Heat health during future summers in eastern Asia: The combined roles of circulation, temperature and humidity

Excessive relative humidity (RH) in combination with high temperature can lead to heat stress, often measured by the Wet Bulb Globe Temperature (WBGT). The Clausius‐Clapeyron (CC) relationship implies that warming reduces RH if no extra moisture is added. Over coastal regions like eastern Asia, however, the predominant summer monsoon favours increased moisture transport from surrounding oceans as a result of enhanced evaporation driven by surface temperature increase. This would lessen the RH reduction by potentially two‐thirds. Based on two ensembles of climate model simulations, this paper examines the competing effects of near‐surface temperature, humidity and circulation patterns in this region and the consequential health risks. Under a high emissions scenario (RCP8.5/SSP5‐8.5), surface temperature could increase by 4 – 7°C with WBGT increases of several degrees by the end of the 21st century. Devastating extreme heat health events could therefore become a frequent occurrence as a result. Overall, our results show how humidity can be just as important as temperature when considering the risks to society of excessive heat.


| INTRODUCTION
Excessive heat is a well-known danger to most forms of life, including to humans (Ebi et al., 2021;Hajat & Kosatky, 2010;Koppe et al., 2004) and livestock (Lara & Rostagno, 2013;Renaudeau et al., 2012;St-Pierre et al., 2003). Heatwaves in Europe and Russia in 2003 and 2010 respectively, for example, each resulted in several thousands of human fatalities (Conti et al., 2005;Fouillet et al., 2006;Pirard et al., 2005;Porfiriev, 2014). Extreme heat events have also been noted in eastern Asia (Ding et al., 2010;Tan et al., 2004;Wang et al., 2018;Wang & Gaffen, 2001) with further increases expected in coming decades (Sun et al., 2014) as a consequence of background global scale warming. However, while warming is also expected over Europe and Russia (IPCC, 2014), an additional factor, relative humidity (RH), could result in conditions in eastern Asia becoming even more dangerous. This is due to the warmest season each year, in the region, typically also being the wettest thanks to moisture advected from nearby seas and oceans. Excessive values of RH exacerbate the danger of heat by compromising the efficiency of the cooling process that humans and many animals rely upon to regulate their core body temperature (see Ebi et al., 2021;Sherwood & Huber, 2010;Steadman, 1979;Wang et al., 2021).
Understanding how RH may change in eastern Asia is therefore an essential exercise in understanding and projecting how the danger from excessive heat may change in the region in the future.
RH however is, itself, strongly related to the temperature since it is highly dependent on the saturation vapour pressure, which increases by 7% per K of warming by virtue of the Clausius-Clapeyron equation (Clapeyron, 1834;Clausius, 1850). Overall, most models project little change in globally averaged, vertically integrated relative humidity due to increases in evaporation over oceans being sufficiently large enough to make up for the increased water vapour holding capacity of the atmosphere (Held & Soden, 2010). The global averages however, hide substantial variability. Land-sea contrasts in near-surface warming, for example, as argued by Byrne and O'gorman (2016) can result in sharper land-sea contrasts of the water holding capacity of the atmosphere, enhancing land-sea contrasts of RH due to constraints in local evaporation over land and insufficient advection of moisture from the ocean. Observed evidence of the landsea contrast of RH changes in recent decades has already been reported by Willett et al. (2020).
Exploring the land-sea contrasts of temperature and humidity change, close to the Earth's surface in future East Asian summers, within the background of global warming, and their potential impacts on a widely used heat health measure is the key objective of this paper. A second objective is to quantify the role of circulation variability in producing anomalies of the measure. Section 2 will introduce the methods and data used. Results will be given in Section 3, and conclusions in Section 4.

| Climate simulations
To determine how the temperature and relative humidity, 2 m above the Earth's surface (hereafter: 'nearsurface') could change in a future warmer world, we use model simulations from two ensembles of fully coupled ocean-atmosphere models. The first (denoted PPE), is from a collection of 15 perturbed physics variants of the Met Office HadGEM3-GC3.05 model (Yamazaki et al., 2021) driven by observed greenhouse gas concentrations up to 2005 and thereafter following a (Relative Concentration Pathway) RCP8.5 (Moss et al., 2010) scenario to 2100. The simulations are the same as those in Clark, Wu, et al. (2021), with a horizontal resolution, in mid-latitude regions of approximately 60 km with 85 vertical levels. The second ensemble is of simulations from the 12 models given in Table 1, each chosen, for having a resolution similar to that of the PPE, submitted to the CMIP6  archive. Simulations chosen for analysis followed a (Shared Socioeconomic Pathway) SSP5-8.5 (O'Neill et al., 2016) pathway, very similar to that of RCP8.5 but slightly less aggressive. Both are representative of a world of future rapid economic development and convergence.

| Heat health measure
Several quantities are available for quantifying the environment's impact on heat health. The simplest is the wet bulb temperature (WBT), measured by wrapping an analogue, dry bulb thermometer in a wet cloth, to give a combined representation of the temperature and humidity. By combining the WBT with the temperature (BGT) measured from inside a black globe sphere, Yaglou and Minard (1957) yielded a new index: the 'Wet Bulb Globe Temperature' (WBGT), more closely correlated to human heat stress than WBT alone. Their WBGT effectively became a function of the dry-bulb temperature, relative humidity (RH), sunlight and wind speed. Real world estimates of the  (Epstein & Moran, 2006;Fischer & Knutti, 2013;Liu et al., 2018;Willett & Sherwood, 2012).
Here, we use the simplified version (Grundstein and Cooper, 2018) of the WBGT given by Equation (1), where T is the dry-bulb temperature in C and ea (water vapour pressure) and es (saturated water vapour pressure) are given as in Equations (2) and (3). This can be interpreted as the WBGT that would be perceived by someone indoors. When describing the levels of WBGT in Section 3 of this paper, we will use the following WBGT thresholds adopted by the US military, reported by Willett and Sherwood (2012) to describe the heat health risk to fit, healthy adults: 28 C: high, 32 C: very high and 35 C: extreme.
The PPE and CMIP6 simulations, described in Section 2.1, above, are our principal tool for determining the severity of WBGT values in the future. Their constituent simulations however have biases compared to the observed present-day values. To account for these, we add the projected changes of WBGT values to equivalent values from the ECMWF ERA5 reanalysis (Hersbach et al., 2020) which we use as estimates of observations of the present-day. Throughout, our focus is on JJA (June, July and August) days of the 43-year 2057 to 2099 period, compared to the equivalent length ERA5 1979 to 2021 period.

| Circulation clustering
From Section 3.4 onwards, we focus on the role of circulation on future WBGT estimates using eight clusters of daily sea level pressure as proxies of atmospheric circulation close to the surface. These have been compiled by applying k-means clustering to daily ERA5 JJA sea level pressure analyses over eastern Asia as in  but with data from 1979 to 2021 inclusive. As shown in  and Clark, Wu, et al. (2021), the sea level pressure patterns match the surface circulation very closely.
Regarding our use of reanalyses as a representation of observations; although observations are available of temperature, RH and sea level pressure, we use ERA5 for reasons of data completeness and coherency between the three quantities. Its use is also strongly supported by comparisons of its water vapour with observations, by Zhang et al. (2019) and Yao et al. (2020).
F I G U R E 1 HadGEM3 PPE future (2057-2099) timeslice minus recent past  timeslice ensemble mean, mean JJA 2 m (a) temperature and (c) relative humidity (RH). Panel (b) shows the hypothetical change of RH that would have been expected solely from the temperature change (shown in (a)) in the absence of any change in the dewpoint temperature (see Supplementary Section of paper for details). All values were calculated separately for each PPE simulation before calculation of ensemble mean.

| Warming and relative humidity changes
As described in Section 2, temperature and humidity both play large roles in the WBGT. At the same time, RH is related to the temperature. A simple illustration of this is given in Figure 1. The first panels show the projected ensemble mean near-surface mean JJA temperature for 2057 to 2099 relative to 1979 to 2021 from the PPE, showing warming of at least 4 C for everywhere over land and between 6 C and 7 C in northern China, Mongolia and Russia. Figure 1b shows the hypothetical change of the near-surface RH that such a warming would imply, following the Clausius-Clapeyron relation, if there was no change in the amount of moisture present, that is, if the dewpoint remained the same. Decreases of between 16% and 24% would be expected over most land areas. If such large decreases occurred, they would be expected to reduce some of the increased heat health impact of the warming. The projected change, by the PPE of the nearsurface RH however, is given in Figure 1c with much smaller decreases of between 3% and 9% over the mainland. The greatest differences between the hypothetical and projected RH changes (shown in Figure 1d) occur over the sea. Overall, the values illustrate that approximately two-thirds of the RH decrease which would have been solely produced by the warming is potentially reversed. The role of circulation in this will be explored later. Figure 2 gives some examples of WBGT changes implied by the changes shown in Figure 1. If the warming of Figure 1a was to be accompanied by the near-surface RH decrease of Figure 1b, the WBGT increases (see Figure 2a) would be mostly between 2 C and 4 C, depending on the location. If the warming was to be accompanied by a situation in which the near-surface RH remained the same, much greater WBGT increases (see Figure 2b), of between 5 C and 9 C would be expected. These two outcomes are purely hypothetical but they clearly demonstrate the non-negligible impact of RH on the heat health index. Figure 2c gives the WBGT change from the projected temperature and RH changes of Figure 1a,c, with increases of between 5 and 7 C. Figure 2d gives a third hypothetical outcome of the F I G U R E 2 Future (2057-2099) timeslice 2 m wet bulb globe temperature (WBGT, C), relative to recent past  timeslice, from changes in (a) PPE projected temperature (but not relative humidity), (b) PPE projected relative humidity (but not temperature), (c) PPE projected temperature, and hypothetical relative humidity change due to the temperature change in the absence of any dewpoint change. (d) PPE projected temperature and relative humidity. HRH, hypothetical relative humidity; RH, relative humidity; T, temperature. WBGT change if the projected near-surface RH decrease occurred in the absence of any warming. The 0.5-2.0 C negative effect on the WBGT change is of a much smaller magnitude than the warming effects seen in the other panels of Figure 2.

| Recent past and future WBGT and PPE model bias
We now consider near-surface WBGT values in the recent past and future periods in the context of the impact thresholds described in Section 2. From Figure 3a, ERA5 JJA values averaged during the 1979 to 2021 period widely exceeded 28 C in SE China, representing a high risk to heat health. A very high risk is only seen in coastal areas of China's Guangdong province. These are reproduced very well by the PPE ensemble simulations with ensemble mean biases between 0 C and 1 C in the regions of greatest WBGT and up to 2 C over most of the wider eastern Asia region, suggesting that the PPE is a suitable tool for estimating WBGT values for the future 2057 to 2099 period. These are given in Figure 3c by adding the projected PPE mean increases to those from ERA5. The very high health risk threshold of 32 C is exceeded throughout the region. The extreme risk (35 C) is breached in parts of China and almost everywhere in the Western Pacific.

| Effect of circulation on nearsurface temperature and relative humidity
To extract information of a temporal granularity greater than that of the broad climate means shown in Figure 3c, we now examine the dependency of WBGT in the region on the local circulation. This is useful since it highlights excessive WBGT events which are more likely to have F I G U R E 4 (a)-(h) Mean ERA5 2 m temperature ( C) during JJA 1979-2021 days of k-means based clusters of ERA5 sea level pressure defined over eastern Asia, relative to the mean (i) of all JJA 1979-2021 days, regardless of cluster. Green contours show mean pressure anomalies (hPa) of the days in each cluster.
F I G U R E 5 (a)-(h) Mean ERA5 near-surface meridional thermal transport (10 m meridional wind speed * 2 m temperature) during JJA 1979-2021 days of kmeans based clusters of ERA5 sea level pressure defined over eastern Asia, as a fraction of the mean (i) of all JJA 1979-2021 days, regardless of cluster. Green contours show mean pressure anomalies (hPa) of the days in each cluster.
F I G U R E 6 (a)-(h) Mean ERA5 cloud cover during JJA 1979-2021 days of k-means based clusters of ERA5 sea level pressure defined over eastern Asia, as a fraction of the mean (i, also a fraction) of all JJA 1979-2021 days, regardless of cluster.
impacts. It also allows a better understanding of their causes. Here, we firstly examine near-surface temperature and RH dependence on circulation. To do this, we gather together daily fields of ERA5 mean sea level pressure of the region 80 E, 15 N to 130 E, 50 N from 1979 to 2021 JJA days using the k-means clustering technique into 8 clusters. This number of clusters was found in , to be a suitable number for the east Asia region during JJA. The mean sea level pressure anomaly patterns of the clusters are shown by the green contours in Figure 4a-h. As discussed in  and Clark, Wu, et al. (2021), the pressure patterns are strongly representative of the near-surface circulation (shown in Figure S1a), with anomalous clockwise circulation anomalies encircling areas of positive pressure anomalies and the reverse around negative pressure anomalies. The positive pressure anomalies to the east of China in Cluster 1, for example are characterized by anomalous meridional flow from the South China Sea.
Figure 4a-h also shows the mean near-surface temperature, relative to the overall long-term summer average (denoted 'regardless of circulation'), shown in Figure 4i, during JJA days of 1979 to 2021.
The largest temperature anomalies occurred during days of circulation Clusters 2, 4, 5 and 6. To understand the anomalies, though, requires comparisons with both the near-surface meridional thermal transport and cloud cover anomalies shown in Figures 5 and 6. Relationships between all three, however, are complex with the effects of the thermal transport appearing to compete against those of the cloud cover. The warm anomalies of Clusters 2 and 6 appear to be associated with stronger than normal thermal transport (see Figure 5b,f), for example, while those of Clusters 4 and 5 appear related to anomalously negative cloud cover (see Figure 6d,e). In these two clusters (4 and 5), the (lack of) cloud cover appears to even cancel out the effects of the weaker than normal thermal transport seen in Figure 5d,e. The competing roles of cloud and thermal transport are also seen in Clusters 1 and 3, where anomalous cloud cover appears to cancel out stronger than normal thermal transport, with resulting cooler conditions. The cloud cover effects seen here thus align with the findings of Schneider (1972) in which cloud was found to have a greater role on surface heating from incoming solar radiation, than on surface cooling from heat loss back out to space.
The cloud cover itself is strongly dependent on circulation cluster type, with anomalous cloud downstream of large-scale moisture sources. This can be seen, for example, over China, downstream of the South China Sea in Clusters 1 and 3 (Figure 6a,c) and over the Korean peninsula, downstream of the Yellow Sea in Cluster 2 (Figure 6b).
Some interesting results can also be seen where the cloud cover and temperature anomalies were both positive, indicating that thermal advection also has a role in the temperature anomalies. Examples include northern China in Cluster 1 and central to eastern parts of China between 25 N and 30 N in Clusters 2 and 6. Separation of the role of thermal advection on the temperature, from that of the cloud, though, is very difficult throughout the region since both are favoured by southerly winds.
Anomalies (Figure 7a-h) of near-surface relative humidity (RH) are also strongly dependent on circulation with greater than normal RH downstream of moisture sources and similar spatial patterns of anomalies to those of the cloud cover. They are also closely related, but anticorrelated with the temperature anomalies. Where temperatures are greater than normal, the Clausius-Clapeyron relation broadly implies that the near-surface RH is likely to be less than normal. This is seen, for example, in north-west China in circulation Clusters 1, 2 and 6 ( Figure 7a,b,f). There are a few regions, however, where warm anomalies do occur without less than normal humidity, during days of specific circulation clusters. These are mostly coastal with onshore wind anomalies such as in eastern China in Cluster 5 (Figure 7e). Figure 8a-i shows the WBGT anomalies, resulting from the near-surface temperature and RH, in each circulation cluster type, relative to the mean in Figure 8i. Over large parts of the region, circulation Clusters 2, 5 and 6 are clearly the most likely to give heat health impacts, while those of Clusters 1, 3 and 7 afford a degree of respite. The mean values of greatest concern (Figure 8i) are in southern China and Vietnam. In some of the circulation types, actual WBGT values (shown in Figure S2) widely exceed 31 C in the region, almost reaching the 'extreme' heat health risk threshold (32 C) mentioned in Section 2. By subtracting the corresponding (i.e., circulation cluster relevant) near-surface temperature from WBGT values, we can see where the humidity has its greatest influence on the heat health measure. Anomalies of this influence are shown in Figure 9a-h, relative to the mean influence shown in Figure 9i. The regions where the mean influence is greatest (e.g., southern China in Figure 9i) correspond remarkably well with those of the greatest WBGT concern. Along the coast, the near-surface RH effectively enhances the heat health measure by approximately 4 C compared to the air temperature. In northwest China, Kazakhstan and Mongolia, where the air is much drier (Figure 7i), the opposite can be seen (Figure 9i) with an overall cooling effect in the driest locations. The role of the near-surface humidity on the WBGT in this region can also be visually appreciated by comparing the cooling effect in this region in circulation Clusters 1, 2 and 6 ( Figure 9a,b,f) with the warming effect of Clusters 4, 7 and 8 (Figure 9d,g,h), corresponding to the opposing signs of the RH anomalies in these two sets of clusters in Figure 7. Also notable, is the effect of the positive RH anomaly during days of Cluster 1 (Figure 7a) in central and southern China. During these days, the temperature anomaly is, on average, between 0.3 C and 0.6 C cooler than the mean (see Figure 4a). The humidity, however, results in WBGT values between 2 C and 4 C (calculated by adding the values of Figure 4a to those of Figure 4i) warmer than the actual temperature. The humidity is sufficiently large enough to limit the respite that the cooler conditions provide.

| Circulation dependent changes of near-surface temperature, RH and WBGT
We now switch our attention to the future 2057-2099 period (FP, as defined in Section 2) and determine how F I G U R E 9 (a)-(h) Mean ERA5 2 m wet bulb globe temperature minus temperature ( C) during JJA 1979-2021 days of k-means based clusters of ERA5 sea level pressure defined over eastern Asia, relative to the mean (i) of all JJA 1979-2021 days, regardless of cluster. the near-surface temperature, RH and WBGT may change according to the circulation cluster patterns shown in Figures 4-9. To do this, the Euclidean distances between these patterns and the sea level pressure pattern of each JJA day of each PPE simulation were compared to determine the pattern most likely to have occurred on each day of the simulations. This method readily allows comparisons between ERA5 and PPE recent past circulation dependent WBGT values and between the recent past and future periods but does make an assumption that the circulation cluster patterns in the PPE simulations are the same as in ERA5, and that no fundamentally new circulation patterns emerge in the future. Figure 10a-h gives the projected near-surface temperature changes between the recent past 1979 to 2021 period and the FP, relative to the mean (regardless of cluster) changes, shown in Figure 10i. Warming is projected in all circulation clusters, but is greatest during days of circulation Clusters 1, 2 and 3 and smallest in Cluster 4. The corresponding near-surface RH changes, given in Figure 11, highlight how the circulation type may modify the perception of the circulation dependent warming. Although the mean near-surface RH is projected to decrease over all land areas in the region (Figure 11i), changes are expected to vary considerably with circulation type. Onshore, cyclonic flow anomalies, such as in eastern China in circulation Clusters 4 and 5 result in smaller projected decreases in RH, for example (Figure 11d,e). This could be expected to exacerbate the heat health impacts of the warming in this region. Of note, also are the projected positive near-surface RH changes over northern China in the clusters (1, 2 and 3) of the greatest warming. The pressure contours of Cluster 3 suggest that transport of moisture from the South China Sea, on the west side of the anticyclonic anomaly could be responsible for this.
Maps of the circulation influence on the WBGT from the combined effects of the projected near-surface temperature and RH changes are given in Figure 12. For almost the whole region, but especially in northern China, increases are greatest during days of circulation Clusters 1, 2 and 3. During the recent past period, F I G U R E 1 0 (a)-(h) Ensemble mean, mean HadGEM3 PPE future (2057-2099) minus recent past  timeslice 2 m temperature ( C), during JJA days of k-means based clusters defined using daily JJA ERA5 sea level pressure over eastern Asia, relative to the mean (i) change of all JJA days, regardless of cluster. Red (+ve values) and blue (−ve values) areas in panels (a)-(h) thus show locations where the future warming (relative to the recent past) is, respectively, greater and smaller than the mean (regardless of cluster) warming. however, the greatest WBGT anomalies were in Clusters 4, 5 and 6 over central and eastern China, and the greatest absolute values in the mean were over south-east China. To identify the circulation where the actual future period WBGT values are likely to be greatest; however, requires careful removal of the PPE's bias. This, as shown by Figure S3, is also circulation dependent. Because of this, the most straightforward approach is to thus add the projected changes to the ERA5 values, using their corresponding circulation dependent values. The circulation dependent nature of the bias also suggests that this approach is more suitable than that of quantile mapping, commonly used in the literature (Li et al., 2010).

| Estimates of future WBGT
Using the method above, Figure 13a-h gives estimated, bias-corrected, near-surface FP WBGT anomalies for each of the ERA5 circulation cluster types, relative to the mean (Figure 13i) of the future period. These mean values are widely estimated to exceed the extreme (35 C) in the region shown, especially in Southern China (between 20 N and 23 N) and the Yangtze river basin (between 26 N and 31 N). In Southern China, any respite from the extreme conditions provided by circulation type variability is likely to be very limited, with mean circulation dependent anomalies smaller than 1 C. For the Yangtze region, some respite is seen, for example, during circulation Clusters 4 and 7. Even more challenging conditions, though, are indicated in Clusters 2 and 6, characterized by days when the West Pacific Subtropical anticyclone is stronger than normal.
Equivalent projections from the CMIP6 ensemble described in Section 2, relative to those from the PPE, are provided in the supplementary information ( Figure S4). Estimates of future WBGT from CMIP6 are generally 0.5-1.5 C cooler than from the PPE, although this difference is circulation dependent. The smaller warming is likely to be related to the HadGEM3-GC3.1 model, from which the PPE was produced, having a higher effective climate sensitivity to CO 2 increase than typically found F I G U R E 1 1 (a)-(h) Ensemble mean, mean HadGEM3 PPE future (2057-2099) timeslice minus recent past  timeslice 2 m relative humidity (%), during JJA days of kmeans based clusters defined using daily JJA ERA5 sea level pressure over eastern Asia, relative to the mean change (i) on all JJA days, regardless of cluster. Red areas (−ve values) in panels (a)-(h) thus show locations where the future drying (relative to the recent past) in a circulation cluster is greater than the mean drying.
F I G U R E 1 2 (a)-(h) HadGEM3 PPE ensemble mean, mean 2 m wet bulb globe temperature ( C) during JJA 2057-2099 days of k-means based clusters of ERA5 sea level pressure defined over eastern Asia, minus equivalent days of JJA 1979-2021, relative to the mean difference (i) from all JJA days, regardless of cluster.
F I G U R E 1 3 (a)-(h) Ensemble mean, mean projected 2 m wet bulb globe temperature (WBGT, C) during JJA 2057-2099 days of k-mean based clusters of ERA5 sea level pressure defined over eastern Asia, relative to the mean (i) of all projected JJA 2057-2099 days, regardless of cluster. Projected values were produced by adding circulation cluster dependent PPE WBGT changes to corresponding ERA5 WBGT values. The additions were undertaken separately for each PPE simulation. Black box in (i) shows region 105 E,20 N to 125 E,40 N,referenced in Figure 14. in other models (Andrews et al., 2019). The results from the PPE however, do strongly support the WBGT increases reported by Liu et al. (2018) in their study using RCP8.5 simulations from the earlier CMIP5 ensemble.
To put the difference between the PPE and CMIP6 ensembles, of the WBGT estimates for the future period, into context with those of the recent past period from ERA5, Figure 14 shows the percentage of the black box of Figure 13i, where WBGT values are estimated to exceed a variety of thresholds. The black box is positioned to include the greatest human population centres in the region. Furthermore, by ranking the results from the constituent simulations in each ensemble, the dashed lines of Figure 14 also show the uncertainty within the ensembles. Although the percentages of grid-boxes from the CMIP6 ensemble in Figure 14 are less than of the PPE in all circulation cluster types, the envelopes of the two ensembles do overlap and are clearly very distinct from the blue line in each panel, showing results from ERA5. The threshold of 34 C, unreached in ERA5 for example, is attained in at least a third of the domain in 7 of the 8 circulation types according to both the CMIP6 and PPE ensemble means.

| Future estimates of WBGT during extreme events
All circulation dependent results presented so far are of means calculated from all days in each circulation type and are therefore summaries from several hundred days. Although certain circulation types are clearly of greater concern than others, most days from which the means are calculated, will be of conditions unlikely to cause the greatest impacts. To examine the days which are more likely to have an impact, we now examine those days (hereafter, 'warmest WBGT days') of the greatest 10% of WBGT values in each circulation type, that is, the 90th percentile. By choosing the top 10%, instead of, for example, the top 1%, we can obtain estimates which are likely to be reasonably robust but are also of events deserving attention because of their frequent occurrence. Figure 15a-h gives estimates, for each circulation type, of the projected WBGT (using the same methodology to calculate the projections as in Section 3.5) for the warmest WBGT days for the 2057-2099 timeslice, from the PPE mean. In almost all circulation types, there are locations where 10% of the days in each type thus reach F I G U R E 1 4 Percentage of land gridboxes of the region 105 E, 20 N to 125 E, 40 N (shown by the black box in Figure 13i) where average daily mean wet bulb globe temperature (WBGT) reaches the thresholds shown (y axis) during ERA5 1979-2021 (blue) and projected 2057-2099 (black and red) JJA days of each circulation cluster of Figure 4. Projected values were produced by respectively adding circulation cluster dependent PPE (black) and CMIP6 ensemble (red) WBGT changes to corresponding ERA5 JJA 1979-2021 WBGT values. The additions were undertaken separately for each PPE simulation. Solid black and red lines are from the ensemble means of the PPE and CMIP6 projections respectively while corresponding dashed lines (2 from each ensemble) show the 10th and 90th percentile range from across the individual simulations in the two ensembles.
at least 40 C, that is, 5 C beyond the 35 C reported by Willett and Sherwood (2012) as the limit of safety. Interestingly, the region between 30 N and 34 N appears to be at an even greater risk than the region in southern China identified in Figure 3a as being the most prone, climatologically, during the 1979-2021 period. The emergence of this second region of concern is also apparent in Liu et al. (2018) and could be the result of a region influenced by both larger temperature increases to the north and smaller relative humidity reductions to the south, rather like the patterns for the mean JJA day changes shown in Figure 1a,c.
Equivalent plots to those of Figure 14, but for the 90th percentile days, are given in Figure 16. In circulation Cluster 2, 60% of the black box has WBGT temperatures for the 2057-2099 period of at least 36 C. By definition, the spatial coherence of the circulation clusters also suggests a strong possibility of extreme days impacting multiple locations simultaneously, possibly limiting the benefit of population movements to regions of less impact.

| CONCLUSIONS
Due to the Clausius-Clapeyron relation, a warmed nearsurface atmosphere can hold more moisture resulting in a reduced near-surface relative humidity unless there is a change in moisture available, potentially alleviating some of the health impacts of excessive summer heat, from the warming. Model projections of warming in eastern Asia, following the RCP8.5 greenhouse gas concentration scenario, suggest that the near-surface relative humidity could decrease by approximately 20% in such a hypothetical situation. As part of the East Asian monsoon however, the prevailing circulation in the region during summer favours an increase in the import of moisture into the region in future, from nearby sea areas and the Pacific Ocean, which our results show, could cancel twothirds of the potential relative humidity reductions. For this region, the future welfare of people, livestock and wild animals during future summers is thus inextricably linked to future changes in both the temperature and humidity. With model projected dry-bulb temperature F I G U R E 1 5 (a)-(h) Ensemble mean, mean projected 2 m wet bulb globe temperature (WBGT, C) during the hottest 10% of WBGT JJA 2057-2099 days of k-mean based clusters of ERA5 sea level pressure defined over eastern Asia and the mean (i) of the hottest 10% of all projected WBGT JJA 2057-2099 WBGT days, regardless of cluster. Projected values were produced by adding circulation dependent HadGEM3 PPE (2057-2099 minus 1979-2021) changes of the hottest 10% of WBGT days to corresponding ERA5 WBGT values. The additions were undertaken separately for each PPE simulation. Black box in (i) shows region 105 E,20 N to 125 E,40 N,referenced in Figure 16. increases over land in east Asia, of potentially between 4 C and 7 C during the final four decades of the 21st century, compared to the most recent four decades (up to 2021), the resulting heat health impacts during summer in most parts of the region are thus likely to be substantial. Average conditions in almost all circulation types in at least 20% of the most populated part of the region could exceed heat health thresholds deemed dangerous in past research, for example, according to the mean of high resolution models contributing to the CMIP6 ensemble. Projections from the UK Met Office's Had-GEM3 perturbed physics ensemble suggest that at least 40% of the region could exceed the threshold.
During days of circulation types that favour anomalously elevated temperatures and humidity simultaneously in southern China, the Yangtze region, and along the Pacific coast, model projections even suggest a non-negligible risk of conditions becoming devastating. The problem could be even greater if these types were to become more frequent in future. Research is thus required to determine if this could happen.
The current study, however, has only examined projections following the RCP8.5 and SSP5-8.5 radiative forcing pathways. These could be considered as being some of the worst-case scenarios. Future research is envisaged looking at how mitigation may help to reduce the risk.
Further research is also required regarding factors which play a role in heat health assessments, that we have neglected, including, for example, changes in human behaviour during extreme heat events and the effects of wind and direct sunlight.
Assessments of the effects of humidity in other regions of the tropics, where moisture is also readily available during the warmest season should also be made.
AUTHOR CONTRIBUTIONS Robin T. CLARK: Conceptualization; methodology; software; validation; investigation; data curation; formal analysis; visualization; writingoriginal draft; writingreview and editing. Peili WU: Conceptualization; writingoriginal draft; writingreview and editing; resources; supervision; investigation; methodology. F I G U R E 1 6 Percentage of land gridboxes of the region 105 E, 20 N to 125 E, 40 N (black box in Figure 15i) where warmest wet bulb globe temperature (WBGT) days (defined as JJA days greater than then 90th percentile of daily WBGT values) reach the thresholds shown during ERA5 1979-2021 (blue) and 2057-2099 (black and red) on JJA days of each sea level pressure cluster shown in Figure 4. Values of the 2057-2099 period were obtained by firstly adding either the PPE (black lines) or CMIP6 (red lines) ensemble changes (between 1979 to 2021 and 2057 to 2099) to equivalent ERA5 1979-2021 WBGT values. Solid black and red lines are from the ensemble means of the PPE and CMIP6 projections, while corresponding dashed lines (2 from each ensemble) show the 10th and 90th percentile range from across the individual simulations in the two ensembles.