Projections and patterns of heat-related mortality impacts from climate change in Southeast Asia

This study aims to investigate the impact of climate change on heat-related mortality in Southeast Asia in the future. The ensemble mean from five General Circulation Models (GCMs) including the Flexible Global Ocean-Atmosphere-Land System Model: Grid-Point Version 3 (FGOALS-g3), Max Planck Institute Earth System Model Version 1.2 (MPI-ESM1-2-LR), EC-Earth3, The Meteorological Research Institute Earth System Model Version 2.0 (MRI-ESM2-0), and Geophysical Fluid Dynamics Laboratory Earth System Model Version 4 (GFDL-ESM4) was used to project severe temperatures and heat indices in Southeast Asia under the Coupled Model Intercomparison Projects Phase 6 (CMIP6). This data was used to correlate with mortality data from the Global Burden of Disease database to quantify heat-related mortality in the region. The ensemble mean results show a reasonable level of accuracy in capturing temperature patterns in the Southeast Asian region with an R2 of 0.96, root mean square error (RMSE) of 0.84 and a standard deviation of residual (SDR) of 0.02. When compared to the baseline (1990–2019), temperature extreme indices are rising across all climatic scenarios, with a substantial increase in the SSP3–7.0 and SSP5–8.5 scenarios, ranging from 10% to 50% over the regions, with the heat index predicted to peak in the middle of the century. The two low-emission scenarios, SSP1-2.6 and SSP2-4.5, on the other hand, anticipate more moderate increases, indicating a potentially less severe impact on the region. As a result, under high-emission scenarios, there is expected to be a significant increase in heat-related mortality across Southeast Asia. The expected impact is estimated to affect between 200 and 300 people per 100,000 people from 2030 to 2079, accordingly. Our results highlight the critical need to address health-related impacts of climate change in this region.


Introduction
Global communities are currently dealing with a significant and complex issue due to ongoing human-induced greenhouse gas emissions.According to the World Meteorological Organization's report in (2021), the average global temperature at the surface in 2020 increased by 1.2 °C ± 0.1 °C relative to the pre-industrial period.This temperature rise places 2020 within the top three hottest years ever recorded.Climatic and extreme weather conditions will probably persist in the next few decades (Stott et al (2010)).The primary objective of the Paris Agreement is to hold global warming to well below 2.0 °C above pre-industrial levels and pursue efforts to limit it to 1.5 °C (Amnuaylojaroen and Parasin 2022).As a result, according to some studies (Rogelj and Knutti 2016, Smith et al 2019, IPCC 2021), achieving this goal appears to be extremely difficult.
Undoubtedly, climate change has wide-ranging effects on the environment, economy, and society, posing especially serious implications for human health.The relationship between high temperatures in the environment and many health outcomes, including illness and mortality, has been extensively studied and consistently shown to be positively correlated (Basu 2009).The onset of heat-related illnesses becomes apparent immediately following exposure (Chung et al 2015).Consequently, there is a growing recognition that mortality rates tend to rise gradually when temperatures fall below or exceed a temperature limit that is peculiar to a particular region (Braga et al 2002, Baccini et al 2011).Over the past few decades, several studies have been conducted to evaluate the potential health consequences of increased temperatures.This research has employed the connection between temperature and mortality as well as anticipated future temperatures derived from various climate change scenarios (Baccini et al 2011, Heaviside et al 2016, Díaz et al 2019).However, a recent systematic review shows that many health impact assessment (HIA) studies have been too narrow in their scope, even though most research papers agree on the expected rise in deaths related to heat (Sanderson et al 2017).Previous studies have predominantly relied on a limited number of models, concentrated on moderate-to-high emissions scenarios, and disregarded future socioeconomic projections.However, recent observational investigations have examined historical temperature data alongside medical records.These studies have revealed a notable decline in the severity of heat-related mortality impacts (Hondula et al 2015, Todd and Valleron 2015, Barreca et al 2016, Chung et al 2017).Moreover, several studies have employed climate models to predict the potential consequences of global warming on heat-related mortality.For example, a climate model with global scope predicted a surge of 250% in the number of deaths caused by heat annually by the 2050s.Based on the work of McMichael et al (2003), high-emission scenarios predicted a 75% increase in the annual mortality rate due to heat among people 65 years of age and older in six temperate towns in Australia by 2050.It has also been anticipated that the number of deaths in Lisbon, Portugal, caused by heat-related conditions during the summer season might potentially rise by a factor of six during the 2050s (Dessai 2003).
The region of Southeast Asia (SEA) is well-known for its notable population density and varied topography, which renders it intrinsically vulnerable to the consequences of global warming (Davidson et al 2017).In recent decades, SEA has witnessed an escalation in the occurrence of severe events, resulting in a significant increase in heat-related hazards.Thirumalai et al (2017) reported that in April 2016, there were unprecedented droughts and prolonged periods of severe temperatures, leading to significant loss of life and extensive economic consequences.Liu and Qin (2023) found that mainland SEA has consistently witnessed an increase in heatwave occurrences across all thermal indices during the past six decades.Furthermore, there has been a significant intensification of heatwaves in the region, particularly in the previous 30 years.Gasparrini et al (2017) reported the influence of climate change on heat-related mortality in SEA.Their report predicted that this region would face a significant increase in heat-related mortality.The study also revealed, based on CMIP5 climate projections, that the highest emission scenario could result in a net increase in excess mortality of up to 12.7% by the end of the century.Further, Singh and Dhiman (2012) emphasized the significance of climate change in causing an estimated loss of more than 2.5 million Disability-Adjusted Life Years (DALYs) across SEA.This loss of life is primarily attributed to heat waves.The studies by Zhu et al (2020a) and Dong et al (2021) suggest that the occurrence of extraordinary heat extremes is expected to increase due to the ongoing rapid global warming.Furthermore, considering the population exceeding 600 million in the SEA region, it is important to note that the intensification of severe weather events linked to the continuing process of global warming will result in an increased vulnerability for a larger number of individuals (Liu et al 2017b, Kumar and Mishra 2020).
It is crucial for municipalities in these areas to accurately assess the anticipated changes in population exposure to extreme temperatures.This information is vital for the effective implementation of adaptation and mitigation strategies.Nevertheless, there is less of studies determining a direct correlation between mortality rates in this region and the heat caused by climate change in SEA.Understanding the potential consequences of projected climate change on mortality rates connected to heat-related occurrences in SEA is of utmost importance since this knowledge is essential for protecting the well-being of the diverse populations in the region.The existing amount of scholarly literature pertaining to the correlation between mortality rates and exposure to temperature extremes in the SEA has a dearth of comprehensive coverage.This study aims to investigate the potential risks of projected climate change based on CMIP6 projection on heat-related mortality in SEA.

Methodology
SEA is defined by its geographical boundaries, extending from 10°S to 28.5°N in latitude and from 92°E to 130°E in longitude.The domain under consideration in this study can be further subdivided into three discrete subregions (figure 1).The Indochina Peninsula subregion comprises Thailand, Myanmar, Laos, Vietnam, Cambodia, and Malaysia, spaning from a latitude of 1°N to 28.5°N and from a longitude of 92°E to 110°E in an eastward direction.The second subregion, ranging from 5 to 20°N, and from 118 to 130°E, covers the country of the Philippines.The third subregion comprises three main islands of Indonesia, namely Kalimantan (KAL 4°S-6°N, 109°to 118°E), Sumatra (SUM 8°S to 6°N, 95°to 108°E), and Sulawesi (SUL 6°S to 3°N, 118°to 126°E).
In this study, the monthly temperature and temperature extreme indices are derived from five Global Climate Models (GCMs) acquired from the Coupled Model Intercomparison Project Phase 6 (CMIP6), including FGOAL-g3, MPI-ESM1-2-LR, EC-Earth3, MRI-ESM2-0, and GFDL-ESM4 under four Shared Socioeconomic Pathway (SSP) scenarios including SSP1-2.6,SSP2-4.5, SSP3-7.0, and SSP5-8.5 (Mauritsen et al 2019).To evaluate the performance of MPI-ESM1-2-LR during the period of 1990-2019, the datasets of the Modern-Era Retrospective Analysis for Research and Applications, Version 2 (MERRA2) are selected for comparison in this study (Gelaro et al 2017).The models conducted a regridding process to interpolate their data onto a standardized grid prior to employing Bayesian Model Averaging (BMA) for analysis of the multi-model ensemble.The temperature from the ensemble mean was used to correlate with all-cause mortality from the Global Burden of Disease (GBD) during 1990-2019 to estimate future heat-related mortality under several climate scenarios.

Data used
We utilized data from five Global Climate Models (GCMs) acquired from the CMIP6 for this study (table S1) (Eyring et al 2016).These models exemplify a wide range of methods for simulating the climate system, exhibiting the differing expertise and abilities of international climate research institutions.The presence of diverse elements is essential in order to encompass an extensive range of physical, chemical, and biological phenomena, as well as their intricate interconnections within the Earth system.Each model provides a distinct mix of resolution and complexity, which is essential for effectively depicting atmospheric, oceanic, and land dynamics.For example, the high-resolution air component of MRI-ESM2-0 enhances the detailed vegetation dynamics in EC-Earth3 Veg, offering a comprehensive perspective of the climate system.The diverse range of resolutions and intricacies enable thorough evaluations, encompassing worldwide trends and unique local nuances that are crucial for comprehensive climate forecasts.The rigorous validation of these models against historical climate data underscores their reliability (Kim et al 2020).Their demonstrated capacity to precisely replicate established climate patterns and trends instills confidence in their projections for future scenarios.The accuracy of past simulations is crucial since it serves as a fundamental indicator of a model's dependability in forecasting future climates.
The FGOALS-g3 model, developed by the Chinese Academy of Science (CAS), is a comprehensive system that combines the atmospheric, ocean, land, and sea ice components.It is designed with a moderate level of complexity and resolution.It is intended for conducting both historical and projection simulations as part of the CMIP6 framework (Li et al 2020).The Max Planck Institute Earth System Model, specifically the low-resolution version (MPI-ESM1-2-LR), is renowned for its ability to accurately depict the various components of the earth system.This makes it particularly useful for making long-term projections of climate change.The model has been widely utilized in the CMIP6 ensemble (Mauritsen et al 2019).EC-Earth3 integrates sophisticated atmospheric dynamics with a meticulous depiction of land surface processes.The CMIP6 is enhanced by its contribution, which specifically targets climatic variability and change across several temporal scales (Dösche 2021).The Meteorological Research Institute Earth System Model version 2.0 (MRI-ESM2-0), developed by Japan, incorporates extensive analysis of the atmosphere, ocean, and biosphere.The model is renowned for its atmospheric component's exceptional level of detail (Yukimoto et al 2019).The GFDL-ESM4, developed by the Geophysical Fluid Dynamics Laboratory, is a cutting-edge Earth System Model that accurately simulates past, present, and future climate conditions.The notable feature of this is its focus on biogeochemical cycles (Dunne et al 2020).The models were chosen based on their varied depictions of climate systems, enabling a thorough evaluation of future climate forecasts.In this work, the outputs of the models were adjusted to a uniform grid resolution to guarantee comparability among the ensembles.The temperatture extreme indices including TX90p and TN90-and Heat Index from ensemble mean were used to ass the heat event in SEA during 1990-2100 (table S2) (You et al 2011, Sillmann et al 2013, Sheikh et al 2015).
MERRA2 is the most recent generation of atmospheric reanalysis in the modern satellite era, with a resolution of roughly 0.5°× 0.625°and 72 levels covering the earth's surface up to 0.01 hPa.This dataset was produced by NASA's respected Global Modeling and Assimilation Office (GMAO 2015).MERRA2 incorporates observational modalities previously unavailable in MERRA, as well as improvements to the Goddard Earth Observing System (GEOS) model and assessment framework, allowing for the establishment of a reliable and ongoing climate assessment that expands over the spatial regions of MERRA (Gelaro et al 2017).The Goddard Earth Observing System (GEOS-5.12.4) atmospheric data assimilation system is used to construct the MERRA-2 dataset (Wu et al 2002, Putman and Lin 2007, Rienecker et al 2008, Kleist et al 2009b, Molod et al 2015).The MERRA-2 system shares many core properties with the GEOS-5.2.0 (Rienecker et al 2011), but with a few notable improvements.A comprehensive overview of the preceding updates is provided herein, together with supplemental material from the related publications as cited.In the absence of explicit requirements, the other aspects of the system setup, data input, and method of preparation follow the standards provided in Rienecker et al (2011).The aspects encompass advancements made to the forecast model, enhancements to the analysis algorithm, refinements to the observation system, improvements in the methods employed for radiance assimilation, and modifications to the bias correction techniques utilized for aircraft observations.The mortality data were extracted from the Global Burden of Disease (GBD) database.It represents an unparalleled scientific endeavor of immense magnitude, aimed at comprehensively assessing and quantifying the prevailing levels and dynamic patterns pertaining to health across the globe.The GBD initiative consistently generates periodic assessments pertaining to overall mortality rates, cause-specific fatalities, premature mortality-induced years of life lost (YLLs), years lived with disability (YLDs), and disability-adjusted life years (DALYs).The essential benchmarks for continuous estimation encompass periodic revisions to the GBD estimation.To achieve the utmost comprehensiveness and comparability in the estimates, a thorough reestimation of the entire time series dating back to 1990 is conducted for each cycle, utilizing all available data.The data were collected from various sources, including hospitals, governmental institutions, surveys, and global databases (GBD 2019).The GBD database mortality statistics show a vast and intricate framework.This dataset spanning from 1990 to 2019, gives a time-series perspective that allows researchers to identify and investigate long-term trends.The GBD studies health patterns on a global scale from a geographical standpoint.The study's primary focus, on the other hand, is on SEA, allowing for more accurate regional investigations.This dataset can be used to identify multiple categories.The general mortality rate, which measures the frequency of fatalities in relation to population size and is commonly expressed as a ratio per 1,000 or 100,000 people, is an important aspect to examine.Because of its unique temporal depth, geographical specificity, and categorically categorized variables, the GBD's mortality data is a great resource for understanding health trends in Southeast Asia.

Bayesian model averaging for multi-model climate ensembles
The ensemble consists of five Global Climate Models (GCMs) acquired from the CMIP6 including FGOAL-g3, MPI-ESM1-2-LR, EC-Earth3, MRI-ESM2-0, and GFDL-ESM4.Each model provides its surface temperature and associated indices, such as TX90p, TN90p, and Heat Index projections, for both the historical period (1990 to 2014) and the future period (2015 to 2100).These predictions are then adjusted to a similar spatial resolution to maintain uniformity across the group.The regridding procedure utilizes linear interpolation to map model outputs onto a predetermined grid of latitude and longitude, enabling straightforward comparison.We employ a regridding procedure to interpolate model data onto a standardized grid.
The regridding procedure entails determining an interpolated value for each point on the target grid, which is composed of coordinates (x t , y t ), based on a source grid with coordinates (x s , y s ).The interpolated value ′V′ on the target grid is calculated for a variable V specified on the source grid, using a selected interpolation method.An extensively employed technique is bilinear interpolation (Jone 1999), which can be mathematically represented as: The expression ( • ) V x y x i x i denotes the values of the variable at the four closest adjacent grid points to the source grid surrounding the target point (x t , y t ).The weights, wi, are determined by the relative distances between each of these four locations and the target point.
The weights in bilinear interpolation are calculated as follows: The variables Δx and Δy represent the distances between nearby points on the source grid.The variables x dist i and y dist i represent the horizontal and vertical distances, respectively, between the ith neighboring point and the target grid point.
This study use Bayesian Model Averaging (BMA) (Hoeting et al 1999) to combine projections from five global climate models (GCMs).BMA is a statistical framework that allows for the combination of predictions made by various models, using probability theory.The principle behind this approach is Bayes' theorem, which can be mathematically formulated in the context of model averaging as follows: The expression ( | ) p M y i represents the posterior probability of model Mi given the data y.Similarly, ( | ) p y M i denotes the likelihood of the data y under model M i .The term ( ) p M i refers to the prior probability of model M i , while p(y) represents the marginal likelihood of the data y and serves as a normalizing constant.
The BMA predictive distribution for a new observation  y is calculated by taking a weighted average of the predictive distributions from each model.
The equation involves the predictive distribution,  ( | ) p y M , i which represents the probability distribution of the new observation  y given the model M .
i Additionally, ( | ) p M y i serves as the weight for model M , i and it is calculated as explained earlier.
The ensemble of climate models provides a combined Bayesian Model Averaging (BMA) prediction for the surface temperature (T) at a specific grid point (x, y) and time (t).
Ti(x, y, t) represents the forecast generated by model i, whereas w i (x, y, t) is the Bayesian Model Averaging (BMA) weight assigned to model i at grid point (x, y) and time t.This weight is determined based on the posterior probability p(M i |y), which is particular to the given place and time.
The quantification of uncertainty in the BMA combined prediction is determined by the standard deviation of the BMA predictive distribution.
where σ BMA (x, y, t) denotes the level of uncertainty at the specific coordinates (x, y) and time t on the grid.The variability of the posterior distribution indicates the level of uncertainty in the aggregated forecast and offers understanding about the level of agreement among the models.

Evaluation of model performances
The horizontal resolution disparity between MERRA2 and ensemble mean necessitated the interpolation of MERRA2 data to align with a standardized grid.This was accomplished through the utilization of the regridding function in the NCAR Command Language (NCL) during the evaluation of the model's performance with several statistical measures including the mean bias error, standard deviation of residuals (SDR), correlation coefficient(r), root mean square error (RMSE), and R-squared (R 2 ).The mean bias was calculated as follows equation (1): In the context of this study, let M represent the model data, while O denotes the observed data.The calculation of the standard deviation of residuals (SDR) was performed according to equation (2).
Also, the observed data denote as X O and the model data as X .M Furthermore, X O represents the arithmetic mean of the observed data, while X MO symbolizes the arithmetic mean of the model data.It is important to note that n represents the number of both the model and observed data sets.
The Pearson correlation coefficient was computed in accordance with the prescribed mathematical equation (3): where r is the correlation coefficient.
The computation of the root mean square error was performed utilizing the prescribed mathematical expression, denoted as equation (4): where M is the model data, and O is the observed data.

Estimations of heat related mortality.
A health impact assessment was conducted to ascertain the prevailing and prospective mortality rates associated with heat-related conditions within SEA.To effectively delineate the extant mortality risk associated with heat among populations residing in SEA across diverse climatic regions, we employed a uniform modeling methodology and employed heat-mortality risk functions derived from the seminal work of Leone et al (2013).The functions in equation (5) under consideration pertain to the collective influence exerted by modeled temperature (T) and relative humidity (RH), which is mediated by the dew-point temperature (T m ).
= -+ + [ ( )] ( ) AT 2.653 0.994T 0.0153 T T, RH 5 As described in Hajat et al (2023), this method utilized spline functions to effectively represent exposuremortality relationships in various circumstances.These spline functions demonstrated a linear increase in mortality risk beyond specific heat thresholds, allowing for the quantification of heat coefficients using linearthreshold models.
Subsequently, we employed the temperature-mortality functions and projections of population in conjunction with the climate data to derive estimations regarding forthcoming heat-attributable mortalities across diverse climate change and socioeconomic scenarios.The heat-associated mortality (HM t ) attributable to monthly apparent temperature was calculated for each grid cell in accordance with the methodology defined by Hajat et al (2023) as shows in equation (6).
The coefficient denoted as b serves as a quantitative measure for assessing the risk associated with mortality resulting from heat-related incidents.The heat threshold is a critical parameter that warrants consideration.The variable denoted as AT represents the average apparent temperature observed monthly.DM, on the other hand, signifies the fundamental all-cause mortality rate.P 0 denotes the initial population size at a given time point, while P t represents the population size in subsequent time periods.The estimation of future population size was derived through the utilization of exponential change, which can be mathematically represented by the subsequent equation ( 7 The heat-exposure coefficient, as estimated by Nguyen et al (2023), indicates that a rise of 1 °C in the annual mean temperature corresponds to a 2.73% increase in the monthly mortality rate.To address variations in heat limits across heterogeneous geographical areas, a hypothesis was formulated positing that the threshold value within each individual grid cell would align with the 95% confidence interval of the temperature distribution.While the absolute values of heat thresholds may vary across different geographical locations, there is often a notable level of consistency when these thresholds are examined in relation to the percentiles of temperature distributions specific to each respective locality.The all-cause mortality rate utilized data from the Global Burden of Disease (GBD) study for the baseline period spanning from 1990 to 2019 (table S3).In contrast, for future estimations, a constant value was employed based on the average value derived from the baseline period.
The estimation of forthcoming yearly mortality resulting from heat-related causes was conducted for three distinct time intervals: 2021-2039, 2040-2079, and 2080-2100.In conjunction with the presentation of estimations, an investigation was undertaken to evaluate the effect of anticipated heat-induced mortality on the comprehensive mortality rate, presuming a consistent mortality rate across all causes.As described in Martínez-Solanas et al 2021, Hajat et al 2023, to account for differences in heat thresholds across various geographical regions, we adopted the assumption that the threshold value for each grid cell corresponds to the temperature at the 95th percentile of the unique temperature distribution within that cell over the reference climatological period spanning from 1990 to 2019.While there may be variations in the absolute values of heat thresholds across different locales, there is typically a level of consistency when these thresholds are assessed in relation to the percentiles of local temperature distributions.

Descriptive analysis
Generally, it denotes the expansion of the population, with differing rates of growth observed among countries, and the enduring patterns that will influence the socioeconomic structure of the region.Furthermore, various points have been observed from this data (figure S1).Firstly, it is anticipated that the population of SEA would sustain its growth trajectory throughout the 21st century.In the year 1990, the region's population was estimated to be at 414 million people, with projections indicating a steady growth in the following decades.According to projections, the global population is anticipated to surpass 760 million by the year 2100, exhibiting a growth rate that exceeds twofold.Although the anticipated annual growth rate is predicted to decrease in comparison to historical data, SEA is still likely to undergo significant expansion in the foreseeable future.The rate of population increase varies among countries, with certain nations anticipated to see more accelerated population expansion compared to others.. Table S4 provides data of the mortality rates associated with heat-related incidents in SEA between the years 1990 and 2019, as collected by the GBD dataset.It is evident that there exists considerable disparity in the rates of mortality associated with heat between the countries.As an illustration, in the year 1990, Vietnam recorded the most elevated death rate of 0.093 per 100,000 people, although Laos exhibited the lowest rate of 0.001.In the year 1990, Malaysia witnessed a significant rise in death rates, which was then followed by a gradual decrease in the subsequent years.On the other hand, certain countries have persistent patterns in their death rates.Throughout the whole period, Cambodia and Thailand consistently exhibit low mortality rates, indicating a rather steady condition in terms of heat-related deaths within these countries.Moreover, specific years stand out owing to elevated mortality rates across various countries, such as 1994 and 2007.While Malaysia, for example, had much higher death rates in 1991, 1992, and 1993, with rates of 0.422, 0.410, and 0.399, respectively.The observed successive increase in temperature may indicate the presence of a prolonged and intense heatwave throughout this time period.In a comparable trend, it is noteworthy that Vietnam had a death rate of 0.093 in 1990.The observed rise may imply a significant heat event occurring in that particular year, indicating the need for deeper research on the existing climatic conditions during that timeframe.On the contrary, it is critical to stress that not all cases with low fatality rates should be taken as indicating less severe heat events.Using the Philippines as an example, it is demonstrated that the country had extraordinarily low death rates in 1995 and 2019, with recorded values of 0.009 and 0.021, respectively.However, it is crucial to note that these lower numbers do not always signify minor heat occurrences.Lower death rates may be attributed to factors such as increased preparedness, the use of effective response tactics, or a population that is less susceptible to heat-related consequences.

Climate data evaluation of performance
Figure 2 depict the validation process pertaining to the disparity in monthly mean temperature, specifically for the time span of 1990-2019, as it is averaged over the Ensemble Mean and the MERRA2 datasets.Figure 2(b) shows the spatial distribution of uncertainties across five GCMs shows ranging from 0 °C-2 °C.In general, it can be asserted that the Ensemble Mean exhibits a reasonable level of accuracy in its projections of monthly temperature patterns.The temperature model demonstrates a predominantly cold bias across the entire region, although it is worth noting that a hot bias was observed in select areas of Myanmar, Thailand, and Indonesia (refer to figures 2(a)-(c)).The probability distribution function (PDF) reveals a notable resemblance between the datasets derived from the ensemble mean model and the MERRA2 dataset.Both datasets exhibited temperatures within the range of 0 °C-30 °C.However, it was observed that the MERRA2 dataset displayed a greater concentration of temperatures within the range of 25 °C-30 °C.The statistical computations pertaining to the comparison between the model data and the MERRA2 data have been aggregated on an annual basis, spanning the time period from 1990 to 2019 as listed in table 1.The model exhibits a notable capacity for capturing temperature, as evidenced by the R-squared values of 0.96.While the correlation coefficient between the variables was observed to be 0.98, while the mean bias error (MBE) exhibited a value of 0.001 °C.The root mean square error (RMSE) and standard deviation of residual (SDR) were observed to be 0.84 and 0.02, respectively.
Table 1 shows the uncertainties of multi-model ensemble, consisting of climate models including FGOALS-g3, MPI-ESM1-2-LR, EC-Earth-3, MRI-ESM2, and GFDL-ESM4, exhibits.The FGOALS-G3 model indicates the most extreme warming circumstances, with a mean temperature of 25.4 °C.Additionally, it exhibits a significantly increase median temperature, indicating that its trends lean towards the warmer end of the value for over 50% of the time.On the other hand, the GFDL-ESM4 has the lowest average temperature but has the most variability, with a standard deviation of 5.65 °C.This suggests a wide range of temperatures, indicating more uncertainty in its results.MPI-ESM1-1-LR and EC-Earth-3 have reduced average and median temperatures.Among the ensemble, EC-Earth-3 demonstrates the least variability, as seen by its lowest standard deviation of 5.18 °C.This implies that it may provide more consistent and dependable simulate in terms of internal variability.The MRI-ESM2, exhibits a mean temperature that is around the highest value in the ensemble, along with the second-largest standard deviation.This indicates that while its average simulate lean towards warmer temperatures, there is also a considerable range of anticipated temperatures.When these models are analyzed together using Bayesian Model Averaging, the ensemble encompasses a range of potential climatic results.The narrow range of standard deviations among the models, ranging from around 5.18 °C to 5.65 °C, suggests that despite their different geographical origins, the models exhibit a consistent level of uncertainty in their simulations.

Projection of extreme temperature in Southeast Asia
Figure 3 illustrates the temporal patterns of annual surface temperature, heat index, and temperature extreme indices, specifically TN90P and TX90P, in SEA across various SSP scenarios until the end of the century.In a broad sense, the surface temperature, heat index, and extreme indices exhibit comparable patterns across the various scenarios until the mid-century (2040), after that divergences emerge until to the end of the century.The most extreme escalations occur in scenarios SSP5-8.5 and SSP3-7.0,suggesting substantial increases in temperatures by the end of the century, with SSP5-8.5 possibly reaching up to 30 °C.However, a notable departure from this pattern is anticipated to commence in the year 2050 for each scenario.At the end of the century in SEA, the respective surface temperature maxima for the scenarios SSP1-2.6,SSP2-4.5, SSP3-7.0, and SSP5-8.5 were recorded as 25 °C, 26 °C, 27.5 °C, and 28 °C, respectively.The patterns of the heat index were similar to surface temperature, exhibiting a general rise but with a more pronounced increase after 2050, especially under the scenarios SSP5-8.5 and SSP3-7.0.By 2100, the projected heat index is expected to surpass 42 °C and 40 °C under these scenarios, which is significantly higher than the range of 30 °C to 33 °C observed during the baseline period.At the end of the century in SEA, the heat index reached their respective maximum values of 32 °C, 33 °C, 36 °C, and 38 °C for the SSP1-2.6,SSP2-4.5, SSP3-7.0, and SSP5-8.5 scenarios.The TN90p and TX90P index reveal a sharp increase in the number of warm nights, particularly under the SSP5-8.5 scenario.This implies that by 2100, 90% of both nights and days could surpass the baseline threshold and experience much higher temperatures relative to the baseline.
Figure 4 illustrates the alterations in surface temperature on an annual basis for forthcoming timeframes in relation to the reference period spanning from 1990 to 2019.According to the model's projections, it is anticipated that the average temperature in SEA will reach approximately 4.5 °C during the period of 2080-2100, under a high emission scenario (SSP5-8.5).The SSP1-2.6 scenario predicts a relatively little change in temperature.By the early century, warming is expected to reach up to 1 °C (figure 4(a)), with increases of roughly 1 °C by the mid-century (figure 4(b)).Projections for the late century show a warming of up to 1.5 °C  ).The Indochina region, which includes Thailand, Myanmar, Laos, Cambodia, and Vietnam, as well as particular parts of Malaysia and Indonesia, experiences the most significant rises in heat index.These regions are more vulnerable to the combined impacts of elevated temperature and humidity, determined by the HI.They are projected to have the highest risk of heat-related consequences on human health under the high-emission SSP5-8.5 scenario.
The depiction of the spatial distribution of the pattern of extreme temperature change in SEA can be observed in figures 6 and 7, utilizing the indices TX90P and TN90P, respectively.In figure 6 evident that there is a significant increase in the frequency of exceptionally high temperatures, particularly in relation to the highemission scenario SSP5-8.5.During the period of 2080-2100, it is projected that the TX90p index tends to increase significantly by 30% to 50% throughout the region, with Indonesia being particularly affected by these changes.Throughout the early period (2020-2039), there is a projected increase of 10% to 20% in TX90p, which further intensifies to a 10% to 40% rise by mid-century (2040-2079).The projections indicate that by the end of the century, this index might increase by 30% to 80% across all scenarios, with the greatest values observed in  SSP5-8.5. Figure 7 displays increases in TN90p, which follow a consistent rising trend, particularly in the SEA region where the increments are more noticeable.During the early part of the century (2020-2039), there is expected to be an increase ranging from 10% to 20%.However, projections for the mid-century (2040-2079)   indicate a rise of 30% to 60%.Significant changes are expected to occur primarily in the last decades of the century (2080-2100), with projected increases ranging from 40% to over 90% in some regions.The greatest percentage of the TN90p changes is located in Indonesia, where the index is projected to increase by over 90% from 2080 to 2100 under the SSP5-8.5 scenario.

Projection of heat-related mortality in Southeast Asia
Figure 8 depicts the temporal patterns of annual mortality rates until the end of the century across various SSP scenarios.In most nations, the distribution of burdens across various scenarios exhibits a comparable pattern until the mid-century.The analysis indicates that there is a rise in mortality rates due to heat over the whole region.This increase is most prominent under the high-emission scenarios SSP5-8.5 and SSP3-7.0.Thailand, Cambodia, and Myanmar are projected to experience the most substantial increases in mortality rates throughout the century.In Thailand, the mortality peak is projected to emerge around 2070, resulting in an additional 120 deaths.For Cambodia and Myanmar, the mortality peak is expected to occur around 2050, leading to an additional 90 and 150 deaths, respectively.Notably, the mortality rates in the Philippines and Malaysia are expected to be comparatively lower, potentially because of their more moderate population growth estimates in comparison to other Southeast Asian nations.In general, the anticipated patterns of heat-related deaths are strongly linked to the rising heat indices seen in the area, as shown in figure 5. Regarding the changes in annual death rates (figure 9), it is anticipated that Southeast Asian countries are expected to experience the most significant rises between 2040 and 2079 under the SSP3-7.0 and SSP5-8.5 scenarios.The lower-emission scenarios SSP1-2.6 and SSP2-4.5 indicate a comparatively smaller increase in mortality rates, varying from 10 to 60 per 100,000 individuals.Nevertheless, in the scenarios where emissions are higher, nations such as Laos, Myanmar, Cambodia, and Thailand may experience increases ranging from 10 to 80 per 100,000.This emphasizes the immediate requirement for climate adaptation and public health efforts to alleviate these effects.

Discussion
This study employed the ensemble mean approach to incorporate estimates from different GCMs.The approach entails computing the mean values of the results obtained from multiple models, including FGOALS-g3, MPI-ESM1-2-LR, EC-Earth3, EC-Earth3 Veg, MRI-ESM2-0, and GFDL-ESM4, in order to offer a comprehensive evaluation of future climate conditions.The value of this technique is in its ability to combine various simulations, resulting in a unified projection that is more dependable than the outputs of separate models.Nevertheless, it is important to recognize its limits.An important limitation of the ensemble mean method is its capacity to mitigate severe values by averaging out the fluctuations and extremes projected by different models.This phenomenon can lead to a miscalculation of forthcoming fluctuations and severe weather occurrences that are vital for evaluating their effects (Tebaldi and Knutti 2007).In addition, the conventional approach of giving equal importance to each model does not include variations in performance, which means that less accurate models might have the same influence on the ensemble average as more skillful models (Weigel et al 2010).The assertion that all models are independent is another subject of disagreement.Due to the shared components and developmental lineage among several GCMs, there is a risk of overestimating the ensemble's range as a measure of uncertainty (Pennell & Reichler 2011).In addition, it is important to note that the ensemble mean does not inherently rectify biases and errors.It is possible for shared biases among models to still be evident in the combined output (Pierce et al 2009).Furthermore, the ensemble approach may not adequately capture the intricate and non-linear feedback mechanisms present in the climate system, resulting in uncertainty, particularly in regional climate change estimates (Bony et al 2006).The proficiency of models can fluctuate across various locations, indicating that global model performance does not ensure correctness at the regional level (Christensen et al 2013).
Another point that should be recognized is regridding processes in this study.Basically, it is a prevalent method employed in climate modeling to align datasets from many models onto a standardized grid prior to conducting ensemble averaging.This method is crucial when merging outputs from distinct models, which may include diverse native resolutions and grid patterns.The main issue associated with regridding is the interpolation inaccuracy.Interpolating model data from their original grid to a shared grid involves a certain degree of approximation.Interpolating data might result in data smoothing, geographic detail loss, and the possibility of introducing biases, particularly if the interpolation method does not preserve important values (Jones 1999).The selection of the regridding technique can have a substantial impact on the outcomes, especially for variables that display abrupt changes or discontinuities, such as topography or coasts (Laprise et al 2008).Furthermore, the resolution of the shared grid can have an effect on the average outcomes of the group.A higher resolution can capture more intricate details, but it can also enhance noise and biases that are peculiar to the model being used.On the other hand, a less detailed resolution may weaken significant local characteristics that are crucial for impact studies (Giorgi and Gutowski 2015).Hence, achieving a harmonious equilibrium between computing efficiency and the preservation of vital spatial information is a fundamental factor to take into account during the regridding process.Another factor to take into account is the preservation of physical characteristics.Certain regridding techniques might not uphold the conservation of resources such as energy or mass, resulting in disparities in integrated values throughout space.This could provide significant challenges for research involving water and energy budgets (Gleckler et al 2016).Preserving conservation during the process of regridding is essential for keeping the physical integrity of the climatic data.Meanwhile there is a difference between the ensemble mean temperature and the MERRA2 surface temperature data in Southeast Asia.The difference is mostly negative in places like Indochina and Indonesia.This is because of the complicated topography and distribution of land and sea, which has created a complicated climate pattern (Robertson et al 2011).The land-sea system provides rise to atmospheric circulation patterns that cause uneven seasonal temperatures and rainfall over the region (Yoneyama and Zhang 2020).The results of CMIP6 modeling performance may have been modified by these factors (Hamed et al 2022).Furthermore, the bias may stem from the inadequate representation of complex topographic regions in the coarse topography of the GCMs.This results in the models underestimating the height of mountains and, thus, the surrounding air being warmer than it is.In addition, places with complex topography are susceptible to terraininduced circulations at the meso-to micro-scale, which have a significant impact on the local temperature patterns.These circulations are not accurately captured in global climate models (GCMs) due to their limited spatial resolution (Carvalho et al 2013 and2014).
While the temperature extreme events in this study align with the research conducted by Dong et al (2021) on a regional extreme event in SEA.Furthermore, as demonstrated by Perkins-Kirkpatrick and Gibson (2017) and Seneviratne et al (2012), that the occurrence and severity of extreme events are expected to increase in the future because of global warming.This statement posits that SEA may exhibit a higher susceptibility to the impacts of global warming in comparison to other regions within Asia.Henceforth, with respect to the context, Dosio et al (2018) conducted an analysis on the global prevalence of heatwave incidents under two distinct magnitudes of temperature increase.The observed increase in the HI aligns with the findings of Li (2020), wherein the ECMWF reanalysis dataset was employed to quantify heatwaves in the Southeast Asian region.The individual conveyed that within this area, there is a discernible escalation in all aspects pertaining to heatwave phenomena.This study is in alignment with a prevailing global pattern characterized by a progressive rise in the frequency, duration, and magnitude of heatwaves.Moreover, the study has ascertained that the El Niño-Southern Oscillation (ENSO) and the Indian Ocean Dipole (IOD) exhibit diverse impacts on temperature, precipitation, and heatwaves across different geographical areas (Amnuaylojaroen and Chanvichit (2019), Amnuaylojaroen 2021).
Based on the results in this study indicated that SEA is indisputably one of the most vulnerable regions worldwide to heat-related mortality.It is worth noting that the anticipated rise in heat-related mortality resulting from climate change in the Southeast Asian region is expected to exhibit an upward trend in the future, as highlighted by the research conducted by Sun et al (2022).Despite the anticipated rise in temperature and its corresponding extremes throughout the remainder of the century, our research findings suggest that the heatrelated mortality in the Southeast Asian region will occur during the mid-century period spanning from 2040 to 2080 that correspond to the risen of heat index at the same time.This result indicated that temperature is not only a factor controlling heat-related illness, but humidity also plays a crucial role in heat-related illness.As described in Amnuaylojaroen et al 2022 the heat index derived from wet-bulb temperatures exhibit significantly greater magnitudes of trends compared to those derived from dry-bulb temperatures within the Indochina Peninsula.This observation underscores the pivotal role of humidity in shaping the climatic conditions of this region.Moreover, the elucidation of this phenomenon can be predicated upon the pivotal influence exerted by temperature and humidity on the physiological manifestations of heat-related health conditions.According to Sherwood and Huber (2010) and Gagge et al (1937), empirical investigations in the field of physiology have yielded findings that suggest a noteworthy association between elevated levels of humidity and temperature, and the potential hazards they may impose on the well-being of individuals.The extant body of literature in the fields of physiology and biophysics has firmly established the significant influence of ambient humidity on the ability to maintain human core temperature within the boundaries of safety.Concomitantly, in tandem with the escalation of ambient humidity, there is a concomitant reduction in the evaporation rate of secreted sweat, resulting in a higher proportion of sweat either remaining on the skin surface or dripping off the body, thereby failing to contribute to the latent cooling process (Candas et al 1979).Due to the reduced dissipation of heat at elevated levels of humidity, the internal temperature of the body progressively escalates until enough perspiration is produced, thereby facilitating the necessary rate of evaporation to achieve thermal equilibrium.The risk of heat stroke significantly increases when the demand for evaporative heat balance surpasses the physiological threshold, leading to an unsustainable sweat rate.Meanwhile, the main cause of heat-related mortality in Southeast Asia is attributed to a combination of topographical, meteorological, and socio-economic variables.The geographical location of Southeast Asia near the equator renders it susceptible to extended and intense periods of elevated temperatures, hence leading to a sustained heightened vulnerability to health problems associated with heat (Haines et al 2006).The vulnerability of the region is further intensified by the tropical environment, as the high levels of humidity enhance the thermal stress experienced by the human body (Watts et al 2018).Additionally, the phenomenon of urban heat islands has arisen because of the swift urbanization occurring in Southeast Asia.This phenomenon is characterized by elevated temperatures within urban areas, surpassing those observed in the adjacent rural regions.The phenomenon of urban heat intensifies the health hazards associated with high temperatures, especially for vulnerable groups residing in informal settlements that have restricted healthcare accessibility (Hertel et al 2010).
As shown in figures 5-8, the spatial distribution of extreme temperature variations in Southeast Asia (SEA), a region known for its dynamic climate and diversified topography, is influenced by a complex interaction of geographical, atmospheric, and oceanic components under different climate scenarios.The Indochina Peninsula, which includes Thailand, Myanmar, Laos, Vietnam, Cambodia, and Malaysia, is expected to experience a significant rise in surface temperature and heat index.This would be most pronounced under the high emission scenarios of SSP3-7.0 and SSP5-8.5, occurring mainly from 2040 to 2100.The Philippines, characterized by the second subregion, exhibits a very modest rise in temperature and heat index fluctuations.Finally, the islands of Indonesia, specifically Kalimantan, Sumatra, and Sulawesi, demonstrate diverse reactions.Temperature projections show a steady and continuous increase, with the heat index changes being most extreme under the SSP5-8.5 scenario.Understanding the various characteristics of land and ocean responses to climate change can explain the varied responses of surface temperature and heat index in various subregions of Southeast Asia under the three different scenarios.According to Sutton et al (2007), land areas warm more rapidly than the oceans due to differences in heat capacity, moisture availability, and changes in surface albedo, among other factors.This can lead to a larger land-sea temperature contrast, especially under high greenhouse gas emissions scenarios Joshi and Gregory (2008) further explain that the land-sea warming ratio is dependent on the nature of external forcings and feedback processes in the climate system.On the Indochina Peninsula, a continental climate with less influence from the ocean makes it more susceptible to higher temperature rises.The Philippines, being an archipelago, experiences a more moderate change due to the ocean's buffering effect.Meanwhile, the diverse topography and extensive forest cover in Indonesia may lead to more complex and localized climate responses, influenced by land-use changes and deforestation.In addition, Manabe et al (1991) explained that when there are amounts of water or wet surfaces, it is probable that a significant portion of the additional energy can be utilized to increase evaporation.This is because evaporation is highly responsive to variations in surface temperature as a result of the Clausius-Clapeyron relationship.The energy budget will be mostly offset by an increased upward latent heat flux.In contrast, on a somewhat arid land surface, there is limited capacity to increase evaporation.As a result, a larger proportion of the additional energy will be utilized to increase the temperature.The energy budget can be balanced by increasing the sensible and longwave heat fluxes, which change less when the surface temperature changes than the latent heat flux.
Our study focused thoroughly on the climatic factors that influence heat-related mortality in Southeast Asia.It became clear that a variety of external circumstances have a substantial impact on these outcomes.However, other factor also effects on heat mortality, for example, socioeconomic aspects were found by Smith et al (2019) as a crucial component that is strongly related to community resilience.According to the findings of Harlan et al (2019), locations with low socioeconomic capacity are more vulnerable to heat-related mortality.This discovery highlights the interdependence of economic development and the ability to withstand and recover from climaterelated impacts.The demographic proportions have added levels of complication to our understanding.According to Bobb et al (2014), locations with a significant population of older people, who are inherently more vulnerable to temperature extremes, reported elevated fatality rates during high-temperature occurrences.The demographic trend emphasizes the importance of tailored measures developed expressly to meet the needs of older persons in the context of severe temperature occurrences.The impact of urban infrastructure on outcomes was similarly significant.The concept of 'heat islands,' which refers to the phenomenon in which densely populated urban areas with minimal vegetation amplify the effects of heat, was especially relevant in the study of Oke (1982).In stark contrast, places that used urban design strategies that included the incorporation of green spaces and reflective surfaces saw a significant reduction in heat-related mortality (Stone et al 2010).The availability and efficacy of medical therapies were also important determinants in the overall result.Heat-related illnesses and mortality were significantly reduced in regions with robust healthcare systems and aggressive heatwave response measures (Kovats & Hajat 2008).By combining various components and adding our core temperature-centric findings, our work provides a comprehensive understanding of heat-related mortality in Southeast Asia.This method enables a more sophisticated understanding of the problem.This statement emphasizes the complex interaction between climatic conditions, socioeconomic situations, population variances, infrastructure challenges, and healthcare capacity.In addition, the region's vulnerability is exacerbated by the low capacity of numerous governments in Southeast Asia to effectively respond to these difficulties.The effectiveness of interventions to heatwaves might be impeded by resource limitations, insufficient healthcare infrastructure, and a dearth of public knowledge (Oka et al 2023).

Conclusion
In summary, the findings of the study present a clear and pressing depiction of the challenges that will confront Southeast Asia in the forthcoming century.The Southeast Asian region is anticipated to have substantial population expansion, which would result in considerable pressures on infrastructure, healthcare systems, and resources.The implementation of effective planning and adaptation techniques is crucial to protect the welfare of the expanding population.One of the most alarming discoveries is the anticipated rise in mortality rates associated with heat, particularly in countries such as Thailand, Cambodia, and Myanmar.The projections exhibit a strong correlation with increasing temperatures and altering heat indices, underscoring the urgent necessity for comprehensive public health interventions and adaptive strategies.In addition, the research highlights the significance of precise climate modeling.Although regional biases are evident, it is crucial to consider these biases when formulating policies and devising strategies for adaptation, notwithstanding the overall satisfactory performance.Moreover, the increasing occurrence of extreme heat events necessitates prompt intervention, encompassing efforts to reduce greenhouse gas emissions and the adoption of measures to protect susceptible communities.The findings might be interpreted as a strong indication of the urgent need for comprehensive, collaborative initiatives across all sectors to mitigate the consequences of climate change in Southeast Asia.The results emphasize the urgency of acting right now, as the health and overall welfare of a significant population in Southeast Asia are at stake within a global context that is progressively characterized by a rising climate.

Figure 1 .
Figure 1.Domain of Southeast Asia with subregion, commonly known as the Indochina Peninsula (a green border), Philippines (a red border), and Malaysia, and Indonesia (a blue border).

Figure 2 .
Figure 2. Spatial distribution of surface temperature from (a) Ensemble Mean, (b) uncertainty of Ensemble Mean, (c) MERRA2, (d) different between Ensemble Mean and MERRA2, and (e) probability distribution function between Ensemble Mean and MERRA2 during 1990-2019 over Southeast Asia.

(
figure 4(c)).According to the SSP2-4.5 scenario, the region is projected to experience a temperature increase of up to 1.5 °C by the middle of the century (figure 4(e)).By the end of the century, this temperature rise is estimated to reach about 2 °C (figure4(f)).The high-emission scenarios, SSP3-7.0 and SSP5-8.5, predict a more significant increase in temperature.According to figure 4(h), the SSP3-7.0scenario predicts a maximum temperature increase of 2 °C by the middle of the century and up to 3.5 °C by 2100.On the other hand, figure 4(j) indicates that the SSP5-8.5 scenario expects a warming of up to 2.5 °C in the near term, which would escalate to 3.5 °C by the middle of the century, and ultimately result in over 5 °C of warming by the end of the century, as shown in figure4(l).The spatial distributions of monthly-averaged Heat Index (HI) for the baseline period spanning from 1990 to 2019 are juxtaposed with those of projected HI values for three distinct climate change scenarios, namely SSP1-2.6,SSP3-7.0, and SSP5-8.5, as illustrated in figure 5.In general, the spatial distribution illustrates a distinct and consistent upward trend in HI throughout the region.The most notable increases are projected to occur in the latter half of the 21st century, specifically between 2080 and 2100, across all scenarios.Between 2020 and 2039, there are anticipated slight increases in HI under the SSP1-2.6 scenario (figure5(a)).Between 2040 and 2079, the level of HI accelerates significantly, particularly in scenarios SSP3-7.0 (figure 5(h)) and SSP5-8.5 (figure 5(k)), where the escalations are widespread and more apparent.The projections for the end of the century (2080-2100) indicate significant increases, with HI values potentially increasing by up to 10 °C under the SSP5-8.5 scenario (figure 5(l)

Figure 8 .
Figure 8. Annual aggregate estimations of mortality risk associated with heat-related incidents within each country, as derived from the comprehensive CMIP6 SSP pathways, spanning the temporal domain from 1990 to 2100.Over (a) Thailand, (b) Indonesia, (c) Cambodia, (d) Viet nam, (e) Myanmar, (f) Laos, (g) Malaysia, (h) Philippine, and (i) Southeast Asia.

Table 1 .
The statistics comparing the Ensemble Mean and MERRA2 during 1990-2019 over Southeast Asia.