Feasibility of the Olympic marathon under climatic and socioeconomic change

There are concerns about the impact of climate change on Olympic Games, especially endurance events, such as marathons. In recent competitions, many marathon runners dropped out of their races due to extreme heat, and it is expected that more areas will be unable to host the Games due to climate change. Here, we show the feasibility of the Olympic marathon considering the variations in climate factors, socioeconomic conditions, and adaptation measures. The number of current possible host cities will decline by up to 27% worldwide by the late twenty-first century. Dozens of emerging cities, especially in Asia, will not be capable of hosting the marathon under the highest emission scenario. Moving the marathon from August to October and holding the Games in multiple cities in the country are effective measures, and they should be considered if we are to maintain the regional diversity of the Games.

www.nature.com/scientificreports/ marathon runners including elite athletes 14,15 , and since the WBGT standards suggested by the International Institute for Running Medicine (IIRM) 16 and American College of Sports Medicine (ACSM) 17 are referred to in the management of various marathons, we adopt the WBGT as an indicator of the feasibility of hosting an Olympic marathon. We utilize the seven GCMs included in the S14 retrospective meteorological forcing dataset (S14FD) 18 , which was developed based on the Coupled Model Intercomparison Project Phase 5 (CMIP5) 19 . We reproduce hourly urban climatic characteristics by applying a bias correction using meteorological observation data based on the method of Takakura et al. 20 . We consider future socioeconomic conditions using the parameters of population 21,22 , the gross domestic product (GDP) 23 , and its growth rate in the 2030-2050 period and the 2070-2090 period, both are 10 years before the evaluation periods, based on the five Shared Socioeconomic Pathways (SSPs) 24 . We quantify the effects of AMs that have been adopted or considered in recent Games, such as holding the marathon late at night or early in the morning. The main text presents the results for the late twenty-first century (2080-2099), while the results for the mid-twenty-first century (2040-2059) are presented in the Supplementary Information (SI).

Results
Climate change impacts under current socioeconomic conditions. First, to clarify the impact of climate change alone, we evaluated the number of cities that could host the Olympic marathon in August during the late twenty-first century (2080-2099) under the socioeconomic conditions in 2020, changing only the climatic conditions according to the four RCPs. A total of 70 cities in 25 countries were selected based on (1) socioeconomic conditions as of 2010 (an urban population of 2.5 million or more, a national GDP of 300 billion dollars or more (purchasing power parity (PPP), Int'l $ 2005) as of 2010 and a GDP growth rate above 0% as of 2010-2015); (2) an elevation of less than 1,600 m 11 ; and (3) the availability of meteorological data for WBGT correction using the method of Takakura et al. 20 . See the Methods section for the rationale behind the selection criteria above. As evaluation criteria for cities, the WBGT levels were set as follows based on the four alert levels (low, moderate, high, and extreme) in the IIRM Medical Care manual 16 .
WBGT level 1 (good): There is a greater than 90% probability that the WBGT will fall below 18°C for at least three consecutive hours between 7:00 and 21:00 in August. WBGT level 1 corresponds to a low alert level and "good conditions" in the IIRM's manual. WBGT level 2 (caution): Level 1 does not apply, and there is a greater than 90% probability that the WBGT will fall below 22°C for at least 3 consecutive hours between 7:00 and 21:00 in August. WBGT level 2 corresponds to a moderate alert level and "less than ideal conditions" in the IIRM's manual. WBGT level 3 (warning): Levels 1 and 2 do not apply, and there is a greater than 90% probability that the WBGT will fall below 28°C for at least 3 consecutive hours between 7:00 and 21:00 in August. WBGT level 3 corresponds to a high alert level and "potentially dangerous conditions" in the IIRM's manual. WBGT level 4 (cancel): Levels 1, 2, and 3 do not apply. WBGT level 4 corresponds to an extreme alert level and "event cancelled / extreme and dangerous conditions" in the IIRM's manual.
We determine that cities with WBGT levels of 1 to 3 can host the Olympic marathon, while those with a WBGT level of 4 cannot. It has recently been found that the performance of elite marathon runners improves as the dry bulb temperature rises 25 , which is closely related to the WBGT. However, there would have been few cases where the WBGT exceeded 28 °C in the Berlin Marathon races covered in the study even momentarily, based on the dry bulb temperature range in the races. Therefore, the use of WBGT Level 4 (28 °C) as a threshold for the impossibility of holding a marathon is consistent with the results of the aforementioned research. The 90% criterion is set in reference to previous studies on the feasibility of hosting the summer and winter Olympics 11,26,27 , and the duration (3 h) and timing (between 7:00 and 21:00 in August) is set based on the general competition time of the Olympic marathon, where all Olympic marathons since the 1980 Moscow Olympics were held, except for the men's and women's marathons at the 1988 Seoul and 2000 Sydney Olympics, the women's marathon at the 1996 Atlanta Olympics, and the women's marathon at the 2020 Tokyo Olympics, which was moved up by an hour the day before to start at 6:00 [28][29][30][31][32][33] .
As a result, globally, the number of cities that can host the Olympic marathon (WBGT levels 1-3) significantly decreases as the amount of greenhouse gas emissions increases toward the late twenty-first century (2080-2099) (  Table 2. Regionally, there are three patterns in the decrease in the number of cities that can host the Olympic marathon (Fig. 2). The number of cities by country is shown in Supplementary Fig. 3    Pattern 2: North America, Latin America and the Caribbean, and Africa. The number of cities that can host the event (WBGT levels 1-3) only slightly decreases even under the highest emission scenario, while the number of cities with a relatively low heat risk (WBGT levels 1-2) will decrease due to climate change. This finding may be because many cities are in the high latitudes of the Northern Hemisphere (North America) or the Southern Hemisphere (Latin America and the Caribbean, and Africa) and are thus less likely to be hot and humid in August than Pattern 1 (Asia). Climate change impacts under future socioeconomic conditions. Next, we estimate the number of cities that can host the Olympic marathon in the late twenty-first century (2080-2099), considering future socioeconomic scenarios (SSPs 1-5) (Fig. 3). Here, all combinations of the four RCPs and five SSPs are shown to identify the wide range of impacts due to climate change and socioeconomic conditions, but it should be noted that the combinations of RCP2.6 and SSP3 and RCP8.5 and SSPs 1-4 are infeasible 24  We selected target cities using the same criteria as in the evaluation of 70 cities for current socioeconomic conditions. The numbers of selected cities for the five SSPs are 94 for SSP1 (sustainability scenario), 120 for SSP2 (middle of the road scenario), 79 for SSP3 (regional rivalry scenario), 90 for SSP4 (inequality scenario), and 126 for SSP5 (fossil-fueled development scenario).
Globally, as in the results under current socioeconomic conditions, the number of cities that can host the Olympic marathon tends to decrease as the amount of greenhouse gas emissions increases, especially in Asia. The number of cities under each emission scenario is the highest for SSP5 and the lowest for SSP3. Under all SSPs, the number of possible host cities increases from the current socioeconomic conditions (2020). For example, under RCP2.6, the number of cities ranges from 72.7 (SSP3) to 122 (SSP5), while under RCP8.5, the number of cities ranges from 57.6 (SSP3) to 99.9 (SSP5).
For all SSPs, the number of possible cities under RCP8.5 is approximately 20% lower than that under RCP2.6, and the impact of the emission scenarios is apparent. Dozens of emerging cities, mainly in Asia, that could host www.nature.com/scientificreports/   Fig. 19) and for all 165 cities in the mid-and late-twenty-first centuries ( Supplementary Figs. 20-31).

Discussion
The main contribution of this study is that it more comprehensively explains the feasibility of hosting the Olympic marathon in the future than previous studies 7, 11 , considering the variations in climate change projections, the hourly climatic characteristics of cities, future socioeconomic conditions, and the effect of adaptation measures. As a result, the number of the 70 cities that can host the Olympic marathon under the socioeconomic conditions in 2020 decrease by approximately 27% globally under RCP8.5 in the late twenty-first century, with a particularly large decrease seen in Asia. When compared with the results of Smith et al. 11 , who assessed the situation as of 2085 under RCP8.5, the decrease in the number of potential host cities is more moderate in this study.
Considering that the WBGT varies significantly within a day (see Supplementary Fig. 32 for an example), the Adaptation measure 4 (AM4: All) Implementing AMs 1 to 3 simultaneously. This measure has not been implemented or considered for the Olympic Games to the best of our knowledge www.nature.com/scientificreports/ fact that the evaluation is based on hourly WBGT rather than daily average WBGT, as in Smith et al. 11 , may have contributed to the difference in results. The use of seven GCMs in this study is also likely to contribute to the difference in results from Smith et al. 11 , who used two GCMs, since the error bars in the figures show significant variations among GCMs. Furthermore, we quantified the impacts of socioeconomic conditions (SSPs 1-5) and the effectiveness of adaptation measures in addition to the impacts of climate change.
The number of countries participating in the Games has increased from 14 in the first modern Games in Athens in 1896 37 to 206 in Rio de Janeiro in 2016 38 , and since the Munich Olympics in 1972, each of the Games has been held on a different continent from the previous one, which suggests that regional diversity is becoming increasingly important in the Games. If the Games are to be held in various regions of the world in the future, adaptation measures including the four presented in this paper, are worth considering, in addition to mitigating the impacts of climate change by reducing greenhouse gas emissions. In particular, these adaptation measures will be necessary for many cities to be able to host marathons with relatively low heat risk (WBGT levels 1-2).
Hosting the Olympics in October (AM2) was found to be an effective adaptation measure, but it requires negotiation with broadcasters in North America (US and Canada), which pays the majority of the International Olympic Committee's broadcasting rights fees 39 . In North America, July and August are off-season for sports, which may be the reason why most of the recent Games have been held in those months 40 . The share of broadcasting rights fees from Asia, Europe, Central America, South America, and the Caribbean is increasing 39 , but it is not clear whether they will be sufficient to offset the possible decline in broadcasting rights fees for North America. In addition, another effective measure for holding the Games in multiple cities in a country (AM3) may increase the cost related to infrastructure, security, and logistics; therefore, careful consideration of costeffectiveness is necessary.
For example, holding the Games in a city with existing infrastructure and experience in hosting large-scale sporting events and/or limiting the number of spectators and people involved in the Games would reduce the costs associated with AM2 and AM3 41,42 . The above proposal is consistent with the proposal by Muller et al. 10 from the viewpoint of the sustainability of the Games (e.g., drastic reduction in the scale of the Games or rotating the Games among the existing host cities).
Some issues should be considered in the future. (1) The Games include many outdoor events other than the marathon. It is desirable to evaluate the thermal conditions for such events and to consider possible competition programs within 16 days, which is the basic duration of the Games 43 . (2) Although the focus of this study is on athletes, the heat risk to spectators, who make up the majority of those involved, should be considered as well. (3) In this study, the evaluation was based on the WBGT of representative points in each city, and therefore, microclimatic conditions in the city and adaptation measures, such as the construction of heat-shielding pavements, planting of trees, and misting, were not considered. Such evaluations require a large computational resource for modeling and/or detailed measurements in the field; therefore, it is realistic to use these methods for a smaller number of cities. (4) The cities selected in this study may differ from those that can host the Olympic Games in the future due to the lack of relevant data. Factors such as political rights, experience in hosting world championships, the population support for the Games, dispute with the IOC, and existing stadium infrastructure are also important for being selected as a host city 44 , but it is difficult to predict the situation of such factors in the mid-and late twenty-first centuries. Therefore, we did not include them in the city selection. Additionally, many cities particularly in Africa and Latin America are excluded due to the lack of sufficient meteorological observation data to reproduce the climatic characteristics of cities, even though they have sufficient socioeconomic conditions to be selected. Moreover, from the perspective of sustainability, Olympic Agenda 2020 45 , presented by the IOC as a reform plan for the Olympic Movement, encourages the use of simple temporary facilities rather than new venues, and it allows the games to be held in locations other than the host city (e.g., where facilities already exist). Therefore, it is possible that smaller cities than those targeted in this study will be allowed to host the Olympics in the future. Based on the above, it is necessary to review the conditions for selecting cities in the future, taking into account the improvement in related data and the trend of sustainability improvement in the Olympics. (5) The WBGT, which was used as the index in this study, has several limitations, including the tendency to underestimate heat in environments where sweat evaporation is limited (high humidity and low wind speed), the inability to take into account conditions such as exercise intensity and clothing, and the lack of evidence on the relationship with health risks in sports 46,47 . In the future, it is necessary to develop more robust methods for assessing health risks in marathon running and sports in general.

Selection of cities for evaluation.
Referring to the papers that studied the selection of host cities for the Games 44,48 and the conditions of the historical host cities, we selected 165 cities in 66 countries based on the following four perspectives: (1) the urban population, (2) national GDP and its growth rates, (3) elevation, and (4) the availability of meteorological observation data to reproduce the climatic characteristics of the city. Regarding socioeconomic conditions, (1) and (2), since the selection of host cities generally starts approximately 10 years before the Games 48 , this study use data from 2010 (for 2020), the 2030-2050 period (for 2040-2059), and the 2070-2090 period (for 2080-2099), 10 years before the evaluation period. The specific conditions for the four perspectives are described as follows.
Condition 1: The city has a population of at least 2.5 million. A study of bids for the 1992-2020 Summer Olympic Games pointed out that a city population of 2.5 million or more is an important factor in being selected as a host city 44 . In reality, competitions are rarely all held in a single city, and in most cases, competitions are also held in neighboring cities. Therefore, in this study, the selection is made based on the population of urban agglomerations ("the population contained within the contours of a contiguous territory inhabited at urban density levels without regard for administrative boundaries" 21 ). We use data from 2010 (for 2020), the 2030-2050 www.nature.com/scientificreports/ period (for 2040-2059), and the 2070-2090 period (for 2080-2099). The data for each period is acquired or calculated using the following methods. Condition 2: The national GDP is no less than $300 billion (PPP, Int'l $ 2005) and the GDP growth rate is above 0%. Olympic-related expenditures are often made on a national basis and on a city basis. After the 1992 Summer Olympic Games in Barcelona, no host city had a national GDP of less than $300 billion 10 years before the Games 49 . Therefore, we set the threshold at $300 billion. Regarding GDP, its medium-term growth rate is important for being selected as a host city 44 , and after the 1992 Summer Olympics in Barcelona, there was no host city with an average national GDP growth rate of less than 0% from 15 to 10 years before the Games. For these reasons, we set a GDP growth rate of above 0% as a condition. We use data from 2010 (GDP for 2020), the 2010-2015 period (GDP growth rate for 2020), the 2030-2050 period (GDP and its growth rate for 2040-2059), and the 2070-2090 period (GDP and its growth rate for 2080-2099) 22 . For each SSP, the national GDPs in 2010, the average GDP from 2030 to 2050, and the average GDP from 2070 to 2090 are used 22 .
Condition 3: The elevation of the city is less than 1,600 m 11 . Given the negative effects of a high elevation on long-distance running, including marathons that became apparent at the 1968 Mexico City Olympics 9 , cities with elevations higher than 1,600 m should not be viable for hosting marathon competitions.
Condition 4: There is a meteorological station within 20 km of the representative point of the city, with at least 2,500 data points collected from May to October (summer season in the Northern Hemisphere) over multiple years at intervals of 3 h or less. This condition was established to adequately reproduce the site-specific WBGT, referring to the SI of Takakura et al. 20 . We used terrestrial weather station data 51 and solar radiation data 52 corresponding to the weather stations.
Calculating site-specific hourly resolution WBGT. To consider holding a marathon in a particular city for a few hours of the day, it is necessary to calculate the site-specific hourly resolution WBGT. The environmental conditions during Olympic competitions vary even from start to finish during the same race 9 . In many cases, the output of GCMs does not provide diurnal variations, and the spatial resolution is coarse (e.g., horizontal resolution > 50 km), so it is not suitable for this study. Dynamical downscaling of GCM output using regional climate models (RCMs) is a common solution for spatiotemporally detailed calculations, but it is computationally expensive and not suitable for extensive global studies, such as this one. Therefore, we adopt the method of Takakura et al. 20 , which can provide site-specific hourly resolution WBGT prediction with only relatively simple calculations using GCM outputs, and we can achieve the objective of this study with a reasonable amount of computation.
In Takakura et al. 20 , bias correction of WBGT calculated from each grid of GCMs was performed using the observation records of WBGT in six cities in Japan. In this study, however, we use the hourly resolution meteorological data of each city to calculate the equivalent WBGT data using the method of Liljegren et al. 53 since most of the cities in the world do not have published WBGT observation records. Additionally, we used natural spline interpolation to create 24-h data for cities where data existed at 3-h intervals or less and excluded other cities since many cities do not have a complete set of 24-h observation data. For more explanations of the calculation, see Takakura et al. 20 and Supplementary Fig. 33.
Statistical analysis. The feasibility of hosting the Olympic marathon in each city is assessed by comparing the 90th percentile of the WBGT between 7:00 and 21:00 in the RCP/target period (e.g. RCP4.5/2080-2099), averaged over seven GCMs, with the conditions of four WBGT levels defined in "Climate change impacts under current socioeconomic conditions". To mitigate the effects of interannual variability of the climate, the period covered is 20 years, and this setting is commonly used in IPCC reports 1 . The 90th percentile is used to identify the 90% probability of the definition of WBGT levels. In Figs. 1, 2, 3 and 4, the maximum and minimum 90th percentiles of seven GCMs are shown as error bars to represent the variability among climate models.
Setting of adaptation measures. We evaluate the effect of four AMs that have been adopted or considered in recent Games (except AM4) and can be reproduced with the spatiotemporal resolution of this study. However, it has not been confirmed whether these measures were considered heat countermeasures when they were adopted or considered in the past. See Table 1 for details.

Data availability
The datasets generated and/or analyzed during the current study are available from the corresponding author upon reasonable request. www.nature.com/scientificreports/