Increasing heat risk in China’s urban agglomerations

A heat danger day is defined as an extreme when the heat stress index (a combined temperature and humidity measure) exceeding 41 °C, warranting public heat alerts. This study assesses future heat risk (i.e. heat danger days times the population at risk) based on the latest Coupled Model Intercomparison Project phase 6 projections. In recent decades (1995–2014) China’s urban agglomerations (Beijing-Tianjin-Hebei, Yangtze River Delta, Middle Yangtze River, Chongqing-Chengdu, and Pearl River Delta (PRD)) experienced no more than three heat danger days per year, but this number is projected to increase to 3–13 days during the population explosion period (2041–2060) under the high-emission shared socioeconomic pathways (SSP3-7.0 and SSP5-8.5). This increase will result in approximately 260 million people in these agglomerations facing more than three heat danger days annually, accounting for 19% of the total population of China, and will double the current level of overall heat risk. During the period 2081–2100, there will be 8–67 heat danger days per year, 60%–90% of the urban agglomerations will exceed the current baseline number, and nearly 310 million people (39% of the total China population) will be exposed to the danger, with the overall heat risk exceeding 18 times the present level. The greatest risk is projected in the PRD region with 67 heat danger days to occur annually under SSP5-8.5. With 65 million people (68% of the total population) experiencing increased heat danger days, the overall heat risk in the region will swell by a factor of 50. Conversely, under the low-emission pathways (SSP1-2.6 and SSP2-4.5), the annual heat danger days will remain similar to the present level or increase slightly. The result indicates the need to develop strategic plans to avoid the increased heat risk of urban agglomerations under high emission-population pathways.


Introduction
Human health is sensitive to high temperatures (WHO 2008). In very hot and humid conditions, the human body's ability to evaporate heat through sweat will decrease, causing heat stress (Koppe et al 2004, Parsons 2014, Liu 2020, Li et al 2021. Heat stress can increase the human body's core temperature and heart rate. In addition to classical heat illnesses, including heat stroke, cramps and syncope, heat stress can induce diabetes, respiratory disease, cardiovascular disease and others (Guo et al 2018, Casanueva et al 2019. These conditions may cause even deaths (Hoag 2014, Matthews et al 2017, Baldwin et al 2019. For example, the 1995 heatwave resulted in 739 deaths in Chicago (Dematte et al 1998). The 2003 extreme heatwave in Europe killed at least 70 000 people and severely damaged the economy (Robine et al 2008). Under global warming, heat stress will become one of the most widespread and immediate dangers (AghaKouchak et al 2015, Coffel et al 2017, posing a serious threat to human health, energy infrastructure, and outdoor activities, especially in areas with large populations (Mora et al 2017, Schuster et al 2017.
China is the most populous country in the world, has about 59% of its population living in urban areas (http://world-statistics.org), and is one of the countries severely affected by extreme heat. Extreme hightemperature events in China occurred frequently in the past 20 years (Sun et al 2017, Zhang et al 2020a, causing many deaths. For example, the 2013 recordbreaking high-temperature event in Shanghai caused about 160 deaths in Pudong New District (Sun et al 2014). Given its increasing population (Jiang et al 2017) and rising air temperature (Zhou and Ren 2011, Zhao et al 2015, Zander et al 2018, governments should build necessary health infrastructure in advance for the expected changes and must prioritize investment in places where impacts are concentrated. The COVID-19 pandemic is a practical testimony how disaster it could be without early warning and adequate infrastructure in place. However, most existing projections for China were made using either high-temperature event changes (Sun et al 2017, Yu et al 2018, Zhang et al 2020a  Previous studies usually explored extreme heat events at the national level (Li et al 2016, Sun et al 2017, Liu 2020, paying little attention to regional amplification in urban agglomerations. As an inevitable outcome of China's new industrialization, its urbanization has reached an advanced stage of development over the last 30 years (Fang 2015). Given the rapid urbanization (Yang 2013), there is a pressing need to investigate heat stress impacts in China's urban agglomerations (CUAs). However, urban agglomerations remain a weak link in Chinese scientific research and are in urgent need of investigation for strategic planning (Fang 2015, Yu et al 2018.
It is critical that China intervenes in the climate crisis. Strategic mitigation and adaptation to contain climate change and its impacts will provide significant health benefits to China's 1.4 billion people, and incorporating these strategies into detailed pathways to achieve carbon neutrality commitments will ensure improved human well-being (Cai et al 2021). To support strategic planning, we attempt to project the future heat risks in China by considering both population growth and heat stress. Meanwhile, we select to focus on the top five CUAs (figure 1(a): Beijing-Tianjin-Hebei (BTH) region, the Yangtze River Delta (YRD) region, the Middle Yangtze River (MYR) region, the Chongqing-Chengdu (CC) region, and the Pearl River Delta (PRD) region). These densely populated CUAs account for more than 80% of the national commercial carbon emissions and contribute more than 55% of the national gross domestic product (Ma et al 2018(Ma et al , 2019. This study aims to estimate how heat stress risk in these five CUAs will change and the number of CUA residents who will suffer extreme heat stress under future combined emission-population pathways, a topic that has not been addressed in previous studies. Our result will provide a more comprehensive, updated projection of China's heat risks, concentrating on CUAs, for strategic planning and prioritizing health infrastructure needs to effectively adapt to the future of a growing population and warming climate. Our approach to assess future heat risks, emphasizing the necessity to combine climate and population changes, can be readily generalized to address similar problems worldwide.

Population data
The population data come with the four shared socioeconomic pathways (SSPs) (Jones andO'Neill 2016, 2020). Each includes a projection of global and urban populations at 10 year intervals from 2010 to 2100 at a latitude-longitude grid spacing of 0.125 • . SSP1 represents sustainable development, meaning less dependence on resources and fossil fuels; SSP2 represents business-as-usual, maintaining the trends of recent decades, achieving some of the development goals and gradually reducing dependence on fossil fuels; SSP3 represents global regional rivalry, having large regional disparities, large wealth gaps, unattainable development goals and increased dependence on fossil fuels; SSP5 stands for fossil-fueled development, with economic development focused on mitigating challenges and solving socioeconomic problems through the pursuit of self-interest (O'Neill et al 2015).

Coupled Model Intercomparison Project phase 6 (CMIP6) climate projection
In its latest assessment, the Intergovernmental Panel on Climate Change considered comprehensive factors such as population, economy, technological progress, and resource utilization, together with newly proposed SSPs to quantify the relationship between climate change and socioeconomic development (Moss et al 2010). This framework resulted in new emissions scenarios that combine SSPs with the future climate representative concentration pathways (RCPs) in the previous assessment (Kriegler et al 2012, van Vuuren et al 2013, O'Neill et al 2015. Under these new SSPs, the CMIP6 has provided the latest climate projections, which are expected to be more reliable as they are generated using improved and a greater number of climate models than in previous phases (Eyring et al 2016, Jiang et al 2020). The CMIP6 projections also offer a unique opportunity to study future heat Based on daily data availability, this study used surface air temperature and relative humidity simulations from 23 climate models in the CMIP6 archives (Eyring et al 2016). These climate models come from 14 institutions around the world. Table  S1 (available online at stacks.iop.org/ERL/16/064073/ mmedia) (supplementary) lists the basic information of these models, whose details can be found at https://esgf-node.llnl.gov/projects/cmip6. The historical simulations (1850-2014) and future projections (2015-2100) under four emission scenarios were analyzed. These scenarios, SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, represent respectively the global forcing pathways of RCP2.6, RCP4.5, RCP7.0, and RCP8.5 under the socioeconomic conditions of SSP1, SSP2, SSP3, and SSP5. All models used in this study produced multiple realizations with varying initial conditions. Here, only the first realization of each model was selected for the analysis.
The selected four scenarios are in Tier-1 and required for all climate models participating in the Scenario Model Intercomparison Project for CMIP6. SSP5-8.5 and SSP3-7.0 represent high-emission pathways, whereas SSP2-4.5 and SSP1-2.6 represent lowemission pathways. SSP5-8.5 consists of the only SSP scenario with high enough emission (SSP5) to generate a radiative forcing of 8.5 W m −2 in 2100. SSP3-7.0 combines relatively high societal vulnerability (SSP3) with a relatively high forcing of 7.0 W m −2 . SSP2-4.5 combines intermediate societal vulnerability (SSP2) with an intermediate forcing of 4.5 W m −2 . SSP1-2.6 combines low vulnerability with low challenges for mitigation (SSP1) as well as a low forcing of 2.6 W m −2 .

Heat stress index (HSI)
Apparent temperature represents heat stress on the human body by accounting for the effects of environmental factors such as surface air temperature and humidity (Steadman 1979, Russo et al 2017. The HSI is a measure that combines temperature and humidity to derive a human-perceived equivalent temperature (Buzan et al 2015). It is refined from the multiple regression analysis of Rothfusz (1990), which is summarized in supplementary. To highlight the extreme events, this study defines heat danger days as those days in which HSI is at the danger level or extreme danger level (table S2), that is, greater than 41 • C (Buzan et al 2015). The overall heat risk is defined here as the product of the number of heat danger days (HSI > 41 • C) and the population in danger.

Response rate to global warming
The HSI and global mean temperature are averaged over 10 year periods to eliminate interannual variability (Collins et al 2013). Beginning in 2016, averages are calculated every five years to obtain a runningmean time series of 16 values (i.e. the averages of 2016-2025, 2021-2030, and up to 2091-2100). A linear regression between these HSI averages and the global mean temperature is conducted, and the resulting regression coefficient is referred to as the response rate of HSI to global warming. The rate indicates the amount by which HSI will change in response to a global mean temperature rise of 1 • C.

Future changes in population
The urban population projections under the four SSP scenarios for all of China (figure 1(b)) and each CUA (figure S1) show an increasing trend until 2050 and a decreasing trend thereafter. China's urban population would peak around 2050 at about 790-1010 million people and then fall to 620-660 million by 2100. The five CUAs would also reach their peaks in the period 2041-2060. SSP1 and SSP5 give very similar projections as their population change assumptions differ only in internal migration rates. Their projections are higher than those of SSP2 and SSP3 until 2080, after which the situation is reversed. Although the urban population is projected to decrease after 2050, its proportion in China's total population would increase from 40% (2010) to 98% (SSP1), 80%-96% (SSP2), 59%-85% (SSP3), and 98% (SSP5) (figure S2). Thus, the urban population would gradually become the majority, creating a pressing need to better understand how climate change will affect people living in China's urban centers.

Future changes in heat stress
This study used the period 1995-2014 as the baseline and selected two future 20 year windows to capture the major population and climate changes in China. As the urban population is projected to peak around 2050 ( figure 1(b)), 2041-2060 was designated the population explosion period. Likewise, since the annual temperature is projected to increase continuously through 2100 (figure 1(c)), 2081-2100 was named the highest warming period. Figure 2(a) depicts the multimodel ensemble (MME) mean and spread (the range between 25th and 75th percentiles) of surface air temperature projections averaged over China and in the five CUAs. During the highest warming period, depending on regions, the temperature would increase by 1.6 (1.3-2.0) • C, 2.8 (2.4-3.4) • C, 4.3 (3.6-4.9) • C, and 5.5 (4.6-6.5) • C under SSP1-2.6, SSP2-4.5, SSP3-7.0, and SSP5-8.5, respectively. In the population explosion period, the corresponding increases would be 1.5 (1.2-2.0) • C, 1.8 (1.5-2.2) • C, 1.9 (1.6-2.3) • C, and 2.4 (2.0-2.9) • C. The warming is highest under SSP5-8.5 and lowest under SSP1-2.6. The projected warming exceeds the national level in most CUAs under the low-emission scenarios (SSP1-2.6 and SSP2-4.5) but does so only in BTH under the high-emission scenarios (SSP3-7.0 and SSP5-8.5). Among the four scenarios, the temperature in the five CUAs would increase by 1.3 • C-2.4 • C and 1.5 • C-5.4 • C during the population explosion and highest warming period, respectively. The warming is highest in BTH and lowest in PRD.
The relative humidity is projected to change less than 6% for both the population explosion and highest warming periods (figure 2(b)). Most CUAs would become drier, with the largest drying in MYR under SSP5-8.5. Notable humidity increases are projected only in BTH under SSP3-7.0. The future HSI trends (figure 2(c)) mostly follow those of temperature (figure 2(a)). The HSI increases would exceed the national trend in all CUAs except CC. These increases are respectively 1.9 • C-3.2 • C and 2 • C-7.8 • C in the population explosion and highest warming periods. Unlike temperature, which would increase most in BTH, HSI would increase most in PRD. Figure 3 plots the geographic distributions of the MME mean HSI response rates to global warming for the two periods under the four future scenarios. The projected response rates over the five CUAs under SSP2-4.5 (only in the population explosion period), SSP3-7.0, and SSP5-8.5 are larger than 1 • C/ • C, indicating that these regional HSIs would increase faster than global warming. In particular, the rates in southern China (MYR, YRD, PRD) would exceed 1.5 • C/ • C, that is, these regional HSIs would increase at least 1.5 • C per global warming by 1 • C.

Future changes in heat danger days and overall risks
To focus on extreme events, we analyzed the prevalence days in which HSI was at the danger level or extreme danger level (>41 • C), indicating a high risk for heatstroke. The annual number of these heat danger days was first calculated at each grid and then averaged over each CUA. The maximum of the CUAmean annual numbers during the present period (1995)(1996)(1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014) was defined as the baseline from which to identify future record-breaking events. Future heat danger days were similarly calculated for each of the five CUAs in both the population explosion and highest warming periods under the four emission scenarios. Figure 4 compares the MME results along with observations. The MME baseline is close to the observed record for each CUA, and all results indicate that the present danger level is less than three days annually.
The average number of heat danger days is smaller during the population explosion period than the highest warming period, and greater under the high-emission scenarios (SSP3-7.0 and SSP5-8.5) than the low-emission scenarios (SSP1-2.6 and SSP2-4.5). In the population explosion period, the number of danger days is projected to exceed the present record only under the high-emission scenarios, about 3-13 days in all CUAs. In the highest warming period, the number of danger days is greater than the present record under all scenarios. Among all CUAs, PRD is projected to have the greatest number of danger days, about 67 days per year during the highest warming period under SSP5-8.5. In the population explosion period, only 10%-50% of the areas in each CUA are projected to exceed the baseline (figure S3). In the highest warming period, larger fractions of each CUA would do so. For example, under SSP5-8.5, about 75%, 85%, 87%, 56%, and 86% of the areas would exceed the baseline in BTH, YRD, MYR, CC, and PRD, respectively.
From a social and governmental perspective, any assessment of extreme heat risk must also account for population changes (Xu et al 2019). Figure 5 compares the projected population living in the areas of each CUA that exceed the baseline number of heat danger days during the population explosion and highest warming periods. Taking the MME mean as representative, the population living in such areas during the population explosion period, is projected to be similar under SSP1-2.6, SSP2-4.5, and SSP3-7.0: 53-61, 47-52, 28-31, 10-13, and 30-34 million people in BTH, YRD, MYR, CC, and PRD, respectively. In contrast, under SSP5-8.5, there would be 10-20 million more people living under these conditions in each CUA; about 260 million people in sum over the five CUAs are projected to suffer at least three heat danger days per year, accounting for 19% of China's total population. During the highest warming period, the number of heat danger days would significantly increase, while the urban population is projected to decline in all scenarios except for SSP3-7.0, in which population changes very little. As a result, there could be more people experiencing an increased number of heat degree days under SSP3-7.0 than SSP5-8.5 in some CUAs; for example, about 103 versus 98 and 78 versus 74 million people per year are projected to suffer in BTH and YRD, respectively. The gap between SSP3-7.0 and SSP5-8.5 would decrease when summed over all CUAs: about 310 versus 316 million people would experience at least three heat danger days per year during the highest warming period, accounting for 30% and 39% of China's total population. Figure 6 compares the projected changes in heat risk under the four SSP scenarios for the five CUAs. In both the population explosion and highest warming periods, heat risks would be the highest under SSP5-8.5 and the lowest under SSP1-2.6. Heat risks in the CUAs for the two periods would have a 90% probability of reaching 4.1-13 (7.0) and 12-51 (28) times the present level, respectively, under SSP5-8.5, but only 0.6-3.8 (2.5) times and 0.4-3.0 (1.9) times under SSP1-2.6 (the value in the parentheses is the mean of all CUAs). In the highest warming period, heat risks would be higher than those in the population explosion period due to the larger increase in heat danger days. In particular, heat risks would have a 90% probability of reaching 1.9-13 (6.9) and 7.1-32 (18) times the present level under SSP2-4.5 and SSP3-7.0, respectively, in the highest warming period. The corresponding risks during the population explosion period would be only 0.9-5.3 (3.5) and 1.0-6.2 (3.9) times the present level.
Heat risks in the highest warming period, relative to the population explosion period, would increase much more under high than low-emission scenarios. Under SSP3-7.0 and SSP5-8.5, heat risks averaged over all CUAs in this period are projected to be 18 and 28 times the present level, approximately quadrupling those (3.9 and 7.0 times) in the population explosion period. The corresponding heat risks under SSP2-4.5 and SSP1-2.6 would be only 6.9 and 1.9 times the present level, respectively, in the highest warming period, less than twice and even lower than those (3.5 and 2.5 times) in the population explosion period. These changes can be very different among CUAs. Heat risks under all scenarios are projected to be the lowest in CC, only 0.6-1.9 and 0.4-12 times the present level in the population explosion and highest warming periods, respectively. Heat risks in PRD would be the highest, exceeding 30 and 50 times the present level in the highest warming period under SSP3-7.0 and SSP5-8.5, respectively. Previous studies, mainly focused on high-temperature changes alone Figure 6. The probability density function (a), (c) of heat risk changes from the present level at individual grids and regional averages (b), (d) over the five CUAs under SSP1-2.6 (green), SSP2-4.5 (yellow), SSP3-7.0 (blue) and SSP5-8.5 (red) for the population explosion period (a), (b) and the highest warming period (c), (d). (Yu et al 2018, Zhang et al 2020a, projected similarly the greatest heat risks in PRD but relatively low heat risks in northern China. Our study, however, showed that CC (in southwest China), rather than BTH (in northern China), would have relatively low heat risks because much larger population is projected in BTH than CC (figure S1).

Discussion
This study assessed future heat risks for five CUAs by combining regional population growth and heat stress changes based on the latest CMIP6 projections. We found that these CUAs would face heat risks up to 50 times the present level under high-emission pathway scenarios, with the greatest increase in PRD. The result calls for China to prioritize intervenes to contain health risks in the climate crisis. Two important questions are whether the threshold (HSI > 41 • C) for a heat danger day is representative for the hazardous heat exposure and whether the heat risk so defined has a positive relationship with heat-related mortality. Figure S4 compares the observed heat risk based on HSI > 41 • C and the heatwave-related mortality derived by Cai et al (2021) in the provinces identified with the CUAs. Both the heat risk and heatwave mortality showed similar increasing trends. Their interannual variations were also positively correlated, in the range of 0.52-0.84, which are statistically significant at the 99% confidence level. Table S3 shows the interannual correlations between the observed and simulated (present) heat danger days, indicating that MME can capture the past variations and trends in the provinces identified with the CUAs. Furthermore, figure S5 compares the heat risk changes projected using the HSI threshold varying from 39 • C to 41 • C at an interval of 0.5 • C, which covers the heat exposure range of Cai et al (2021). For both the population explosion and highest warming periods, varying the threshold does not significantly alter the heat risk projections. For example, the projected heat risk in PRD for the highest warming period under SSP5-8.5 would be around 50 times the present level for all thresholds. Therefore, the heat risk so defined can both capture the observed heat mortality and represent the projected heat stress changes in a wider exposure range.
To answer the question on estimating future heatrelated deaths, let us assume that the heat risk to mortality link would continue to follow the above statistical relationship based on Cai et al (2021). For the provinces identified with the CUAs, the potential heat deaths during the highest warming period (figure S6) would be 2-4 times the low-emission scenario. In particular, the potential heat mortality of Beijing during the highest warming period would increase to 2740-4110 per year under the high-emission scenario, with 356-1041 potential deaths under the lowemission scenario. Moreover, because physiological and thermoregulatory characteristics differ among genders and ages, heat-related health effects may vary among individuals (Huang et al 2011, Bobb et al 2014, Lee and Kim 2016. For example, extreme heat may pose a greater risk to elder people. Thus, a more detailed analysis with more comprehensive data is needed to better quantity the heat risk impacts on public health. We also found that the simulated urban heat danger days were slightly higher than the rural ones ( figure S7). However, global models generally lack urban representation (Zhao et al 2021), neglecting city-level climate responses, which may lead to underestimation of future heat stress increase in the CUAs (Bounoua et al 2015). Different urban development strategies, e.g. urban land expansion versus urban center acceleration, can produce varying urban heat island effects , suggesting the need to consider future policies. Furthermore, surface moisture still contains large biases in the current CMIP simulations (Dunn et al 2017). The biases can be reduced through dynamic downscaling, although their influence on this study's result is minor since the temperature contribution dominates HSI variation (figures S8 and S9). To reduce these CMIP model uncertainties, especially at regional to local scales, we may seek statistical bias-correction or machine learning methods (Grover et al 2019) and dynamic downscaling approaches (Liang et al 2008, 2019). By then we would be able to make a more rigorous assessment of heat risks in individual cities rather than clustered CUAs.
Several issues related to this study warrant further investigation. As shown above, PRD is projected to have greater heat risks than other CUAs. It is interesting to note that the key rising risks from climate change in PRD are not on heat mortality, probably due to the excellent local healthcare conditions and well-established responses to heat hazards, but mainly for labor productivity losses (Cai et al 2021). Located in this region are many key economic cities in China (e.g. Guangzhou, Shenzhen, and Hong Kong). Higher temperatures will make outdoor work more challenging and increase indoor air conditioning use and more energy consumption (Kinney et al , Dunne et al 2013, Yu et al 2019. These may have negative impacts on the regional economy in addition to public health. Therefore, more factors need to be considered to address the hazards posed by heat stress, and effective strategic plans must be developed for different regions depending on their existing infrastructural capabilities.

Conclusion
Heat danger days would increase rapidly in CUAs under the high-emission pathway scenarios (SSP3-7.0 and SSP5-8.5). In the population explosion period (2041-2060) and highest warming period (2081-2100), 3-13 and 8-67 heat danger days (HSI > 41 • C) per year are projected to occur in the CUAs under these scenarios. As such, approximately 20%-50% and 60%-90% of the CUA areas will exceed the current baseline number of heat danger days (i.e. less than three per year). This would cause respectively about 260 and 310 million people to suffer under these conditions, accounting for 19% and 39% of the total population of China. The corresponding overall heat risks (i.e. heat stress danger days times the population in danger) would have a high probability (90%) of reaching 18 and 28 times the present level in the highest warming period, and 4 and 7 times in the population explosion period for SSP3-7.0 and SSP5-8.5, respectively. Although the projected heat risks for 2041-2060 would not be as severe as 2081-2100, they would be still many times the present level. Given the devastation of the current heat stresses and only 20 years to reach the population explosion, strategic planning of potential heat risks becomes imperative and pressing for society and governments.
On the other hand, heat danger days are projected to be similar to the present level under the lowemission pathway scenarios. The corresponding heat risks for SSP1-2.6 would be 2.5 and 1.9 times the present level in the population explosion period and highest warming period. Since the heat danger days are positively correlated with the observed heatwave mortality and their projections are representative of future heat stress exposure ranges, similar magnitude increases of heat-related deaths would be anticipated, varying with regions and scenarios. The result reinforces the need to minimize global emissions and develop strategic plans to mitigate the escalated heat risk under high emission-population pathways, especially in urban agglomerations.
Projected heat danger levels vary largely among different urban agglomerations in China. Under all future scenarios, heat risks are projected to be highest in the PRD and the lowest in the CC region. In particular, in the highest warming period under SSP5-8.5, PRD would have 67 heat danger days per year and 65 million people (68% of its total population) living under increasing danger, escalating the overall heat risk over 50 times the present level. In contrast, CC would have only 17 heat danger days and 28 million people (52% of its total population) living under increasing danger, elevating the overall heat risk by 12 times. Therefore, considering the future heat risks, the CC would be more livable among the CUAs.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// esgf-node.llnl.gov/projects/cmip6. Baldwin