparison of the impact of global climate
changes and urbanization on summertime future climate in the Tokyo metropolitan area. J. Appl. Meteor.
Climatol., 51, 1441–1454.
Anderson, B. G., and M. L. Bell, 2009: Weather-related
mortality: How heat, cold, and heat waves affect mortality in the United States. Epidemiology, 20, 205–
213.
Ando, M., S. Yamamoto, and S. Asanuma, 2004: Global
warming and heatstroke. Japanese J. Biometeorol.,
41, 45–49 (in Japanese with English abstract).
Basu, R., and J. M. Samet, 2002: An exposure assessment
study of ambient heat exposure in an elderly population in Baltimore, Maryland. Environ. Health Perspect.,
110, 1219–1224.
Byrne, M. P., and P. A. O’Gorman, 2016: Understanding
decreases in land relative humidity with global
warming: Conceptual model and GCM simulations. J.
Climate, 29, 9045–9061.
Chen, K., R. M. Horton, D. A. Bader, C. Lesk, L. Jiang,
B. Jones, L. Zhou, X. Chen, J. Bi, and P. L. Kinney,
2017: Impact of climate change on heat-related mortality in Jiangsu Province, China. Environ. Pollut.,
224, 317–325.
Curriero, F. C., K. S. Heiner, J. M. Samet, S. L. Zeger, L.
Strug, and J. A. Patz, 2002: Temperature and mortality in 11 cities of the eastern United States. Amer. J.
Epidemiology, 155, 80–87.
Doyon, B., D. Bélanger, and P. Gosselin, 2008: The potential impact of climate change on annual and seasonal
mortality for three cities in Québec, Canada. Int. J.
Health Geogr., 7, 23, doi:10.1186/1476-072X-7-23.
Fiala, D., G. Havenith, P. Bröde, B. Kampmann, and G.

613

Jendritzky, 2012: UTCI-Fiala multi-node model of
human heat transfer and temperature regulation. Int. J.
Biometeorol., 56, 429–441.
Fire and Disaster Management Agency, 2019: The emergency conveyance situation due to the heat stroke from
May to September, 2018. information materials. 17 pp
(in Japanese). [Available at https://www.fdma.go.jp/
disaster/heatstroke/item/heatstroke003_houdou01.pdf.]
Fire and Disaster Management Agency, 2021: Commencement of “Survey on the number of people transported
to emergency rooms due to heatstroke during the
summer”. 2 pp (in Japanese). [Available at https://
www.fdma.go.jp/disaster/heatstroke/items/heatstroke_
chousa_kyu124.pdf.]
Flynn, A., C. McGreevy, and E. C. Mulkerrin, 2005: Why do
older patients die in a heatwave? QJM: Int. J. Med.,
98, 227–229.
Fujibe, F., 2013: Long-term variations in heat mortality and
summer temperature in Japan. Tenki, 60, 371–381 (in
Japanese with English abstract).
Fujibe, F., J. Matsumoto, and H. Suzuki, 2018a: Spatial and
temporal features of heat stroke mortality in Japan and
their relation to temperature variations, 1999–2014.
Geogr. Rev. Japan Ser. B, 91, 17–27.
Fujibe, F., J. Matsumoto, and H. Suzuki, 2018b: Regional
features of the relationship between daily heat-stroke
mortality and temperature in different climate zones
in Japan. SOLA, 14, 144–147.
Fujibe, F., J. Matsumoto, and H. Suzuki, 2020: Spatial variability of municipality-wise heat and cold mortality
in Japan with respect to temperature and economic
states. Geogr. Rev. Japan Ser. B, 92, 72–83.
Fuse, A., S. Saka, R. Fuse, T. Araki, S. Kin, M. Miyauchi,
and H. Yokota, 2014: Weather data can predict the
number of heat stroke patient. J. Japanese Assoc.
Acute Med., 25, 757−765 (in Japanese with English
abstract).
Gasparrini, A., Y. Guo, F. Sera, A. M. Vicedo-Cabrera, V.
Huber, S. Tong, M. de Sousa Zanotti Stagliorio Coelho,
P. H. N. Saldiva, E. Lavigne, P. M. Correa, N. V.
Ortega, H. Kan, S. Osorio, J. Kyselý, A. Urban, J. J. K.
Jaakkola, N. R. I. Ryti, M. Pascal, P. G. Goodman, A.
Zeka, P. Michelozzi, M. Scortichini, M. Hashizume,
Y. Honda, M. Hurtado-Diaz, J. C. Cruz, X. Seposo, H.
Kim, A. Tobias, C. Iñiguez, B. Forsberg, D. O. Åström,
M. S. Ragettli, Y. L. Guo, C.-F. Wu, A. Zanobetti,
J. Schwartz, M. L. Bell, T. N. Dang, D. D. Van, C.
Heaviside, S. Vardoulakis, S. Hajat, A. Haines, and B.
Armstrong, 2017: Projections of temperature-related
excess mortality under climate change scenarios.
Lancet Planet. Health, 1, e360–e367.
Gosling, S. N., G. R. McGregor, and A. Páldy, 2007: Climate change and heat-related mortality in six cities.
Part 1: Model construction and validation. Int. J.
Biometeorol., 51, 525–540.
Gosling, S. N., G. R. McGregor, and J. A. Lowe, 2009: Cli-

614

Journal of the Meteorological Society of Japan

mate change and heat-related mortality in six cities.
Part 2: Climate model evaluation and projected
impacts from changes in the mean and variability of
temperature with climate change. Int. J. Biometeorol.,
53, 31–51.
Guo, Y., A. Gasparrini, S. Li, F. Sera, A. M. Vicedo-Cabrera,
M. de Sousa Zanotti Stagliorio Coelho, P. H. N.
Saldiva, E. Lavigne, B. Tawatsupa, K. Punnasiri,
A. Overcenco, P. M. Correa, N. V. Ortega, H. Kan,
S. Osorio, J. J. K. Jaakkola, N. R. I. Ryti, P. G.
Goodman, A. Zeka, P. Michelozzi, M. Scortichini, M.
Hashizume, Y. Honda, X. Seposo, H. Kim, A. Tobias,
C. Íñiguez, B. Forsberg, D. O. Åström, Y . L. Guo,
B.-Y. Chen, A. Zanobetti, J. Schwartz, T. N. Dang,
D. D. Van, M. L. Bell, B. Armstrong, K. L. Ebi, and
S. Tong, 2018: Quantifying excess deaths related to
heatwaves under climate change scenarios: A multicountry time series modelling study. PLoS Med., 15,
e1002629, doi:10.1371/journal.pmed.1002629.
Hajat, S., R. S. Kovats, and K. Lachowycz, 2007: Heatrelated and cold-related deaths in England and Wales:
Who is at risk? Occupational Environ. Med., 64, 93–
100.
Havenith, G., 2005: Temperature regulation, heat balance
and climatic stress. Extreme Weather Events and
Public Health Responses. Kirch, W., R. Bertollini, and
B. Menne (eds.), Springer, 69–80.
Hayhoe, K., D. Cayan, C. B. Field, P. C. Frumhoff, E. P.
Maurer, N. L. Miller, S. C. Moser, S. H. Schneider,
K. N. Cahill, E. E. Cleland, L. Dale, R. Drapek, R. M.
Hanemann, L. S. Kalkstein, J. Lenihan, C. K. Lunch,
R. P. Neilson, S. C. Sheridan, and J. H. Verville, 2004:
Emissions pathways, climate change, and impacts on
California. Proc. Natl. Acad. Sci. U.S.A. (PNAS), 101,
12422–12427.
Honda, Y., M. Kondo, G. McGregor, H. Kim, Y.-L. Guo, Y.
Hijioka, M. Yoshikawa, K. Oka, S. Takano, S. Hales,
and R. S. Kovats, 2014: Heat-related mortality risk
model for climate change impact projection. Environ.
Health Prev. Med., 19, 56–63.
Hondula, D. M., R. E. Davis, M. V. Saha, C. R. Wegner,
and L. M. Veazey, 2015: Geographic dimensions of
heat-related mortality in seven U.S. cities. Environ.
Res., 138, 439–452.
Huber, V., L. Krummenauer, C. Peña-Ortiz, S. Lange, A.
Gasparrini, A. M. Vicedo-Cabrera, R. Garcia-Herrera,
and K. Frieler, 2020: Temperature-related excess
mortality in German cities at 2℃ and higher degrees
of global warming. Environ. Res., 186, 109447, doi:
10.1016/j.envres.2020.109447.
Ikeda, T., and H. Kusaka, 2021: Development of models
for predicting the number of patients with heatstroke
on the next day considering heat acclimatization. J.
Meteor. Soc. Japan, 99, 1395–1412.
Ishizaki, N., H. Shiogama, K. Takahashi, S. Emori, K.
Dairaku, H. Kusaka, T. Nakaegawa, and I. Takayabu,

Vol. 100, No. 4

2012: An attempt to estimate of probabilistic regional
climate analogue in a warmer Japan. J. Meteor. Soc.
Japan, 90B, 65–74.
Jackson, J. E., M. G. Yost, C. Karr, C. Fitzpatrick, B. K.
Lamb, S. H. Chung, J. Chen, J. Avise, R. A. Rosenblatt, and R. A. Fenske, 2010: Public health impacts
of climate change in Washington State: Projected
mortality risks due to heat events and air pollution.
Climatic Change, 102, 159–186.
Keatinge, W. R., G. C. Donaldson, E. Cordioli, M. Martinelli, A. E. Kunst, J. P. Mackenbach, S. Nayha, and I.
Vuori, 2000: Heat related mortality in warm and cold
regions of Europe: Observational study. BMJ, 321,
670–673.
Kimura, F., and A. Kitoh, 2007: Downscaling by pseudo
global warming method. The final report of the
ICCAP. The research project on the impact of climate
changes on agricultural production system in arid
areas. ICCAP Publication, 10, 43–46. [Available at
https://www.chikyu.ac.jp/iccap/ICCAP_Final_Report/
2/4-climate_kimura.pdf.]
Knowlton, K., B. Lynn, R. A. Goldberg, C. Rosenzweig,
C. Hogrefe, J. K. Rosenthal, and P. L. Kinney, 2007:
Projecting heat-related mortality impacts under a
changing climate in the New York City region. Amer. J.
Public Health, 97, 2028–2034.
Kovats, R. S., and S. Hajat, 2008: Heat stress and public
health: A critical review. Annu. Rev. Public Health,
29, 41–55.
Kusaka, H., A. Suzuki-Parker, T. Aoyagi, S. A. Adachi, and
Y. Yamagata, 2016: Assessment of RCM and urban
scenarios uncertainties in the climate projections for
August in the 2050s in Tokyo. Climatic Change, 137,
427–438.
Li, T., R. M. Horton, and P. L. Kinney, 2013: Projections
of seasonal patterns in temperature-related deaths for
Manhattan, New York. Nat. Climate Change, 3, 717–
721.
McGeehin, M. A., and M. Mirabelli, 2001: The potential
impacts of climate variability and change on temperature-related morbidity and mortality in the United
States. Environ. Health Perspect., 109, 185–189.
Morikawa, T., H. Yamada, K. Tanaka, S. Okayama, Y.
Shibata, Y. Nakata, H. Watanabe, and T. Kidokoro,
2021: Air quality estimation in 2050 – JSAE 2050
challenge and air quality estimation in 2050 –. Trans.
Soc. Automotive Engineers Japan, 52, 1261–1266 (in
Japanese).
Nakai, S., T. Itoh, and T. Morimoto, 1999: Deaths from heatstroke in Japan: 1968–1994. Int. J. Biometeorol., 43,
124–127.
National Institute of Population and Social Security Research,
2018: Regional population projections for Japan:
2015–2045. 256 pp (in Japanese). [Available at https://
www.ipss.go.jp/pp-shicyoson/j/shicyoson18/6houkoku/houkoku.asp.]

August 2022

S. NAKAMURA et al.

Nelder, J. A., and R. W. M. Wedderburn, 1972: Generalized
linear models. J. Roy. Stat. Soc.: Ser. A (General), 135,
370–384.
Nishimori, M., Y. Ishigooka, T. Kuwagata, T. Takimoto,
and N. Endo, 2019: SI-CAT 1km-grid square regional
climate projection scenario dataset for agricultural use
(NARO2017). J. Japan Soc. Simulation Technol., 38,
150–154 (in Japanese).
Okada, M., and H. Kusaka, 2013: Proposal of a new equation to estimate globe temperature in an urban park
environment. J. Agric. Meteor., 69, 23–32.
Ono, M., 2013: Heat stroke and the thermal environment.
JMAJ, 56, 199–205.
Piver, W. T., M. Ando, F. Ye, and C. J. Portier, 1999:
Temperature and air pollution as risk factors for heat
stroke in Tokyo, July and August 1980–1995. Environ. Health Perspect., 107, 911–916.
Petkova, E. P., A. Gasparrini, and P. L. Kinney, 2014: Heat
and mortality in New York City since the beginning of
the 20th century. Epidemiology, 25, 554–560.
Rainham, D. G. C., and K. E. Smoyer-Tomic, 2003: The
role of air pollution in the relationship between a heat
stress index and human mortality in Toronto. Environ.
Res., 93, 9–19.
Sato, T., F. Kimura, and A. Kitoh, 2007: Projection of global
warming onto regional precipitation over Mongolia
using a regional climate model. J. Hydrol., 333, 144‒
154.
Sato, T., H. Kusaka, and H. Hino, 2020: Quantitative assessment of the contribution of meteorological variables
to the prediction of the number of patients with heatstroke for Tokyo. SOLA, 16, 104−108.

615

Semenza, J. C., C. H. Rubin, K. H. Falter, J. D. Selanikio,
W. D. Flanders, H. L. Howe, and J. L. Wilhelm, 1996:
Heat-related deaths during the July 1995 heat wave in
Chicago. New England J. Med., 335, 84−90.
Sherwood, S. C., 2018: How important is humidity in heat
stress? J. Geophys. Res.: Atmos., 123, 11808–11810.
Shumway, R. H., A. S. Azari, and Y. Pawitan, 1988: Modeling mortality fluctuations in Los Angeles as functions
of pollution and weather effects. Environ. Res., 45,
224–241.
Smoyer, K. E., D. G. C. Rainham, and J. N. Hewko, 2000a:
Heat-stress-related mortality in five cities in Southern
Ontario: 1980–1996. Int. J. Biometeorol., 44, 190–197.
Smoyer, K. E., L. S. Kalkstein, J. S. Greene, and H. Ye,
2000b: The impacts of weather and pollution on
human mortality in Birmingham, Alabama and Philadelphia, Pennsylvania. Int. J. Climatol., 20, 881–897.
Whitman, S., G. Good, E. R. Donoghue, N. Benbow, W.
Shou, and S. Mou, 1997: Mortality in Chicago attributed to the July 1995 heat wave. Amer. J. Public
Health, 87, 1515–1518.
Vaidyanathan, A., J. Malilay, P. Schramm, and S. Saha,
2020: Heat-related deaths—United States, 2004–2018.
Morbidity and mortality weekly report, 69, 729–734.
Yaglou, C. P., and D. Minaed, 1957: Control of heat casualties at military training centers. Amer. Med. Assoc.
Arch. Ind. Health, 16, 302–316.
Zhang, K., Y. Li, J. D. Schwartz, and M. S. O’Neill, 2014:
What weather variables are important in predicting
heat-related mortality? A new application of statistical
learning methods. Environ. Res., 132, 350−359.

この論文で使われている画像