South Asia is a home to more than 1.5 billion people and is under rapid economic growth with an expected population of 2 billion by the mid-21st century (Jones & O'Neill, 2016; Supporting Information Table S1). Among various environmental stresses, two prominent threats are heat extremes (Dash & Mamgain, 2011) and air quality degradation (Li et al., 2017), both of which are reported to lead to major public health crises (Azhar et al., 2014;Chowdhury et al., 2018).
Heat extremes adversely impact human health by affecting respiratory and cardiovascular systems and can also be associated with high surface ozone concentrations that have negative impacts on human health (e.g., Meehl et al., 2018). The heat hazard for human health is preferably quantified in ©2020. The Authors. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. humidity-related temperature indices (Kovats & Hajat, 2008), such as wet-bulb temperature (Sherwood & Huber, 2010) or heat index (Anderson et al., 2013). These indices are related to the efficacy of releasing heat from the skin to regulate body temperature. Recent global climate model-based assessments show that the probability of reaching certain critical thresholds (jointly defined using temperature and relative humidity) empirically known to be life-threatening will continue to rise, especially over South Asia (e.g., Mora et al., 2017) because of the lower climate variability and the higher background humidity. The South Asia region is projected to experience more frequent heat extremes with longer duration and enhanced severity in the future (Russo et al., 2017), which is consistent with observed trends during the past few decades (Alexander, 2016;Dash & Mamgain, 2011;Basha et al., 2017;Pai et al., 2004;Khan et al., 2019;Yin & Sun, 2018).
While there have been major efforts to cut air pollution emissions in developing nations, South Asia faces a unique challenge because of ongoing industrialization and urbanization processes. The next few decades will witness a continued increase in air pollution emissions (or only slightly decrease) in certain Shared Socioeconomic Pathways (SSP)/Representative Concentration Pathway (RCP) scenarios (Rao et al., 2017), which is opposite to the projected worldwide reduction including East Asia. Thus, local emissions continue to be the primary driver for air quality issues, while the influence of climate change cannot be ignored as well (Xu & Lamarque, 2018;Wu et al., 2019).
Despite limited case studies on the urban heat island effect worsening air quality (Wilby, 2008) and potential positive feedback to further enhance heat stress (Cao et al., 2016) in megacities, a decade-long continentalscale analysis of the co-occurrence of heatwave and air pollution extremes and their future changes is still missing. Recent examples are analyses of the heatwave and ozone episodes, such as Schnell and Prather (2017) using North American observations and Meehl et al. (2018) using global model output.
Similarly, health risks associated with an elevated occurrence of heatwaves and high-PM weather are well studied, but often separately, highlighting a knowledge gap between understanding physical and chemical extremes. The compounding negative effect, when two types of conditions occur simultaneously, has only been studied at limited spatial scales (Doherty et al., 2009;Jackson et al., 2010;Stafoggia et al., 2008;Willers et al., 2016), including wildfire conditions induced by the 2010 Moscow heatwave. However, in the public health field, the synergistic impacts of two factors have raised great awareness on exacerbating health risks (De Sario et al., 2013;Katsouyanni & Analitis, 2009;Li et al., 2011;Nawrot et al., 2007;Qian et al., 2008;Ren et al., 2006).
This section briefly summarizes data sets used in this study, and detailed discussions are provided in the following sections.
1. WRF-Chem model simulation (Kumar et al., 2018) of 8 years for present-day (1997)(1998)(1999)(2000)(2001)(2002)(2003)(2004) and 9 years for the mid-21st century (2046-2054) under RCP8.5 and RCP6.0 emission scenarios. 2. MERRA2 reanalysis products (Randles et al., 2017;Buchard et al., 2017) are used for surface PM 2.5 , and ERA-Interim products (Dee et al., 2011) are used for deriving the wet-bulb temperature. 3. Ground measurement of daily temperature and relative humidity is from select airports (collected by the India Meteorological Department but downloaded free of charge from Weather Underground database). 4. Ground measurement of PM 2.5 in the late 1990s and early 2000s is compiled by Kumar et al. (2018), which is contributed by many observational studies (Balakrishnaiah et al., 2011;Deshmukh et al., 2013;Latha & Badarinath, 2003;Pillai et al., 2002;Tiwari et al., 2009;Tiwari et al., 2013). 5. Populations for present-day and future decades are based on Jones and O'Neill (2016). The spatially explicit population data set is from Jones and O'Neill with a spatial resolution of 1/8°by 1/8°. Before any data analysis related to population exposure, environmental quantities are regridded into the grid cells of population data using MATLAB function (interp2). SSP data are provided every 10 years between 2000 (base year) and 2100 (projections). For example, data are available for 2010, 2020, 2030, and so on. The Decade 2050 population projection (with 2 billion population in South Asia, Table S1) is based on Shared Socioeconomic Pathway (SSP) 5 scenario (fossil-fueled development for the economy), which is consistent with RCP8.5 emission pathway. Other SSP scenarios compatible with other RCPs are available in Jones and O'Neill but are not used in this study.
This study uses multiyear simulations conducted using a Nested Regional Climate model coupled with Chemistry (NRCM-Chem) that is based on the Weather Research and Forecasting (WRF) model coupled with Chemistry (WRF-Chem, Version 3.6.1) as described in Kumar et al. (2018). The model uses Model for Ozone and Related chemical Tracers, Version 4 (MOZART-4) (Emmons et al., 2010), for gas-phase chemistry and simulates major aerosol species including sulfate, nitrate, ammonium, organic carbon, black carbon, dust, and sea salt using the Model of Simulating Aerosol Interactions and Chemistry (Zaveri et al., 2008).
The model domain covers the entire South Asia and surrounding oceanic regions (1.5-44.7°N and 52.6-107.4°E) using two domains with coarser horizontal grid spacing (60 km) for the outer domain and finer horizontal grid spacing (12 km) for the smaller inner domain that encompasses the Indo-Gangetic Plain and the Himalayan region. All grid cells have the same area in this configuration. The simulation within the high-resolution inner domain only covers dry seasons (October to May) of each year. The model includes 51 vertical layers up to 10 hPa.
The Model of Simulating Aerosol Interactions and Chemistry includes a sophisticated aerosol thermodynamics module to simulate the effects of changes in temperature and humidity on gas-particle partitioning and on the intraparticle solid-liquid phase equilibrium. Meteorology and chemistry are fully coupled in NRCM-Chem and feedback to each other at every time step. Aerosols affect the meteorology by interacting with both the radiation and clouds, and the corresponding changes in meteorology (temperature, pressure, winds, solar radiation, planetary boundary layer height, and precipitation) affect trace gases and aerosols via feedback to atmospheric chemical kinetics, dry and wet deposition, transport, biogenic emissions, and boundary layer mixing. Fire emissions and land use types were kept constant between the present-day and future simulations to limit the number of drivers contributing to future changes in air quality.
The historical simulation ("Decade 2000") is from 1997 to 2004, and the future simulation ("Decade 2050") is from 2046 to 2054. The simulation is driven by large-scale meteorological and chemical boundary conditions from a global climate model that has been bias corrected against past ERA-Interim (Bruyère et al., 2014). The evaluation of present-day climate and air quality also shows reasonable agreement (see evaluations in the supporting information), and identified meteorological bias was further corrected before our analysis (see supporting information for details). In a nutshell, we subtracted a geographically varying climatological bias as a function of time of the year (historical simulation against ERA-Interim) from both the historical and future simulations.
Due to high-resolution and complex chemical scheme, Kumar et al. (2018) only performed three sets of decade-long time-slice simulations, as opposed to a continuous century-long transient simulation such as in Xu and Lamarque (2018). Note that the decade-long time span of our simulation is still considerably longer than the most previous simulation with fine-resolution chemistry-climate models that usually lasted for weeks to months. The multiyear simulation with hourly output (averaged in this study to daily mean) of meteorology and chemistry is crucial to capture the behavior of extreme events (heatwave and high PM) and estimate future changes in their frequency.
The Decade 2000 simulation is driven by large-scale meteorological boundary conditions generated by Community Earth System Model Version 1 (CESM1; Hurrell et al., 2013), which is bias corrected towards the reanalysis data (European Reanalysis [ERA-Interim]) (Dee et al., 2011). The bias correction procedure is detailed in Bruyère et al. (2014). The chemical initial and boundary conditions are provided by a global atmospheric chemistry model (Community Atmospheric Model Version 4 with Chemistry, CAM4-Chem) (Lamarque et al., 2012), driven by the same meteorological fields from CESM1. Thus, the meteorological boundary conditions used for WRF-Chem are consistent with the chemical boundary conditions in these runs.
10.1029/2019AV000103
XU ET AL.
The emission data set was taken from the Atmospheric Chemistry and Climate Model Intercomparison Project (ACCMIP) (Lamarque et al., 2013). The simulated PM 2.5 surface concentration for the "Decade 2000" was evaluated against seven observational sites in South Asia, and five out of seven sites have a climatologically monthly mean bias of less than 10%. Note that in this paper we use "PM" as a broader term to refer those health-adverse fine particles (PM 2.5 , particulate matter with a diameter less than 2.5 μm) while excluding the contribution of larger particles (>2.5 μm) that could also be important for surface visibility.
The Decade 2050 simulation is driven by CESM1 output under two future emission scenarios: RCP8.5 (CO 2 equivalent of 630 ppm in 2050) and RCP6.0 (505 ppm at 2050). The two scenarios considered are the two higher ones in the RCP database, and the global CO 2 emission is tracking RCP8.5 closely (as of 2018, Figure 1), justifying the focus on the two higher emission scenarios as opposed to the two lower ones. The PM 2.5 emission in South Asia stays largely the same under RCP6.0, compared to the historical period. But for RCP8.5, a 77% increase in total emission from the historical period level is projected. All four RCP scenarios could not be run because of limited computational and storage resources.
We note that the current global emission of CO 2 is tracking RCP8.5 closely (as of 2018, see Figure 1). The satellite-based SO 2 emission estimate (Li et al., 2017) is even higher than the RCP8.5 projection and more in line with the recently released CMIP6 emission data set (SSP). These provide a strong justification for focusing on the higher emission scenario such as RCP8.5 as opposed to the lower ones.
Many previous heat extreme analyses only considered temperature alone (e.g., Meehl & Tebaldi, 2004;Xu et al., 2018), but more recent studies have accounted for humidity impact on the heat stress (e.g., Kovats & Hajat, 2008). A recent assessment of heat extremes related mortality suggested that a combination of temperature and humidity is a better metric to quantify health risks (Mora et al., 2017). That is, under high humidity conditions, human exposure to a lower temperature can induce the same level of risk compared to higher temperature exposure but under lower humidity conditions. Here, we account for both temperature and humidity variations by computing the wet-bulb temperature (T w ; Sherwood & Huber, 2010). T w should not be confused with the wet-bulb globe temperature that additionally accounts for the effect of wind speed and solar radiation (or the simplified form by assuming moderate radiation and light wind speed as in Willett & Sherwood, 2012;Knutson & Ploshay, 2016).
In practice, wet-bulb temperature (T w ) can be measured by wet-bulb thermometers as the environment saturation ratio of water vapor is reached. Here T w is computed following Stull (2011) from the daily average of T (temperature, "dry bulb"; unit: °C) and RH (relative humidity; unitless, ranging from 0% to 100%).
Depending on the data availability, RH is calculated in the following two ways.
1. From the WRF-Chem model output, RH is calculated from T (temperature; unit: K), p (air pressure; unit: Pa), and q (specific humidity; unitless).
2. From the ERA-Interim data set, RH is calculated from T, p, and T dew (dew point temperature; unit: K).
p-e s p-e dew 100%
In the equations above, e 0 (611 Pa) is the reference water vapor pressure, and e s and e dew are the water vapor pressure at saturation and at dew point temperature, respectively. w and w s are water vapor mixing ratio (water vapor vs. dry air, unitless) at any given temperature or at saturation. T 0 (273 K) is the reference temperature. L v (2.5 × 10 6 J/kg) is the latent heat of water vaporization (from liquid to gas). R a (287 J/kg/K) is the specific gas constant for dry air. R w (461.5 J/kg/K) is the specific gas constant for water vapor. Calculation of daily T w has little differences from the mean-taking method, that is, from the average of hourly T w or from mean daily T and RH (Figure S17).
Here we adopt daily average T w at 25 °C as the threshold for heat extremes in this analysis. This is close to the "deadly" threshold (red line in Figure 2) as reported in Mora et al. (2017) who established this threshold based on hundreds of heat-related deadly events during 1980 and 2014 and recorded daily temperature and humidity (but treated separately, not jointly using T w ).
In the context of weather extremes, the question often arises as to "how extreme" certain thresholds should be. Previous studies have used a higher threshold of 35 °C to identify deadly or even fatal extreme heat (Kang & Eltahir, 2018), which is the physical limit to heat removal from the body. Note that Lelieveld et al. (2014) used daily max temperature of 35 °C as the threshold, which is close to 25 °C T w at 40% RH as in the two India heatwave events we identified (Figure 2), but not the wet-bulb temperature.
10.1029/2019AV000103
XU ET AL.
In our case, using the 25 °C threshold allows more samples to enter the analysis and provides a more robust statistical analysis. However, the results with a higher threshold (e.g., T w of 28 °C, Table 1) would be qualitatively similar, the processes would be the same, and the basic results would not change with the caveat that the higher threshold would, of course, represent more lethal conditions. We also conducted a sensitivity test in Karachi using the threshold of daily maximum T w > 35 °C (Table 1) following Kang and Eltahir (2018), which suggests that RCP8.5 could see a 720% increase in heat extreme frequency.
As acknowledged in Mora et al. (2017), previous assessments on deadly heat events have focused on developed nations in the Northern Hemisphere midlatitudes (Europe and North America). The applicability of the same threshold to tropical and developing nations remains to be tested using large-scale public health data. Here, we justify the robustness of the 25 °C T w threshold with limited case studies over South Asia. The yellow star and purple triangle in Figure 2, both close to T w of 25 °C, correspond to two heat extreme events, which reportedly killed more than 1,300 people (see Figure 2 caption for details).
It has become clear to the climate and health research community that the humidity effect needs to be accounted for in heat extreme health impact studies (Sherwood, 2018). To put T w in the perspective of two other temperature/humidity-related heat metrics, T w during the 2010 Ahmedabad event is 24.8 °C (T = 36 °C, RH = 34%), and this is equivalent to 37.2 °C in "heat index" (using the formula of http://www.wpc.ncep. noaa.gov/html/heatindex.shtml; Matthews et al., 2017, also called "apparent temperature"; Russo et al., 2017;Herring et al., 2016) and 41.6 °C (in "humidex" using the formula of https://memory.psych.mun.ca/tech/js/humidex) (Barnett et al., 2010). Such a high value of heat index is classified as "extreme caution" (https://www.weather.gov/safety/heat-index) by the National Oceanic and Atmospheric Administration, and such a high value of humidex is classified as "great discomfort" by the Canadian meteorologists (https://en.wikipedia.org/wiki/Humidex). Other more complex indices that use factors beyond the relative humidity may be more relevant to health impacts, such as wet-bulb globe temperature (Liang et al., 2011) or Universal Thermal Climate Index (Jendritzky et al., 2012).
Sustained exposure to high PM 2.5 environment (such as 100 μg/m 3 ) is conducive to cardiopulmonary mortality and lung cancer (Turner et al., 2011). The threshold of defining high-PM extremes days is here set to 60 Note. All results are based on the original T w (or T) without bias correction. The threshold of T (daily mean temperature) and T w_max (daily maximum wet-bulb temperature) is deliberately selected so that the Decade 2000 frequency is similar to the frequency when using T w of 25 °C as the threshold (48 days). μg/m 3 of daily mean surface concentration of PM 2.5 following India air quality standard (CPCB, 2009). The 60-μg/m 3 value is larger than the "unhealthy" level of the 25 μg/m 3 recommended by the World Health Organization (2005) and the 55.5-μg/m 3 level of "unhealthy" recommended by the Environmental Protection Agency (2012) of the United States, but it is smaller than the 75-μg/m 3 definition of "severe air pollution" recommended in China (Jin et al., 2016). Sensitivity sensitivities (Figure S7) show the results are not particularly sensitive to the selection of threshold other than the expected absolute value change.
To quantify the occurrence of the heatwave and high-PM extremes, the daily value of T w and surface concentration of PM 2.5 are calculated for all grid points of the model output. Having established a certain threshold (section 4), days with values higher than the threshold are classified as extreme days. The frequency (days/year) and the mean duration (days) of extreme events are calculated for each year, and then, the multiyear average for the Decade 2000 and the Decade 2050 under RCP6.0/8.5 is taken to remove the interannual variability of regional climate. Using a stronger definition of extreme events that requires the duration of any individual events to be at least 2 days (e.g., Xu et al., 2018, and references within), the frequency numbers in Table S2 would be lower (see Table 1), but not significantly, due to the low weather variability and longlasting nature of tropical heat extremes in this region.
In addition to quantities of the number of days of extremes, the severity of extremes is also important. The relative intensity of extreme events is reported here in an anomalous sense, as the difference between quantities averaged within extreme days and the selected threshold. A large relative intensity (°C or μg/m 3 ) indicates a severe departure from the threshold and has been suggested as a predictor for heat stress-related mortality (Rocklov et al., 2012).
We define a fourth metric here, accumulated relative intensity, as the product of frequency (days/year) and relative intensity (°C or μg/m 3 ). The concept of accumulated relative intensity for heat stress is similar to the cooling degree days ((temperature -22 °C) * number of days with the temperature higher than 22 °C) that has been widely used in assessing the demand for air conditioning (Miller et al., 2008;Shi et al., 2016).
For detailed model setups, model evaluation, and wet-bulb temperature calculation, readers are referred to sections 2 to 4 of the supporting information.
With the daily average wet-bulb temperature (T w , as in Stull, 2011) of 25 °C as the threshold, heat extreme frequency is as high as 100-200 days/year over the coastal regions and the Indo-Gangetic Plains during the Decade 2000 (Figure S3), with a prolonged duration of more than 15 days particularly over the foothills of the Himalayas. The high values of T w in the southern coastal regions are due to high humidity and in the Indo-Gangetic Plain due to high temperature (Figure S2). Under the RCP8.5 scenario, the future occurrence of heat extremes is projected to increase from 45 days/year (averaged over the seven countries within South Asia) to 78 days (a 73% increase; Table S2 and Figure 3) and with a mean duration of heat extreme events of over 14 days in cities such as Delhi (Table S5).
The intensification of heat extremes shown above is comparable to previous studies over this region when the uncertainty of regional warming projection is considered. In this study, there is projected regional warming of 1.6 °C from Decade 2000 to Decade 2050 (similar to Chaturvedi et al., 2012;1.4 °C for T w as in Table S2). More informative than the absolute value of change projected by a single model is how much of the enhancement, as shown above, can be mitigated by adopting a (moderately) low carbon emission pathway (e.g., RCP6.0). Our Decade 2050 simulation under RCP6.0 suggests that the increase in frequency and duration in RCP8.5 can be cut by 33% and 66%, respectively, over South Asia (Table S2). The relative magnitude is largely consistent with simulated lower regional warming (1.0 °C increase from now to midcentury in RCP6.0 compared to a 1.6 °C increase in RCP8.5), suggesting the scalability of heat extreme statistics shown here to other low-warming scenarios (e.g., RCP2.6), at least for this region. The fractional increase quantified in previous studies is somewhat different due to various definitions of heat extremes, which are discussed next. The daily averaged T w of 25 °C, at the 88th percentile of climatological T w over South Asia (Table S2), may seem not very "extreme" but indeed corresponds to the level of heat stress experienced in two major multiweek heatwave episodes (May 2010 and May 2015) in India that reportedly led to thousands of deaths (Figure 2). As a sensitivity test, using a weaker threshold of 18 °C of T w (the blue line in Figure 2, when the heat-related causality just started to be reported as in Mora et al., 2017), the Decade 2000 occurrence is more frequent at 168 days/year, and the fractional increase into the future is much weaker (14%, Table 1) than when the 25 °C T w is used as the threshold.
The lower thresholds of 18 and 25 °C T w (following Mora et al., 2017) are established empirically based on numerous multidays to multiweek heatwave events that have led to major casualty to vulnerable groups such as children and elderly. The lower threshold of T w should be clearly distinguished from T w thresholds of 30 to 35 °C adopted in some earlier studies (Kang & Eltahir, 2018;Lemke & Kjellstrom, 2012;Sherwood & Huber, 2010;Van Oldenborgh et al., 2017; Table 1), which refers to a lethal physiologic limit that can cause instantaneous hyperthermia, even to healthy active outdoor workers, within just a few hours of exposure (presumably during the daytime).
When a higher threshold of T w 28 °C is adopted as a sensitivity test, the Decade 2000 frequency is much rarer (2 days/year averaged across South Asia) compared to hundreds of days with heat stress when the lower thresholds of T w were used, and the fractional increase in the future is, understandably, much stronger (>400%, Table 1). This is similar to the case when using a strict definition of heat extremes by requiring individual episodes to be at least two consecutive days (e.g., Xu et al., 2018). The relative future increase in frequency under this stricter requirement will also be larger (71% as opposed to 65%, Table 1).
Earlier studies, if using temperature alone without considering the humidity effect, omit the documented evidence that the human body responds negatively to high humidity conditions (Liu et al., 2014). Using temperature alone would underestimate the future increase of heat extremes. For example, if a threshold of T > 31 °C is selected (intentionally) that leads to a Decade 2000 frequency close to 48 days/year (similar to T w > 25 °C, Table 1), the same model projects a 14-29% increase in frequency versus 38-58% using T w and a 13-20% increase in duration versus 50-83% using T w . The reason for the underestimation is that relative humidity over these tropical regions is projected by the latest global climate models to increase with global warming as well (Figures 6 andS2; see also Dai, 2006;Sherwood & Fu, 2014).
Even if the relative humidity stays the same, there will still be an underprediction of heat stress risks if using T alone, just because of the greater health effect of moisture in a warmer climate (Figure 2). The additional benefit of combining temperature and humidity in heat stress assessments is that the model deficiency in simulating the two (Figure S1; Willett & Sherwood, 2012) tends to offset. Similarly, the model discrepancies in projecting temperature and relative humidity tend to be the opposite (Fischer & Knutti, 2013).
A potential underestimation of future increase in heat stress is also likely in previous studies (e.g., Im et al., 2017) if using daily maximum temperature instead of daily mean temperature (unless the specific health and economic concern are lost labor hours and occupational mortality of outdoor workers). There is only a 25-60% increase in frequency when a T w_max threshold of 26 °C is used (with the intention that a similar Decade 2000 frequency is found, Table 1), in contrast to the daily mean T w used in this study (with a 38-58% increase in frequency). The future increase in health risk, when using T w_max instead of daily averaged T w as here, can be underestimated because (a) cooler nights can provide a relief period for the human body to rest and recover (Obradovich et al., 2017) and (b) nighttime temperatures tend to increase faster than daytime temperatures under global warming (Davy et al., 2017).
Are the simulations here (close to 10 years in each case) long enough to provide a robust projection of regional climate change? One may question that a single realization of 8 to 9 years might not be sufficient because a single decade of simulation can be heavily influenced by the phase of decadal variability mode such as AMO. We argue that our results are robust for the following two reasons: First, our simulation for Decade 2000 is highly constrained by observed meteorology (using ERA-Interim as the benchmark for bias correction) and thus represents the real meteorology as observed during those 8 years. Second, our Decade 2050 simulation is driven by boundary conditions provided by multiple runs from a global climate model (CESM1) and thus has effectively accounted for the decadal fluctuation of the climate system. sensitivity compared with other CMIP5 models. For reference, the CMIP5 models yield a mean 2050 warming of 1.6 °C for RCP8.5 (Figure S2, fourth row) and 1.2 °C for RCP6.0 with an uncertainty of a few tenths of a degree (also seen in figure 6 of Chaturvedi et al., 2012).
Due to negative health consequences, it is important to assess human exposure to heat extremes and the reasons for future changes. In the top panels of Figure 4, we show the geographical distribution of population exposure to heat extreme frequency, which has higher values along the populated Indo-Gangetic Plain regions in Decade 2000. We find that population-weighted heat extreme frequency in the Decade 2000 is 83 days/year, larger than the area-weighted estimate (45 days/year), and is projected to increase by 51% to 125 days/year under RCP8.5 (Table S3). The population projection (under the SSP5 scenario; Jones & O'Neill, 2016) is spatially resolved and is consistent with the socioeconomic drivers of RCP emissions. Similar results were also found for the increase in heat extremes duration to 13 days (under RCP8.5) from 9 days in the Decade 2000 (Table S3).
The population-weighted average in Table S3 tends to be larger than the area-weighted results (Table S2) because populations are concentrated in the Indo-Gangetic Plain and coastal regions (Figure S4) where the heat extremes also tend to increase the most (Figure S3). The co-location of extremes and population density is particularly worrisome considering the lower income and GDP over the Indo-Gangetic Plain (Im et al., 2017), which suggests that the most vulnerable population groups will be subject to stronger heat extremes in the future.
Human exposure to heat extremes is dominated by three nations: Bangladesh, India, and Pakistan. Over India, 189.7 billion people-days of heat exposure per year are projected in the Decade 2050 (Table S4), a 149% increase from the Decade 2000. In the bottom panels of Figure 4, we also show another health-related quantity "accumulated relative intensity," which is the product of frequency (number of days) and relative intensity (T w within extreme events minus the selected threshold) (Table S4 and Figure S5). This quantity factors in both the prolonged exposure and the severity of heat extremes. Decade 2050 will see 338 billion people °C days/year (under RCP8.5), a daunting 258% increase from the Decade 2000. The larger relative increase (258% vs. 149%) is consistent with the enhanced severity of heat extremes (with the relative intensity increasing from 1.0 to 1.7 °C) (Table S3 and Figure S3).
The increase in population exposure is due to three factors: future warming, population growth, and, to a lesser extent, population redistribution arising from migration and urbanization. The warming alone explains 41% of the total increase, while the population growth explains about 39% (Table S6). Interestingly, the redistribution of population in India (Figure S4c, while keeping total population fixed) also contributes 1.5% (1.6 billion people-day/year) of the total increase in human exposure to heat extremes (Table S6), which is due to future urbanization and well-captured urban heat island effects in this highresolution regional climate model (Figure S3). We note that the exposure numbers presented here are the maximum potential human exposure (Mishra et al., 2017) that do not account for the time spent indoors with active cooling (which could also change from now to future due to air conditioning penetration into Table 2 The Land Area Fraction Within South Asia That Is Exposed to 60 or More Days of Heat Extremes (Figure S3) and High-PM Extremes (Figure S5) and 60 More Days of Joint Events of Heatwave and High PM (Figure S6) and the Population Fraction household in developing countries; Auffhammer, 2014), which requires estimates of subdaily population distribution in cities.
In addition to the absolute value of human exposure, other important factors worth assessing are the fractions of population and land exposed to the prolonged heat extremes. In the Decade 2000, about 61% of the population within South Asia experienced heat extremes for more than 60 days per year, while in the future, 80% of the population will experience similar extreme heat conditions (Table 2). We estimate the total land fraction impacted by heat extremes for more than 60 days to be 35% in the Decade 2000 (Figure 7). That number will grow to 56% (RCP8.5) or 48% (RCP6.0) in the Decade 2050. Those estimates are robust regardless of whether the model simulated T w is corrected based on reanalysis or not (Table 2).
Air pollution has been recognized as a modulating factor, which can affect the health impact of heat extremes (Gosling et al., 2009). Similar compounding effects have also been found when assessing air quality-related mortality as a function of background temperature (Jackson et al., 2010). We next describe the characteristics of high-PM extremes before discussing the joint occurrence and risk.
XU ET AL.
When using daily average surface PM 2.5 mass concentration of 60 μg/m 3 as the definition of high-PM extremes (CPCB, 2009), we find an increase in the frequency and duration of high-PM extremes by 76% and 125%, respectively, from its Decade 2000 values of 75 ± 9 days/year (frequency) and 4 days (mean duration) (under RCP8.5; Table S2). This is in line with the mean PM 2.5 concentrations increase of 30% driven by an increase in regional PM emissions of 77% in RCP8.5 (Figure 1), while the climate change itself facilitates a stronger removal of PM 2.5 (Wu et al., 2019).
When using other threshold levels suggested by the World Health Organization, Environmental Protection Agency of the United States, or Chinese agencies, the main pattern of high-frequency regions remains the same, but the magnitude of future change would vary (Figure S7). If a lower threshold of air pollution is used, more days (actually most of the days in some cities) will be classified as "high-PM extremes," and its fractional increase into the future will be rather small. We here use a higher threshold of PM 2.5 to illustrate to the "extreme" nature of high-PM issues. Note that we also adopted a similar philosophy in choosing a higher threshold of heat (25 °C in T w ), again, to emphasize the rarity and extremity of those events.
Similar to heat extremes, population-weighted results are considerably higher than area-weighted results for the high-PM extremes. The population-weighted average of high-PM extremes frequency is 118 days/year (Table S3) compared to the area-weighted average of 75 days/year (Figure 7). This is a direct result of the strong co-location of emission sources (Figure 1), PM 2.5 concentrations (Figure S5), and the urban population (Figure S4). The population exposure to high-PM extremes frequency (number of people who experience extremes multiplied with the number of days exposed to the extreme; person * day/year) is projected to increase under RCP6.0 and RCP8.5 scenarios by 154% and 293% (Table S4), respectively. The lower population exposure to high-PM extremes under RCP6.0 is also largely due to lower emission growth. The population exposure to the accumulated relative intensity is 4.2 trillion people * μg/m 3 * day/year in the Decade 2000 and will increase by 293% in the Decade 2050. Note that the larger fractional change in accumulated relative intensity (as the product of frequency and relative intensity) indicates that the severity of high-PM extremes is getting worse (Figure S5). The multifold increase in human exposure is again driven by both population growth and worsening air quality. But in the case of high-PM extremes, the population growth plays a smaller role (19% due to population growth vs. 52% contributed by the hazier atmosphere). This is different from the stronger role of population growth for determining the increase in exposure to heat extremes (43% due to population growth vs. 38% due to warming). The urbanization effect is also more important for high PM (4.7% as opposed to 1.5% for heat shown previously, Table S6).
Since major air quality improvement initiatives have been planned by local governments, we also quantified the high-PM occurrences at the city level. Within South Asia, many cities are subject to a major increase in high-PM extremes but with different levels of severity (Table S5). For example, Mumbai is projected to experience a 34% increase in relative intensity. Cities such as Karachi are prone to the future growth of high-PM weather frequency by 37%, but some other cities appear to already experience ~300 days of high-PM extremes during the Decade 2000. The city-level results are not particularly sensitive to the spatial resolution of the model simulation. When the 12-km resolution simulation over the inner domain is utilized (higher than the 25 km used by Im et al., 2017, but lower than the 4-km grid resolution used by Hu et al., 2015, for the smaller California domain), both the present-day and future PM 2.5 in Delhi remain largely invariant compared to the 60-km simulation. However, the relative intensity for high-PM extremes documented here is slightly higher at 90.9 as opposed to 80.9 μg/m 3 (Table S5b).
Lastly, we quantify the joint heatwave and high-PM hazards (HHH), which has been largely missing in all previous studies. The Decade 2000 frequency for HHH is low at 12 ± 2 days/year for South Asia (Figure 7) and 13 days/year for India (Table S7). In the Decade 2050, the frequency would increase to 33 ± 5 days/year, a 175% rise (under RCP8.5), much higher than the relative increase in heatwave or high PM alone (73% to 76%). A stronger enhancement in HHH is also seen for other extreme quantities such as the mean duration (with a relative increase of 79%) and the relative intensity (with an increase of 0.4 °C and 7.0 μg/m 3 ) (Figures 8 andS6).
The changes in HHH are driven mostly by a larger increase over the spring to summer transitional period, and that results in a greater number of days falling into the high-T w /high-PM quadrant as illustrated in Figure 5 using the data over the four cities. Figure 6 (bottom panels) shows the seasonal variation of temperature, RH, T w , and PM 2.5 . Moist monsoon season is cooler than the pre-monsoon season, but accounting for the humidity effects leads to an extended "hot" season (see T w during April to October in Figure 6). A key feature is the extension of pre-monsoon high PM (pollution season) into the monsoonal season and, concurrently, the extension of heat extremes into pre-monsoon season. These two factors, when simultaneously occurring, contribute to the multifold increase in the frequency of the joint hazards.
The rarity of HHH frequency in the Decade 2000 also means there is a larger relative change in the future for area and population impacted by prolonged HHH events (a factor of 12 increase for exposed land area and a factor of 6.5 increase for the exposed population; Table S2). The multifold increase in the land or population fraction affected by HHH, as opposed to the 31-60% increase in heat-affected and 23-35% high-PM-affected land or population fraction, when computed separately, is the most remarkable message of this study (Table S2 and Figure 7). The multifold increase in land exposed to HHH is illustrated in Figure 7 by the overlapping area of black and red circles and will pose significant difficulties for adaptation.
Given the potential underestimation of HHH health impacts, our results suggest that a major increase in HHH-related mortality is on the horizon. Evidence-based quantification of HHH-related mortality is clearly needed to account for the compounding effects of two types of extremes and also to avoid double counting when linearly adding the mortality estimates from empirical approaches. Although beyond the scope of the paper, one can investigate extreme ozone (e.g., >70 ppb) because many of these regions are very prone to temperature-ozone overlap. Therefore, it will be interesting to assess the occurrence of all three. Our model simulates ozone concentration well (Kumar et al., 2018), even though one limitation of the current WRF-Chem simulations is that it does not include ozone-radiation interactions, which might be not as large as aerosol effects.
However, in general, the monthly mean value rarely exceeds 70 ppb (see Dhaka in figure 4 of Kumar et al., 2018); thus, ozone is less of concern for local air quality as of now. Note that it is possible the NO x to volatile organic compound ratios will change and ozone will be in exceedance in the future, which has started to happen in China. For North America and other regions, the co-occurrence of heat extreme and ozone can also be very important, as recently studied by Schnell and Prather (2017) and Meehl et al. (2018). Thus, the extreme occurrence of all three could be a very interesting question to look at in future studies, for other regions.
Heat extreme occurrence worldwide has increased in the past decades, especially when accounting for the amplification due to the humidity effect and urban heat island influences. At the same time, many cities are facing severe air pollution problems featuring high-PM episodes (high concentration of particulate matter due to various sources) that last from days to weeks. Despite the potential compounding effects on vulnerable population groups and complex dynamical-physical-chemical interactions, the characteristics and potential predictive skills of the co-occurrence of HHH have not been extensively studied.
Although previous studies have suggested common meteorological drivers for these two types of extremes (Schnell & Prather, 2017) and potential amplifying feedbacks (Cao et al., 2016), an integrated assessment of human exposure to the joint occurrence of heatwave and high-PM extremes and possible future changes has been missing (except for a few studies at local scale; Doherty et al., 2009;Jackson et al., 2010).
A regional-scale assessment for the present-day heatwave and high-PM occurrence and future changes is presented here. The most crucial result here is that the frequency of these rare HHH events would increase by 175% in the future, which is in contrast to the 73-76% increase when heatwave or high PM is assessed individually. Consequently, the land fraction affected by prolonged exposure to HHH events will increase by more than tenfold rather than 35% to 60% when the heatwave or high PM are studied separately. The unprecedented worsening of air quality and regional climate, if occurring in just a few decades, poses great challenges to adaptation. If the air pollution emission were not elevated as much as in projected in RCP8.5, then the high-PM extreme will not worsen. For example, under RCP6.0, the frequency of high PM will decrease by 11%, and HHH will only increase by 58%.
Our results suggest that the thermodynamic effect of regional warming leads to the increase in heat extremes and the PM emission increase (as assumed in RCP8.5) is the first-order factor leading to an increase in the high-PM extremes. Other questions remain. How would atmospheric circulation (stagnation) and precipitation play a secondary role? How do the high PM and heat interact with each other (e.g., heat extreme amplifying the high-PM concentration or a high-PM layer mitigating the intensity of urban heat island which seems to be suggested by Figures 8c)? Those will need to be addressed in future studies because of the limitation of the current model setup. The main purpose of this study is to bring forth a greater awareness of the potential larger increase in the coincidence of two stressors.
Our results have broad implications, both scientifically and societally. The quantification, projection, and communication of joint risks of the co-occurrence of physical and chemical weather extremes are important for public health and urban planning. The mechanisms have been examined before for individual cases, but the findings are often scattered among different research communities with limited integration. A holistic view of the health impacts of the HHH is therefore urgently needed.
. All data are for South Asia (Decade 2000 and Decade 2050 under RCP6.0 and RCP8.5). Error bars are the standard deviation showing inter-annual variability. More statistics can be found in TableS2 and FigureS9.10.1029/2019AV000103
XU ET AL.