Characteristics of human thermal stress in South Asia during 1981–2019

Climate change has significantly increased the frequency and intensity of human thermal stress, with relatively more severe impacts than those of pure temperature extremes. Despite its major threats to public health, limited studies have assessed spatiotemporal changes in human thermal stress in densely populated regions, like South Asia (SAS). The present study assessed spatiotemporal changes in human thermal stress characteristics in SAS, based on daily minimum, maximum, and mean Universal Thermal Climate Indices (i.e. UTCImin, UTCImax, and UTCImean) using the newly developed high-spatial-resolution database of the thermal-stress Indices over South and East Asia for the period 1981–2019. This study is the first of its kind to assess spatiotemporal changes in UTCI indices over the whole of SAS. The study also carried out extreme events analysis of the UTCI indices and explored their nexus with El Niño-Southern Oscillation (ENSO) index. Results revealed a significant increase in heat stress in SAS, with the highest human thermal stress in western Afghanistan, the Indo-Gangetic Plain, and southeastern, and central parts. The extreme event analysis showed that the study region is likely to observe more frequent and intense heat extremes in the coming decades. The correlation of UTCI indices with ENSO exhibited a robust positive coherence in southeastern and central India, southern Pakistan, and northwestern Afghanistan. The findings of the study are critical in understanding human thermal stress and adopting effective risk reduction strategies against heat extremes in SAS. To better understand the dynamic mechanism of thermal extremes, the study recommends a detailed investigation of the underlying drivers of UTCI variability in SAS.


Introduction
Climate change, characterized by anthropogenic global warming, has increased the intensity, frequency, duration, and spatial extent of thermal extremes across the globe (Vicedo-Cabrera et al 2021, Wang et al 2020, IPCC 2021. According to Li et al (2018), Raymond et al (2020), and Yan et al (2021), the level of human thermal stress or human thermal discomfort is determined by the combination of multiple bio-meteorological factors. Due to their complex nature, the stress caused by thermal extremes has relatively more severe impacts on public health than pure high-temperature events (Coffel et al 2018, Russo et al 2019. Extreme heat accompanied by other meteorological factors affects the human body's function via thermophysiological processes (Mueller et al 2014, Roshan et al 2018, Mishra et al 2020. The asymmetric heat exchange between the human body and the biothermal environment results in human thermal discomfort (Napoli et al 2018, Luo and Lau 2019, Blazejczyk 2021, which has catastrophic impacts on human health, vulnerable demographic groups, and marginal socioeconomic classes (Mora et al 2017, Yan et al 2021. Besides public health, the stress caused by thermal extremes also decreases labor productivity (Dunne et al 2013, Kumar and Mishra 2020). In recent years, the increasing prevalence of human thermal discomfort due to heat extremes has attracted the attention of the scientific community, still the robust assessment and quantification of the thermophysiological effects of the atmospheric environment are the key challenges in bioclimatic research.
More than 100 different indices and procedures have been developed to study and quantify human thermal stress in response to thermophysiological effects (Blazejczyk et al 2012, de Freitas and Grigorieva 2015, Yan et al 2021. The Universal Thermal Climate Index (UTCI) was developed in 1999 by a group of multidisciplinary experts (thermophysiology, occupational medicine, environmental sciences, physics, biometeorology, and climatology), recommended by the International Society of Biometeorology and later by the European Cooperation in Science and Technical Action 730 (Coccolo et al 2016, Napoli et al 2018, Blazejczyk 2021. The UTCI is a state-of-the-art thermal index that determines the physiological comfort of the human body under specific meteorological conditions (Jendritzky et al 2012, Seo andHonjo 2021, Yan et al 2021). The UTCI is defined as the air temperature of the reference condition, causing the same thermal stress as actual conditions under the background of the related factors (de Freitas and Grigorieva 2015, Coccolo et al 2016, Jacobs et al 2019. It takes into account the effects of both meteorological factors (the ambient temperature, relative humidity, solar and thermal radiation, and wind speed) and human factors (adaptive clothing behavior and physical activity level and type) (Blazejczyk et al 2012, Fiala et al 2012, Havenith et al 2012.
Since its introduction, the UTCI has been widely used to thoroughly assess and quantify human thermal stress in different parts of the world. In Europe, it has been applied to analyze the spati- . At the same time, the UTCI has been used to assess the effects of heat stress on mortality and morbidity (Nastos and Matzarakis 2012, Urban and Kyselý 2014, Burkart et al 2016, Błażejczyk et al 2018. Moreover, the UTCI has also been used in Brazil, Canada, the Republic of Korea, Iran, Japan, Russia, and China (Bröde et al 2012b, Zeng and Dong 2015, Park et al 2017, Roshan et al 2018, Seo and Honjo 2021, Vinogradova 2021. Despite its extensive use worldwide, very few studies have applied the UTCI in South Asia (SAS). Recently, Zeng et al (2020) investigated the observed spatiotemporal changes in UTCI over the China-Pakistan Economic Corridor and found a general increasing trend of 0.33 • C/decade during the period 1979-2018. In another study, Jacobs et al (2019) assessed the spatiotemporal pattern of exposure to thermal stress in three major cities in SAS: Delhi (India), Dhaka (Bangladesh), and Faisalabad (Pakistan). Their results revealed the occurrence of extremely high temperatures and heat stress over prolonged periods in these cities. The above studies have assessed thermal stress based on the mean UTCI (UTCI mean ), while did not consider the minimum (UTCI min ) and maximum (UTCI max ) indices and their thermal stress categories, which are very important indicators of thermal discomfort. Moreover, previous studies have targeted small specific geographical parts of SAS and did not consider the whole of SAS.
To overcome these limitations, the present study assessed spatiotemporal changes in UTCI min , UTCI max , and UTCI mean as well as their characteristics over the whole of SAS. The study also explored the relationship of UTCI indices with the El Niño-Southern Oscillation (ENSO) index during the study period. We preferred the UTCI over other thermal indices as it considers the effects of multiple meteorological and anthropogenic factors in quantifying human thermal discomfort. More importantly, the present study is the first of its kind to thoroughly assess the spatiotemporal changes in UTCI indices and their characteristics over the whole of SAS from a newly developed high-resolution database of thermal stress indices. The comprehensive assessment of the UTCI-based thermal stress helps to better understand and quantify the resultant human thermal discomfort in densely populated regions, like SAS. Moreover, the outcomes of such assessment are critical in meaningful planning and risk management strategies for ongoing and future human thermal discomfort in SAS.

Study area, data, and methods
2.1. Study area SAS is located in the southern part of the Asian continent (figure 1), with geographical coordinates of 5 • -40 • N latitudes and 60 • -100 • E longitudes (Ullah et al 2020). The region has an estimated land area of 5134 613 km 2 , with a complex topography having the world's highest mountains 'Karakoram-Hindukush-Himalayan (HKH) ranges' in the north, the Indian Ocean in the south, and drylands in the central parts (You et al 2017, Xu et al 2020. SAS is one of the world's most populous regions, with an estimated population of 1.5 billion (Xu et al 2020. The region is among the top ten vulnerable regions to climate extremes (Eckstein and Kt 2020, IPCC 2021). During the past few decades, SAS has experienced several hot extremes, including the extreme heat in 1991, 2006, 2014, 2015, 2017, and 2022, which caused extensive human and socio-economic losses in the region (Wehner et al 2016, Mazdiyasni et al 2017, Ullah et al 2019b, Saeed et al 2021. The accumulation of heat stress during these extreme events has posed potential risks to the local people in the form of heat strokes, death, casualties, and health diseases (Saleem et al 2017). It is argued that the sharp increase in temperature coupled with fragile socioeconomic conditions and low adaptive capacity could greatly aggravate the vulnerability of people to hot extremes in SAS (Im et al 2017, Kotharkar and Ghosh 2021, Ullah et al 2022).

Datasets
The study used the daily data UTCI min , UTCI max , and UTCI mean indices from the newly developed highspatial-resolution database of the Thermal-stress Indices over South and East Asia (HiTiSEA) for the period 1981-2019 (Yan et al 2021). The HiTiSEA database is computed from multiple key meteorological variables of the European Centre for Medium-Range Weather Forecasts (ECMWFs) ERA5-Land and ERA5 reanalysis, including air temperature, humidity, wind speed, direct solar radiation, and shortwave and longwave radiation fluxes. Compared to the existing spatial-resolution (0.25 • × 0.25 • ) ERA5-HEAT (Human thErmAl comforT) data of the ECMWF (Napoli et al 2021), the HiTiSEA database has new features in terms of higher spatial resolution (0.1 • × 0.1 • ), with a comprehensive validation based on thousands of weather stations over South and East Asia (including bias and root mean square error for each index at each station released as part of the dataset). The HiTiSEA developers stated that the finer spatial resolution coupled with wider applicability to stress conditions make this dataset a valuable resource for health authorities and researchers to study the evolution of the thermal environment and identify high-risk areas to potential heat or cold stress. Policymakers can also use the dataset to assess and estimate the costs of extreme thermal stress on the economy through reduced labor productivity and high energy demand, especially in Bangladesh, India, and Pakistan. More details about the processing and computational procedures of the HiTiSEA data are provided by Yan et al (2021). The study also used the ENSO4.0 index to assess its effects on thermal stress distribution and correlation with the UTCI indices in SAS. The monthly time series of the ENSO4.0 index is obtained from the National Center for Atmospheric Research for the period 1981-2019 (https://psl.noaa. gov/data/climateindices/list/).

Calculation of UTCI
Generally, UTCI is calculated in terms of an equivalent ambient temperature ( • C) and physiological response in reference and actual environmental conditions, respectively. The calculation of physiological response to meteorological inputs is based on thermoregulation and adaptive clothing models, which consider behavioral changes in clothing insulation related to the actual thermal environment. Moreover, the reference environment refers to the given conditions: (a) calm air with a 10 m wind speed of 0.5 m s −1 , (b) mean radiant temperature (MRT) equals the air temperature (T a ), (c) 50% relative humidity (RH) for T a ⩽ 29 • C, (d) and water vapor pressure e = 20 hPa for T a > 29 • C, where an average person walks at the speed of 4 km h −1 and generates a metabolic rate of 135 W m −2 . The mathematical equation of UTCI is as follows (equation (1)): In equation (1), T a is the 2 m air temperature, V a is the 10 m wind speed (m s −1 ), e is the water vapor pressure (hPa), and MRT is the mean radiant temperature ( • C).

UTCI stress classes
According to the thermal physiological response of the human body that corresponds to the comfort standard of the model, the values of the UTCI are generally divided into ten thermal stress categories ( Table 1 presents the selected stress categories and their respective physiological conditions.

Statistical analyses
The study used the nonparametric Sen's slope estimator (SSE) (Sen 1968 (2013) and Sun et al (2019). A return value for a specified probability (P) is the value that is exceeded by an annual extreme with a return period, i.e. T = 1/P. This study estimated the extreme values of UTCI indices for 5-, 10-, 20-, 50-, 75-, 100-, 150-, 200-, 300-, 400-, and 500-years return periods. A nonparametric Kolmogorov-Smirnov (K-S) test was applied with a 95% confidence level to determine the significance probability of return values and their distributions. In addition, the Pearson's correlation test was applied to estimate the relationship between ENSO and UTCIs, while the two-tailed student t-test was used to determine the significant correlation at the 95% significance level. Figure 2 shows the long-term spatiotemporal changes and trends in UTCI min , UTCI max , and UTCI mean over SAS during 1981-2019. The spatial distribution of daily climatological means of UTCI indices exhibits a significant variability in SAS (figures 2(a)-(c)), with the highest magnitude (10 • C-40 • C) in the Indo-Gangetic Plain (IGP), central, southern, and eastern parts, while the lowest magnitude (−25 • C-10 • C) in the northwestern mountainous region. Recently, several studies reported similar patterns of heat conditions in SAS (Joshi et al 2020, Mishra et al 2020). The larger spatial variability of UTCI in SAS can be attributed to its complex topographic and climatic conditions (Ullah et al 2019b). The results of long-term climatology indicate that the observed means of UTCI min , UTCI max , and UTCI mean over SAS were 2.40 • C, 24.99 • C, and 12.61 • C, respectively (figure 2(d)).

Spatiotemporal changes in UTCI indices
In monthly analysis, June had the warmest thermal conditions followed by May and July with observed intensities of 34.43 • C, 34.40 • C, and 32.71 • C, respectively. In contrast, January had the coldest thermal conditions followed by December and February with the lowest intensities of −5.26 • C, −3.42 • C, and −3.16 • C, respectively. The countrywise statistics show that all SAS countries had the coldest thermal conditions in January followed by December (table SM1). However, in terms of hot conditions, Bangladesh and India experienced the warmest thermal conditions in May, while Afghanistan and Bhutan experienced the warmest thermal conditions in July. Interestingly, Nepal and Pakistan followed the regional pattern of hot thermal conditions and experienced the warmest thermal conditions in June (table SM3) The spatial distribution of trends in UTCI indices shows a significant increase in thermal stress during the study period (figures 3(a)-(c)). The highest trend (0.5 • C-0.7 • C/decade) of all UTCI indices can be seen in the northwestern and southwestern mountainous parts of SAS. The sharp increasing trend of thermal stress in these areas can be attributed to elevation-dependent warming (Pepin et al 2015, You et al 2020. The central parts have also experienced an increase in all UTCI indices, ranging from 0.25 • C to 0.45 • C/decade. Overall, the magnitude and spatial extent of the warming tendency are larger in UTCI min , followed by UTCI mean and UTCI max . The asymmetric trend in UTCI indices can be attributed to the sharp increase in daily minimum temperature than those of maximum and mean temperatures in SAS (You et al 2017).
The long-term temporal trends in annual anomalies of UTCI indices showed a linear increase during the study period ( figure 3(d)). The temporal distribution reveals an anomalous increase from 1997 and onward, highlighting the tipping points of climate warming and the occurrence of severe heat extremes in the study region (Roshan et al 2018, Jacobs et al 2019. In terms of monotonic trend, the highest increasing trend was found in Interestingly, the majority of the deaths were reported among old-aged, children, women, and disabled people in the affected cities. During the 2015-heatwave event, the central plains of SAS were struck with >45 • C temperature and high humidity for several consecutive days, which resulted in extremely humid-hot conditions (Wehner et al 2016, Im et al 2017. The extended humid-hot period posed potential risks to the local people in the form of heat strokes and health diseases due to the accumulation of heat stress. The above extremes and their catastrophic impacts on the local population confirm the findings of this study.

Spatiotemporal changes in UTCI stress categories
The spatial trend pattern in the frequency of UTCI stress categories over SAS is shown in figure 4. The overall results indicate a rise in human thermal discomfort in SAS, due to an increase in the number of heat stress days. The highest significant increase in heat stress days can be found in the western parts of Afghanistan, the Pak-Afghan border, IGP, and central India. Among all UTCI max -based stress categories, the observed trend (6-8 d/decade) and spatial extent of very strong heat stress days are the highest. This suggests that a large proportion of the population in SAS likely experience periods of strong thermal discomfort (Im et al 2017, Jacobs et al 2019, Kotharkar and Ghosh 2021). Moreover, the highest trend and spatial extent were observed in moderate stress and strong heat stress days under UTCI min and UTCI mean , respectively. On the other hand, the higher thermal stress categories (i.e. very strong and extreme heat stress days) based on UTCI min and UTCI mean did not show any trend, which can be attributed to their low intensities observed in SAS, which could not exceed their thermal ranges.
The spatial distribution of trends in the intensity of UTCI categories over SAS is presented in figure 5. The overall results indicate an increasing trend in the intensity of human thermal discomfort. The highest increase (0.2 • C-0.8 • C/decade) in thermal stress intensity can be found in western Afghanistan, IGP, southeastern coastal, and central parts of SAS. The increasing intensity of heat stress in the western parts of Afghanistan and IGP can be linked to elevationdependent warming (You et al 2020, Shen et al 2021 since these parts are located in the HKH region, which experiences relatively more warming than those of the low-lying areas in SAS. Similarly, the southeastern parts of SAS have a hot climate and are located in the proximity of the Indian Ocean and the Bay of Bengal, which acts as a major source of moisture and water vapor (Pathak et al 2017a, Ullah et al 2021b. The movement of moisture and water vapor contents from these sources to the land results in a hothumid climate over the southeastern coastal belt of SAS (Wehner et al 2016, Pathak et al 2017a, Mishra et al 2020, which induced significant heat stress in the local population. The persistent occurrence of extreme heat stress poses potential heat-related risks to humans in SAS, particularly to the elderly, children, women, the disabled, laborers, and other marginalized groups. Such health-related risks include heat stroke, sunstroke, dehydration, and excess morbidity and mortality in affected areas due to cardiovascular and respiratory diseases (Napoli et al 2018, Luo and Lau 2019, Blazejczyk 2021. In addition, the rising trend of human thermal discomfort coupled with fragile socioeconomic conditions and limited adaptive capacity could further exaggerate the exposure of the local population to extreme heat stress in SAS (Im et al 2017, Kotharkar andGhosh 2021). Given the UTCI min and UTCI mean indices, some of the higher stress categories did not show any trend, which can be attributed to their low intensities in SAS, which could not exceed their thermal ranges. Figure 6 illustrates the temporal patterns of annual anomalies and linear trends in the frequency and intensity of UTCI stress categories over SAS during the period 1981-2019. In terms of frequency, the number of days with heat stress increased during the study period. The temporal distribution of annual anomalies reveals a sharp increase in the number of   thermal stress days during 1997 and onward, indicating a shift towards a hotter climate for a prolonged period in the study region. In terms of monotonic trend (table 2) In terms of intensity, most of the stress categories exhibited an increasing trend, confirming that the study region has predominantly experienced significant heat stress during the study period. The temporal pattern of annual anomalies shows that the intensity of heat stress categories experienced a sharp rise in the study's late period. A set of the literature revealed that the major parts of South Asia have witnessed several episodes of extreme heat during the last two decades (Rohini et al 2016, Im et al 2017, Panda et al 2017, Ullah et al 2019b, which confirms the findings of the current study. Given the monotonic trend (table 2), the intensity of the slight cold and no thermal stress days has decreased at the rates of −0.03 and −0.06 • C/decade, respectively. Whereas the intensities of moderate heat stress and strong heat stress classes have increased at the rates of 0.01-0.37 and 0.02 • C-0.03 • C/decade, respectively, confirming their evolution into higher stress categories with stronger thermal discomfort. The results show that the estimated frequencies of 5-500 year return periods of all UTCI indices in SAS are gradually decreasing, while their respective intensities are increasing, leading to more frequent and intense thermal extremes, which would have catastrophic impacts on human health. Climate extremes with relatively stronger magnitudes have greater adverse effects over a larger spatial extent than those with low magnitudes extremes (Karl and Knight 1997, Frías et al 2012, Zahid et al 2017, Sun et al 2019. The analyses revealed that the estimated magnitude of UTCI min , UTCI max , and UTCI max in the selected return periods (5-500 year) could be 21.84 • C-25.56 • C, 38.95 • C-44.80 • C, and 29.41 • C-33.78 • C, respectively.

Extreme events analysis of the UTCIs' frequency and intensity
Figures 7(d)-(f) shows the histogram of daily observed UTCI indices in SAS during 1981-2019. The UTCI min and UTCI mean histograms exhibit a positively skewed bimodal distribution, with higher frequencies of 12-14 and 10-13 and maximum observed densities of 22 • C-24 • C and 27 • C-31 • C, respectively (figures 7(d) and 6(f)). The bimodal distribution of UTCI min and UTCI mean indices infer an increase in winter UTCI, exposing more population to mild to moderate heat stress. As the bimodal distribution has two peaks on the right and left sides of the UTCI min and UTCI mean time series, indicating positive shifts in their frequencies and intensities. These positive shifts further confirm the frequent and intense occurrence of heat stress in SAS during the study period. The results can also be confirmed from the spatial distribution of UTCI min and UTCI mean indices, which exhibited the lowest climatological values (−25 • C-10 • C) with maximum spatial extents during the study period. The histogram of UTCI max exhibits a normal pattern with the highest density of 35 • C-38 • C (figure 7(e)), indicating the frequent occurrence of thermal stress days in this range.
From the ECDFs' pattern (figures 7(g)-(i)), it can be seen that the extreme value distribution fits the observed data of all UTCI indices significantly. The UTCI min observed and fitted lines indicate a significant similarity with a moderate overestimation of 5 • C-17 • C and an evident underestimation of 18 • C-24 • C (figure 7(g)). In terms of UTCI max (figure 7(h)), the observed and fitted lines are distributed significantly, indicating a close pattern of both time series. The UTCI mean demonstrates a similar pattern to that of the UTCI min , with overall significant distribution and overestimation at 16 • C-24 • C and underestimation at 25 • C-28 • C (figure 7(i)).  Several studies have reported that high sea surface temperature intensified the evaporation of moisture and water vapor contents from the ocean, which move towards the coastal belt of SAS (Pathak et al 2017a, Ullah et al 2021b. The movement of moisture-rich winds results in warming surface temperature, which evolves into hot-humid conditions in the coastal areas (Trenberth and Fasullo 2012, Wehner et al 2016, Pathak et al 2017b. The Asian monsoon precipitation regional variability could be another driver of the UTCI variability that is directly sensitive to oceanatmosphere interactions (Ullah et al 2021a, Singh et al 2022. Moreover, northern SAS, covering the northwestern parts of Pakistan and India, and most parts of Nepal and Bhutan, exhibited a negative correlation pattern (−0.3 to −0.6) between ENSO and UTCI indices. These parts of SAS are located in the jurisdiction of the HKH region and are influenced by two key global weather systems, i.e. western disturbance and monsoon (Ullah et al 2021b, Abbas et al 2022). The negative correlation between ENSO and UTCI indices in northwestern parts of SAS could be attributed to the atmospheric dynamics of westerly and monsoon weather systems, which have strong effects on the local and regional climates in SAS (Niranjan Kumar et al 2013, Joshi and Kar 2018, Hong et al 2019, Hussain et al 2021. It should be noted that both these weather systems bring a significant amount of precipitation to northwestern parts of SAS (Abbas et al 2022), which could have significant impacts on the distribution of temperature, humidity, cloud cover, solar radiation, and wind at the local scales.

Conclusions
The present study assessed spatiotemporal changes in UTCI indices over SAS during 1981-2019. For the first time, this study comprehensively assessed observed changes in characteristics of human thermal stress over the whole of SAS using a newly developed high-resolution database of thermal stress indices called HiTiSEA. Results of the study revealed a significant increase in human thermal stress, with the highest increasing trend in western Afghanistan, IGP, and southeastern and central parts of SAS. The temporal characteristics of UTCI indices and their stress categories have experienced a relatively dynamic and increasing trend in the late 1990s and onward, indicating more human thermal stress during this period. In terms of the return period, the study region is likely to observe more frequent and intense thermal extremes with significant adverse impacts on human health, particularly the elderly, children, women, disabled, labor, and other marginalized groups. Moreover, the spatial correlation of UTCI indices with ENSO exhibited a dipole pattern, with a positive correlation in southeastern and central India, southern Pakistan, and northwestern Afghanistan, while a negative correlation in northwestern Pakistan and India, and most parts of Nepal and Bhutan. This dipole pattern could be monsoon and westerly driven, as both weather systems are sensitive to the oceanic forcing but ENSO is a major driver of the regional climate warming. Further causal attribution and detection analyses are recommended, which will further explore and complement this pattern. Overall, the findings of the study provide potential implications for policy-makers to design climate change adaptation and mitigation strategies in the region.

Data availability statement
The data that support the findings of this study are openly available at the following URL/DOI: https:// doi.org/10.6084/m9.figshare.c.5196296.