Climate change dominates the increasing exposure of global population to compound heatwave and humidity extremes in the future

Abstract


Introduction
Under climate change, the frequency and/or intensity of various extremes (such as heatwaves, droughts, and heavy precipitations) increases, causing huge loss of life and economic damage (IPCC, 2021;Kurz et al., 2008;Lesk et al., 2017).For instance, 2003 European heatwave event killed tens of thousands of people and the 2010 Russian heatwave event led to higher global food prices than normal years (Haines et al., 2006;Wegren, 2013).In the future, global warming will continue due to human emissions of greenhouse gases, and the frequency and intensity of heatwave will continue to increase, projected to affect more people (Byrne and O'Gorman, 2018).Research on extreme events has received significant attention in recent years due to their severe impact on human societies and economies (Alexander et al., 2006;Cook et al., 2020;Q Zhang et al., 2022).Compound events can be defined as two or more extreme events occurring simultaneously or successively, such as compound flood and hot events (Gu et al., 2022), combination of tropical cyclones and moist heat (Rajeev & Mishra, 2022), compound drought and heatwave events (Zhang et al., 2022) and compound heatwave and humidity (IPCC, 2021;Coffel et al., 2018).Compared with single extreme events, compound extremes may have more devastating effects on water resources management, human society and ecosystem (Ridder et al., 2020;Rogers et al., 2021;Zscheischler et al., 2018;Tripathy et al., 2023).Specially, the compound heat and humidity extremes (CHHE), which can be defined as the co-occurrence of heatwave and high humidity, have much more severe threat to human than single heatwave event (Coffel et al., 2018;Raymond et al., 2020).Some studies suggested that heat-humidity extremes have increased faster and effect more people than heat extremes (Li, 2020;Rogers et al., 2021).Recent researches found that both high temperature and humidity caused the high mortality in India and Pakistan during the 2015 heatwave event (Wehner, Stone, Krishnan, AchutaRao, & Castillo, 2016).With the acceleration of urbanization, anthropogenic heat emission will increase, which can aggravate urban heat island (Huang, Song, Wang, Chui, & Chan, 2021).The use of cooling and dehumidifying facilities leads to increase urban moisture island (Shi et al., 2019).The hot and humidity cities will regulate region climate and increase heat-related morbidity and mortality.Hence, it is essential to investigate the exposure of global population to compound heatwave and humidity extremes, which is beneficial to provide useful information for the development of adaptation strategies.CHHE affect human body through multiple factors, such as temperature, humidity, wind speed, radiation, human's activities and clothing.The heat index (such as wetbulb temperature (Tw), wet bulb globe temperature (WBGT), discomfort index (DI), and etc.) is widely used to quantify CHHE (Epstein & Moran, 2006;Li, 2020;Sherwood & Huber, 2010).Most of heat indices focused on how to measure the high air temperature and high humidity conditions as both temperature and humidity affect the heat exchange between the human body and the environment (Petkova et al., 2013).At high air temperature, the human body can effectively dissipate heat through evaporation if the humidity is low, but the human body becomes less efficient at evaporating heat with high temperature and high humidity conditions, which can increase body temperature and eventually lead to heat stroke or even death (Raymond et al., 2020).
The threat to human health from heatwave events cannot be accurately assessed if only air temperature is considered (Diffenbaugh et al., 2007;Dunne et al., 2013;Fischer and Schär, 2010).Some heat indexes (such as WBGT) consider many variables such as wind speed, radiation, air temperature, humidity, and these indices may characterize heat stress to human body theoretically.However, the data needed in these index are often limited in availability and quality (Rogers et al., 2021).Previous studies also suggested that the ambient heat stress reflected by WBGT is strongly influenced by clothing and human activity, while Tw establishes clear thermodynamic limits that eliminate these effects (Sherwood and Huber, 2010).Numerous studies have been conducted to analyze the heatwave event at different scales using Tw.Under current climate conditions, Tw rarely exceeds 35℃ (Pal and Eltahir, 2016;Raymond et al., 2020;Schär, 2016).Some studies suggest that Tw will exceed 35°C in South Asia and the Middle East (Im et al., 2017;Pal and Eltahir, 2015), and the number of high-risk days will increase 10-30% in West, Central and Northeast Africa by the end of the 21st century (Fotso-Nguemo et al., 2023).
Population exposure is a common metric to assess the impact of compound events on human society (Feng et al., 2022;Yu and Zhai, 2020).For example, Li et al. (2020) speculate that1.22 billion people will be exposed to extreme wet heat events if the globe warms by 3°C.Population exposure is affected by both climate change and population change (G.Zhang et al., 2022).Coffel et al. (2018)  Generally, previous studies mainly focused on the regional change of CHHE during the historical or future period.Although almost all the studies suggested CHHE is experiencing rapidly increasing trend during the historical period and will continue to increase in the future, there are serval studies evaluating the projection characteristics globally, and only limit future scenarios (representative concentration pathways, RCP 4.5 and RCP 8.5) were selected (Ballester et al., 2023;Rogers et al., 2021;Wehner, Stone, Krishnan, AchutaRao, & Castillo, 2016).Such as Coffel et al. (2018) evaluated the change of CHHE and population exposure using CMIP5 data under RCP 4.5 and RCP 8.5.With the development of CMIP6 data, the higher resolution and improved physical processes data can be used to projected future change (Zhang et al., 2022).
The global land daily gridded maximum temperature product provided by Climate Prediction Center (CPC-Unified) for the period from 1979 to 2014 with a spatial resolution of 0.5°×0.5° is used to validate the GCM's outputs.The CPC-Unified dataset is widely used in hydrometeorological studies due to its strict quality control and high accuracy (Mukherjee and Mishra, 2021;Nashwan et al., 2019;Tarek et al., 2021).The daily surface pressure relative humidity data from ERA5 (the fifth generation of European Center for Medium Weather Forecasting atmospheric reanalysis) is also used to calculate daily Tw.Both CMIP6 model data and observed data were re-gridded to 0.5°×0.5°using bilinear interpolation.We used the Quantile Mapping (QM) method (Cannon et al., 2015;Maraun, 2013) to correct the daily maximum temperature and relative humidity data of CMIP6 based on CPC-unified and ERA5 data.The QM method has been widely used in the bias correction of climate data due to its accuracy and simplicity (Maurer et al., 2010;Tang et al., 2021;Thrasher et al., 2012).And we use root mean square error (RMSE) to assess the performance of bias correction (a smaller value of RMSE suggests a better performance of the bias correction).Additionally, we used 44 global land regions from IPCC AR6 (IPCC, 2021) to assess the spatial variability of CHHE (Figure 1).

Definition of Compound Heatwave and Humidity Extremes
We used Tw to measure the intensity of compound heatwave and humidity extremes.Tw was calculated by the algorithm proposed by Davies-Jones (Davies- Jones, 2008).When the daily ambient Tw is greater than 35°C, the evaporative heat dissipation efficiency of the human body through the skin will be greatly reduced and the body will not be able to maintain a stable body temperature (H.Chen et al., 2022).To maintain the body temperature at 37°C, Tw needs to be less than 35°C (Pal and Eltahir, 2015;Sherwood and Huber, 2010).However, data from the 2003 European heatwave showed that a Tw of around 30°C can cause thousands of human deaths (Fouillet et al., 2008), and empirical data show that working outdoors above 32°C is also very dangerous (Buzan et al., 2015;Liang et al., 2011).Therefore, we mainly investigated the change in Tw with threshold exceeding Tw 35°C and we also explore the case of Tw exceed 32°C as supplementary analysis in this study.This provides an early warning for vulnerable people, such as children and the elderly, in these areas, as well as an indication of higher-risk areas in the future.
The calculation of Tw is given as follows: where Tw is wet bulb temperature (K), TE is equivalent temperature (K), rs is saturation mixing ratio calculated by Tmax, HR is relative humidity, p is surface pressure (Pa), p0 is standard atmospheric pressure (Pa).

Attribution of Population Exposure to CHHE
Population exposure is defined as the product of the number of compound event days and the number of people in each pixel (H.Chen and Sun, 2021;Tuholske et al., 2021).We used the approach developed by Jones et al. (2015) to analyze the contribution of population and climate change to the increase of population exposure, which has been widely applied to the attribution analysis of extreme events (Ullah et al., 2022;Weber et al., 2020).We attribute the change of population exposure to three parts: population change, climate change and population-climate interactions: where ΔE represents the change of population exposure, CEH and PH represent the occurrence of compound events and population in the historical period, and ΔCE and ΔP represent their change in the future compared to the historical period, δ represent data bias.The contribution rate of each item can be calculated as follows: where CRP, CPclim and CRint represent the contribution rate of population change, climate change and population-climate interactions, respectively.
3 Results and discussions

Projected Changes in Tw characteristics
We first examine the performance of the raw and corrected multi-model ensembles average in simulating the Tmax and Hr (Figure 1).For Tmax, we find the raw CMIP6 data tends to underestimate the annual average.According to Figure 1b, QM method is better for high temperature corrections.As this study mainly focuses on mid to low latitude regions, the effect of the poorer correction in low temperature on the results can be ignored.For Hr, we find that the raw CMIP6 data slightly overestimates the annual average, and the correction effectively reduces the error.Overall, the QM method effectively reduces the RMSE of the model data and the corrected CMIP6 data can simulate the Tmax and Hr of the historical period well, so we speculate that it is reasonable to use the corrected CMIP6 data to calculate the future compound extremes (Figure 2).pixels under four future scenarios.The spatial distribution of Tw increases in the three scenarios with higher emissions (SSP2-4.5,SPP3-7.0,SSP5-8.5) is similar, with the greatest warming in northern North America, central Africa, the Qinghai-Tibet Plateau, and the Malay Archipelago than the remaining regions.In the near term, there is no significant difference in increase between the four future scenarios, the warming in all scenarios is around 1.8°C.But for the long term, the SSP3-7.0 and SSP5-8.5 scenarios have very significant warming, with most of the global land pixels warming above 5°C and some regions will warm by more than 7°C.We then calculated the full time series of the global average Tw and probability density function (PDF).According to Figure 4, the four future scenarios do not differ significantly in warming magnitudes until 2050.After 2050, the warming in the SSP1-2.6 scenario almost stops and remains at 1.8°C due to lower emission levels and mitigation measures, while the other three scenarios show continuous warming, with the SSP5-8.5 scenario showing the fastest warming, reaching 6.7°C at the end of 21 st century.The same conclusion can be drawn from the probability density plot, where both the mean and variance are smaller for scenario SSP1-2.6,implying that the warming in this scenario is smaller and closer to historical values.The higher emissions scenarios have larger means and variances, implying higher warming level.We divided the global land area into 46 sub-regions based on the IPCC AR6 and analyzed 44 non-polar sub-regions.Figure 5 shows the Tw of history period and four future scenarios over different regions.In the historical and near term, the average Tw is below 32°C in all regions of the world, and there was little difference in warming between sub-regions.In the mid-term, the average Tw of CAR and SEA is close to 35°C, meaning that in some areas Tw has already above 35°C, which can pose a serious threat to human body.In the long term, the average Tw of CAR, SEA, WAF, CAF, NEAF, SEAF and SEA is close or more than 35°C, meaning that the equatorial region will face serious threats in the future.By the end of 21 st century, the GIC, NEN, CAR, NEAF and SEAF sub-regions will have warmed much more than the global average, approaching 9°C in scenario SSP5-8.5.This puts a huge strain on the adaptive capacity of the local population.Although the latitude of NWN, NEN and GIC is high and Tw is still safe for human in the future, the significant increase of Tw in these regions may pose other environmental problems that will also require attention in the future.

Projected Changes in Population Exposure
Since Tw is a measure of environmental heat stress on humans, it is necessary to combine population data to assess the impact of Tw rise on humans.In the historical period, only a few land pixels of Tw occasionally exceeded 35°C.Figure 6(a-d) shows the spatial distribution of the number of days in a year when Tw exceeds 35°C for different emission scenarios at the end of 21 st century.In scenarios SSP1-2.6 and SSP2-4.5, the exposed areas at the end of 21 st century are very similar to the current ones due to the low level of warming, and only a few pixels such as CAR, WAF and SEA have some exposure.In scenarios SSP3-7.0 and SSP5-8.5, exposure occurs in regions near the equator, in the Caribbean, and in southeastern China.The area exposed under the scenario SSP3-7.0 is small, but the excessive population makes the exposure under this scenario almost equal to scenario SSP5-8.5 which has the higher temperature.In addition, the area of exposure in the Indian region is not large (308 and 642 thousand square kilometer under scenario SSP3-7.0 and SSP5-8.5, respectively), but the exposure remains large due to the extremely high population density in the region.Figure 6.Spatiotemporal features of future population exposure and exposure area.(ad) Spatial distribution of exposure under four scenarios (SSP1-2.6,SSP2-4.5,SSP3-7.0,SSP5-8.5) in the long term, respectively.(1-3) is an enlargement of focus area of (a-d).(e) Sub-regions exposure area for each scenario at 35°C threshold.(b) Subregion exposure area exceeded different thresholds under scenario SSP5-8.5.
We also analyzed the time series of population exposure (Figure 7).Population exposure is low for each scenario until 2050, and after 2050 there is a slow increase in exposure in the SSP2-4.5 scenario and a rapid increase in exposure in SSP5-8.5 and SSP3-7.0.In the middle and late stages of scenario SSP1-2.6,population exposure tends to decrease as Tw stops increase and the population decreases, but in the SSP2-4.5 and SSP5-8.5 scenarios, the increase in Tw offsets the decrease in population and the population exposure remains on an upward trend.Due to the rapid population growth in the scenario SSP3-7.0 and the strong temperature rise in the scenario SSP5-8.5, the population exposure in both scenarios increased rapidly, reaching 10 5 million personday at the end of 21 st century.Furthermore, we examined the number of days of exposure under other thresholds in the scenario SSP3-7.0 and SSP5-8.5 (Figure 8).As the threshold decreases, the exposed area gradually expands to both sides of the equator.When Tw threshold decreases 3℃ to 32℃, exposure increases ten-fold in subregion EAS and ENA, where increase faster than other regions at the same latitude (Figure 9(b)).It may be due to the high urbanization rate and the significant urban heat island effect in these areas.

Contributions of population and climate change to increased exposure
To better understand how heatwaves affect humans and to develop mitigation measures, we analyzed the contribution of future climate and population changes to exposure changes globally and in 44 sub-regions.Population changes accounts for about 20% in 2020 and decreases over time, converging to 0 after 2050 (Fig. 10).In the long term, the contribution of each factor is relatively stable under scenario SSP1-2.6, with climate change accounting for about 80% and the population-climate interactions for 20%.In scenarios SSP2-4.5 and SSP3-7.0, the proportion of interactions gradually increases and approaches 50% at the end of 21 st century, indirectly indicating a further increase in the relative role of population change.In scenario SSP5-8.5, the proportion of interactions increases and then decreases due to strong warming effects, with an extreme value in 2060 and a share of about 10% at the end of 21 st century.These results suggest that climate change dominates future increases in population exposure, particularly in Africa, and the impact of population growth cannot be ignored.Figure 10.Contributions of climate, population and population-climate interaction effects to change in total population exposure in four scenarios.(a-d) Global average contribution proportion for scenarios of SSP1-2.6,SSP2-4.5, SSP3-7.0,SSP5-8.5, respectively.
The contribution of each factor was calculated for the scenarios, considering the variability between regions (Fig. 11).Because not every sub-region has exposure and attribution analysis for a smaller number of exposure sub-regions would have a larger error, we only considered sub-regions with more than 100 exposure pixels.In the scenario SSP3-7.0, the African region has a rapid population increase, so the interaction proportion is significantly higher in the WAF, CAF, NEAF and SEAF sub-regions than in the other regions.In the exposed areas, only EAS has some population decrease, so the interaction proportion is negative.In scenario SSP5-8.5, the EAS, SAS, SEA and Americas all experience population declines at the end of 21 st century, so the interaction proportions for this region are all negative.There is a small amount of population growth in Africa, so the interaction ratio is positive, but less than the scenario SSP3-

Discussion
In this study, we first examined projected changes in the future CHHE.Our result show that global Tw will increase significantly in the future due to climate change.This is consistent with pervious studied in Africa and China (Fotso-Nguemo et al., 2023;H. Chen et al., 2022).Even in the scenario SSP1-2.6 with minimum emissions and mitigation measures, where anthropogenic greenhouse gas emissions are significantly reduced to reach carbon neutrality, there is a 1.8°C warming in Tw, which would result in Tw exceeding 32°C in the CAR SEA region, and outdoor work in summer would be affected.In contrast, under scenario SSP5-8.5 global average warming of Tw would reach 6.7°C, with some areas approaching 9°C.Large areas near the equator would have Tw exceeding 35°, and without cooling facilities, prolonged outdoor activities would become unfeasible.
Under scenarios SSP3-7.0 and SSP5-8.5, the region near the equator will have a large area exposed to Tw>35°, while the scenarios SSP2-4.5 and SSP1-2.6 are safer with few area of exposure.Presumably high forest cover and high relative humidity in areas such as SEA NSA and CAF, resulting in a high heat index.On the other hand, the African region is experiencing higher warming, so the heat stress index is also rising faster.In the future, much of Africa is expected to still lack adequate adaptation and mitigation measures, especially considering the rapid population growth (Asefi- et al., 2018;Weber et al., 2018).Without the establishment of effective cooling facilities in these areas, heatwave and humidity extremes could lead to severe heat stroke and fatalities (Thiery et al., 2021).

Najafabady
In the European heatwave of 2003, a 32°C Tw already caused the deaths of elderly people and children (Coffel et al., 2018;Fouillet et al., 2008).To explore potential future risk exposure and impact on outdoor labor, we calculated exposure under different thresholds for scenario SSP3-7.0 and SSP5-8.5.At the 32°C threshold, there is a several-fold increase in exposure in both equatorial regions.In addition, the EAS and ENA sub-regions have much higher increases in exposure than other regions at the same latitude, but these two sub-regions are more developed in the future, are better able to develop continuous mitigation policies and have more cooling facilities, and the impact of high temperatures on human health will be less than in low-income areas.However, heatwave events can affect outdoor work and increase stress on power systems.In these areas, future studies should pay more attention to the impact of heatwaves on infrastructure and the economy (Rajeev and Mishra, 2022;Yin et al., 2023).
Climate change, population change, and their interaction effect are main factors causing increased exposure (Coffel et al., 2018;Raymond et al., 2020).When we calculated the relative contribution of population, we used historical meteorological data, which rarely exceeded the threshold, making the contribution of population change almost zero (Chen et al., 2022;Coffel et al., 2018).Population exposure in 2020s is small but its uncertainty range is large that means population exposure has large relative bias in 2020s.The large uncertainty make sum of three contribution proportion exceed 1.The portion above 1 represents the effect of data uncertainty.As population exposure increase, the influence of uncertainties decreases, the sum of three As populations in Africa and South Asia will continue to increase in the future, the contribution of population-climate interactions in these regions is greater than in other regions, exceeding 50% in scenario SSP3-7.0.In scenario SSP5-8.5, the population decrease in many sub-regions, leading negative contribution of interaction.
Only several Africa sub-region which have much population having small positive contribution of interaction (Coffel et al., 2018).There is an urgent need to reduce anthropogenic greenhouse gas emissions and reduce the extent of global warming on a global scale.In developing countries in Africa and SEA, mitigation policies, universal access to cooling facilities and controlled population growth should be put in place to help reduce losses in the areas most affected by extreme heatwave events (Fotso- et al., 2023;Tuholske et al., 2021).
Recently, some studies started to focus on the effects of radiation on human health.Previous indicated solar radiation can reduce human endurance exercise capacity (Otani, Kaya, Tamaki, Watson, & Maughan, 2016).Human's head directly expose to solar radiation outdoor, and human's brain is vulnerable to the environmental conditions.Exposure to direct solar radiation at lower temperatures may also be hazardous to human health (Piil et al., 2020).There is longer hours of sunlight and stronger radiation in equatorial regions and peoples there may face the greater threat in the future.
Although we used bias correction and multi-model ensemble data to simulate future CHHE, uncertainty of future data is still inevitable.Uncertainties of Tw change and population increase as time.However, with population exposure increase as time, uncertainties of contribution decrease.Because the larger ΔCE bring larger ΔE, and the value of contribution is related to ΔCE/ΔE, the variance between the different model will be smaller than single factor.Scenario SSP1-2.6 has largest uncertainty because of smallest population exposure.Scenario SSP3-7.0 and SSP5-8.5 have very small uncertainties at the end of century.Besides the model uncertainty mentioned above, observed data, resample method and bias correction method we used may also have uncertainties.Recently, several studies used machine learning method to generate high resolution climate data, these new methods provide a new way to improve the accuracy of future projections (Anderson & Lucas, 2018;Yuval & O'Gorman, 2020).
suggested that changes in global extreme humidity and heatwave events mainly caused by population change, climate change, and the population-climate interactions, and they concluded that the increase in global exposure was largely attributable to climate change.Similar work given by Chen et al. (2022) also concluded that the contribution of population change was almost zero in China.We can find that the conclusions from the previous studies on the contribution proportion of population-climate interactions are inconsistent under different regions and scenarios.
Chen et al. (2022) used CMIP6 data to evaluate the change of CHHE under four Shared Socioeconomic Pathway (SSP) scenarios in China.However, the understanding of change of population exposure is far from enough, especially lacks the further assessment of the spatially variability in different regions and their contributing factors of population exposure to CHHE changes.In this study, we calculated the global Tw and population exposure of CHHE in historical (1979-2014) and various future (2015-2100) scenarios (i.e., SSP-RCP scenario) using CMIP6 data.Furthermore, we quantitatively attributed the changes in exposure into three components, i.e., population change, climate change, and population-climate interactions in different regions and future time periods.The objectives of the study include: (a) reveal the spatial and temporal patterns of global historical and future CHHE, (b) investigate the effects of CHHE on human society, (c) attribution analysis of population exposure to CHHE.Our study can provide a theoretical basis for improving population adaptive capacity and developing mitigation measures.

Figure 2 .
Figure 2. Bias correction performance of annual average Tmax form multi-model

Figure 3 .
Figure 3. Spatial features of Tw using 1979-2014 baseline for diverse future scenarios

Figure 4
Figure 4 (a) Time series and probability density function (b) Probability density function of global annual average Tw relative to 1979-2014.Shading areas denote the interquartile ensemble spread, i.e. the range between the 25th and 75th percentiles of the model ensemble, representing the inter-model uncertainty.

Figure 7 .
Figure 7. Time series of future global population exposure to CHHE under four

Figure 8 .
Figure 8. Spatial distribution of exposure at different thresholds of Tw at the end of

Figure 9
Figure 9 (a) Sub-regions population exposure for each scenario at 35°C threshold.(b) Sub-region population exposure exceeded different thresholds under scenario SSP5-8.5.

7. 0 .
According to Figure11, WAF, CAF, NEAF and SEAF sub-regions will experience higher warming than the global average, and their populations will grow rapidly with lack of adaption and cooling infrastructures.

Figure 11 .
Figure 11.Proportion of contribution of population effect, climate effect and interaction effect to population in each sub-region at the end of 21 st century under scenario (a) SSP3-7.0 and (b) SSP5-8.5, respectively.
factors' contribution proportion is close to 1.The proportion of population climate interactions indirectly reflects the impact of population change when the change of population exposure is large and the uncertainties of data cannot influence the result too much.Climate change dominates the increase in exposure at all future times in all scenarios.

Table 1
Selected CMIP6 models in this study UKESM1-0-LL National Centre for Atmospheric Science, UK/ Met Office Hadley Centre, UK 192×144 Figure 1 Geographic location and description of 44 land regions derived from IPCC AR6 (IPCC, 2021)