Asian megacity heat stress under future climate scenarios: impact of air-conditioning feedback

Future heat stress under six future global warming (ΔTGW) scenarios (IPCC RCP8.5) in an Asian megacity (Osaka) is estimated using a regional climate model with an urban canopy and air-conditioning (AC). An urban heat ‘stress’ island is projected in all six scenarios (ΔTGW = +0.5 to +3.0 °C in 0.5 °C steps). Under ΔTGW = +3.0 °C conditions, people outdoors experience ‘extreme’ heat stress, which could result in dangerously high increases in human body core temperature. AC-induced feedback increases heat stress roughly linearly as ΔTGW increases, reaching 0.6 °C (or 12% of the heat stress increase). As this increase is similar to current possible heat island mitigation techniques, this feedback needs to be considered in urban climate projections, especially where AC use is large.


Introduction
In 2018, Japan had the second hottest July on record (since 1883, Japan Meteorological Agency (JMA) official home page: https://www.data.jma.go.jp), with a mean monthly temperature in Osaka 1.63°C higher than the 11 year (July 2000-2010) mean. These elevated temperatures resulted in the highest on record hospitalisations (54,220) and heat stroke deaths (133) (Ministry of Internal Affairs and Communications, Japan 2018). This period was designated a 'heat wave natural disaster' (Nikkei 2018), similar to disasters from typhoons, heavy rainfall and snowfall, and floods.
Heat waves are expected to become more common and more intense with greenhouse gas (GHG)-induced global warming (e.g. IPCC 2013), exacerbated in cities by the urban heat island effect (e.g. IPCC 2014). With cities being home to more than 66% of the population by 2050 (United Nations 2014), the impact of urban climate on public health and energy supply/demand is critical. Already 30% of the world's population are exposed to deadly heat thresholds on at least 20 days per year, and this may increase to ∼74% by 2100 if GHG emissions increase (Mora et al 2017).
To prepare for future heat waves, it is critical to understand how urban heat stress will change and to identify potential feedbacks from GHG-induced global warming and human activities. Although future urban air temperatures have been explored both globally and locally (e.g. Adachi (Fiala et al 2010(Fiala et al , 2012. Heat stress also depends on micro-scale variations in urban morphology (e.g. shading) and differences in individuals (e.g. age, size, movement, activity). Hence, local-scale grid globe temperatures do not capture micro-scale variability or range of values from shading, but rather the mean for the area (section 2.3). However, grid mean heat stress can indicate the most dangerous conditions that outdoor workers will be exposed to, helping risk assessments for human health.
Japan's many megacities have high population densities (e.g. Tokyo and Osaka) where people are exposed to both high temperature and humidity. Hence, there is high risk of both heat stress and heatstroke during heat waves. Additionally, Japanese cities already use air-conditioning (AC) extensively with the associated release of anthropogenic heat (Q F , i.e. Q F, AC ). With warmer temperatures, Q F, AC can increase causing a positive feedback leading to additional urban warming and energy consumption (e.g. Ashie et al 1999, Kikegawa et al 2003, Sailor 2011, Li et al 2014, Kikegawa et al 2014, Salamanca et al 2014, Takane et al 2017, Ginzhurg and Demchenko 2019, Takane et al 2019. In Osaka, this positive feedback is predicted to cause 0.6°C additional warming in early morning August temperatures (based on a four-GCM ensemble for +3.0°C (cf to current) global warming scenario, ∼2070 s). Given this is a similar size to differences or uncertainties within GHG emission scenarios, RCMs, and urban planning scenarios, this feedback need to be considered (Takane et al 2019).
Our objectives are to predict the impacts on heat stress from future climate at 1-km horizontal resolution, considering the feedbacks from Q F, AC . We focus on Osaka, the second largest city in Japan (figure 1), as it has experienced the hottest mean summer temperatures in Japan in the past 30 years (Takane et al 2013). Osaka's humid climate results in greater daytime urban heat island intensities than cities with drier climates (Zhao et al 2014). Moreover Osaka, already a major tourist destination, will host the 2025 World Expo, thus thermal stress is of concern to both local citizens and global visitors.

Methods
In this study we indicate differences between the current and future climate as Δ (e.g., ΔT); and with (→) and without (≠) air-conditioning (AC) feedback (FB) as δ (e.g., δUTCI is the UTCI difference between AC→FB and AC≠FB).
Feedback from AC use (δUTCI AC→FB ) on future urban climates under future global warming scenarios (ΔT GW ) and changes in δUTCI AC→FB related to ΔT GW are estimated. All methods (numerical model, model setup, and climate projections) are as in Takane et al (2019), except for the UTCI and WBGT calculations. The latter are described within the Supplemental Materials.  (Mellor and Yamada 1982, Janjic 1994, 2002; Noah land surface model (Chen and Dudhia 2001); and BEP+BEM urban canopy parameterisation (Martilli et al 2002. At each time step, Q F, AC is calculated from electricity consumption using BEP+BEM for each 1 km grid. Summertime near surface air temperature and AC electricity consumption skill have been assessed for Osaka considering diurnal and spatial variations (Takane et al 2017(Takane et al , 2019.

Model settings
Two model domains (d01 and d02, figure 1(a)) have 126×126 grid points (x, y) at 5-and 1-km resolution, respectively. Vertically, the 35 sigma levels go up to 50 hPa. Land use, land cover (LULC) and topography data are from the Geospatial Information Authority of Japan (GIAJ). In d02, the GIAJ LULC and Osaka geographical information system (GIS) building footprint (polygon) data (figures 1(c), (d)) are used to classify the urban grids into (i) commercial and business (C); and residential with predominantly (ii) concrete fireproof apartments (Rr) or (iii) detached wooden dwellings (Rw). In d01, all urban areas are assumed to be Rw.
Initial and boundary conditions use NCEP-NCAR (National Centers for Environmental Prediction-National for Atmospheric Research) reanalysis (Kalnay et al 1996) and merged satellite -in situ global daily sea surface temperature (MGDSST) (Kurihara et al 2006) data. As 11 Augusts are sufficient for climatological impacts and effects to be considered (Takane et al 2017(Takane et al , 2019, the time integration for each year is from 00:00 UTC July 27 to September 1, with model spin-up. The 2000-2010 period is treated as the control simulation (case AC→FB) (figure 2, red arrow).
The no-Q F, AC (feedback) simulation (case AC≠FB) differs from the control simulation as Q F, AC is assumed to be 0 W m −2 (figure 2, blue arrow); i.e. the larger difference in UTCI between AC≠FB and AC→FB is the Q F, AC feedback effect (δUTCI AC→FB ). Additionally, six future climates are simulated (section 2.2). We estimate δUTCI AC→FB from ΔUTCI AC→FB − ΔUTCI AC≠FB (figure 2), with δUTCI AC→FB for the current climate being 0°C as we assume no long-term climate change (decades) (i.e., no increase in forcing temperature, and ΔUTCI AC→FB and ΔUTCI AC≠FB are 0°C). To determine δUTCI AC→FB , we assume that all conditions (e.g. urban structures and human activities) remain constant except for background climate change. Although unrealistic, this allows the specific impact of interest to be investigated. The climate variables (i.e. wind components, geopotential height, and temperature) differences between the current and future scenarios are estimated (figure 3). For each ΔT GW case, the climate difference for each variable is added to the NCEP-NCAR and MGDSST data (figure 3) but with the relative humidity kept the same as the current climate. Advantages of this regional climate projection (so-called pseudo-global warming (PGW)) method

UTCI calculation
The hourly UTCI is calculated for 11 years for each climate scenario using the Fiala et al where C g is the sensible heat flux from the globe surface (W m −2 ), R g is the longwave radiation emitted from the globe surface averaged for the surface area (W m −2 ), and ε g and ε h are the emissivities of the globe thermometer (assumed to be 1.0) and human clothing (0.98), respectively. C g is a function of globe temperature and air   temperature (Yuge 1960): However, most regional scale heat stress studies use mean radiative conditions (as we do) they allow the regional scale distribution of heat stress or the heat 'stress' island (section 3.1) to be identified, and its change with climate change to be assessed. Regional scale values provide useful initial and/or boundary conditions for higher resolution building resolving models with street level shade and flows around building and trees.

Verification
The model setup (this section) verificiation is presented in Supplementary material (S1). As the urban charactistics of Osaka ( figure 1(d)) do not produce a large difference between the two types of residential area (wooden detached dwellings and fireproof apartments), we only present the results for the area of wooden detached dwellings (hereafter residential) and the commercial and office buildings (commercial).

Results
The ΔT GW changes the temperature, wind, humidity, and radiation in WRF. In the results, wind speed and T mrt increase a small amount with ΔT GW at night but do not change during the day. Hence, their ΔT GW impact on the UTCI could be small. Relative humidity changes a little from the temperature and specific humidity increases.

UTCI increase (ΔUTCI) with global warming (ΔT GW )
The UTCI is greater in Osaka than in the surrounding land areas at 05:00 under all seven climates (current and six future scenarios, figures 4(a)-(g)), we refer to these as urban heat 'stress' islands. In the current climate, Osaka (white line, figure 4(a)) has moderate heat stress but with greater urban warming (ΔT GW ), this area expands to cover the entire plain when ΔT GW =+1.5°C (figure 4(d)), and extends to the low-mountain area (figure 4(g)) with additional warming. People outdoors in this moderate heat stress area will sweat (sweat rate >100 g h −1 ) and experience wet skin (Bröde et al 2012a). The relatively higher heat stress area is in the coastal parts of Osaka and Kobe (black line, figure 4(g)).
At 12:00, UTCI increases with ΔT GW , and feedback effects of AC are projected (figures 4(h)-(n)), but with inland values expected to be higher than those in the coastal area. Under current climate conditions, the entire area, except the high mountains, experiences very strong heat stress ( figure 4(h)). When ΔT GW =+1.5°C, the mountain area is included in that description ( figure 4(k)). Under such conditions, the human body core temperature of people outdoors for 30 min can increase (Bröde et al 2012a). When ΔT GW =+2.0°C, an extreme heat stress area is projected inland from Osaka, covering Kyoto and Nara (black lines, figure 4(l)). When ΔT GW =+3.0°C, it covers most of the plain ( figure 4(n)). Under these conditions, people will sweat at more than 650 g h −1 , show large increases in their core temperature, and have a lower net heat loss (Bröde et al 2012a).
The changes in the diurnal range of UTCI projected for the current and six future temperature scenarios are similar, but the individual mean values of UTCI differ ( figure 5(a)). In the current climate, there is 1 h with no thermal stress (∼05:00), but this disappears with only a small amount of warming (after ΔT GW =+0.5°C) (yellow, figure 5(b)). The midnight-to-morning period of moderate heat stress remains almost constant with ΔT GW , unlike the evening-to-midnight period, which decreases with ΔT GW from (orange, figure 5(b)). Notably, the latter becomes a strong heat stress (red, figure 5(b)) period once ΔT GW =+2.0°C. Under ΔT GW =+3.0°C, the period is projected to persist until midnight. The very strong heat stress daytime period increases with ΔT GW (dark red, figure 5(b)). Under ΔT GW =+2.5°C, extreme heat stress conditions are expected by 12:00, persisting longer with ΔT GW (black in figure 5(b)).

Impact of AC induced feedback on UTCI (δUTCI AC→FB )
The feedback effects of air-conditioning on UTCI (δUTCI AC→FB ) are much greater at night than during the day in residential areas (figure 6(b)), with changing climate expected to have greater influence in the early morning. The size of this feedback increases roughly linearly with the global temperature increases (figures 6(d), (e)). At 05:00, δUTCI AC→FB increases with ΔT GW (figures 7(a)-(f)) but is smaller in the centre of Osaka (figures 7(b)-(f)). However, at 12:00, δUTCI AC→FB does not change with ΔT GW (figures 6(b), (e)). These differences are probably caused by the difference in mixed layer depth, as Takane et al (2019) proposed. In the middle of the day, Q F, AC is large, but the deeper mixed layer reduces its impact on UTCI. At night, although Q F, AC is smaller, the mixed layer is much smaller. Consequently, Q F, AC enhances the mixed depth, and there is a greater impact on UTCI.
Increased temperature from the nocturnal feedback causes an increase in T mrt which could contribute to an UTCI increase. The contribution of δUTCI AC→FB to ΔUTCI AC→FB (figure 6(c)) is influenced by the δUTCI AC→FB diurnal pattern (figure 6(b)), with the contribution for the night-to-morning period being larger than that in the daytime. The early morning contribution is about 12% when ΔT GW =+3.0°C. These results suggest that one reason for the relatively higher ΔUTCI AC→FB at night (figure 6(a)) is the feedback process. The spatial distribution of the contribution of δUTCI AC→FB to ΔUTCI AC→FB (figures 7(g)-(i)) is similar to that of δUTCI AC→FB (figures 7(a)-(f)). 4. Discussion 4.1. Hot and cold summers: consideration of heat waves Differences in UTCI diurnal pattern are expected in a warmer summer climate. From the 11 current summers, we identify a hot (2010, figure 8(a)) and cold (2003, figure 8(c)) summer to compare to the mean ( figure 8(b)). The hot and cold summer temperatures are 30.5°C and 28.3°C, respectively, or 1.52°C warmer and 0.68°C cooler than the 11-year mean. The August 2010 temperature roughly corresponds to the conditions expected when ΔT GW =+1.5°C (i.e. above the summer mean). These individual summers were selected for each of the future climates for comparison ( figure 8).
The patterns of the hot summer ( figure 8(a)) diurnal UTCI classes when ΔT GW =0.0 to +2.0°C are similar to the mean for ΔT GW =+1.0 to +3.0°C ( figure 8(b), solid blue rectangle). Similarly, the cold summer (figure 8(c)) UTCI patterns for ΔT GW =+0.5 to +3.0°C are similar to the mean for ΔT GW =0.0 to +2.5°C ( figure 8(b), solid green rectangle). Therefore, the hot summer UTCI patterns for ΔT GW =+2.5 and +3.0°C provide some insight into more extreme mean climate (e.g. ΔT GW =+3.5 and +4.0°C, dashed blue rectangle). Similarly, the cold summer UTCI pattern at ΔT GW =0.0°C reflects the impact of an urban heat island mitigation of about 0.5°C using current techniques for the current climate (ΔT GW =0.0°C, dashed green rectangle). Comparing these, the need to respond to or modify the future UTCI pattern caused by global warming and urban heat island mitigation techniques can be considered, in addition to the inter-annual summer variability within ΔT GW .
This approach provides a rough estimate of the future climate UTCI for specific heat and cold waves using past hot and cold summers for comparison.

Heat stress metrics
Two heat-related physiological responses, sweat production and human body core temperature, increase nonlinearly once UTCI exceeds 40°C (very strong and extreme heat stresses), whereas human thermal sensation does not (Bröde et al 2012a). In Osaka, daytime UTCI is projected to exceed 40°C during current and future climates ( figure 5(b), table S2). The impact of the feedback on core temperature is estimated to be less than 0.05°C (not shown) and is regarded as not significant in terms of heat stroke vulnerability.
As human thermal sensation does not continue to change with an increase in UTCI, there is the danger that people will not feel the increasing heat stroke vulnerability. The critical UTCI range is 30°C-36°C (moderate to The diurnal variation and spatial patterns of UTCI in Osaka (figures 4-7) are similar to WBGT (Supplementary material), as others have noted (Zare et al 2018). This suggests the widely available WBGT maps can be roughly used to infer probable UTCI spatial patterns.
As the grid average heat stress metrics calculated in this study do not capture the intra-grid variability (e.g. from shade), the values are more applicable to outdoor workers than to individuals who can seek shade outdoors or go indoors to AC areas.

Relative impact of the AC feedback and thermal mitigation to heat stress metrics
The impact of the AC feedback (δUTCI AC→FB ) simulated when ΔT GW =+3.0°C reached 0.6°C for UTCI and 0.4°C for WBGT (Supplementary Material) with 24-h means 0.23 and 0.15°C, respectively. These are of similar size to some proposed thermal mitigation strategies. For example, the estimated decreases in UTCI with Figure 7. Impact of AC use in Osaka on the August monthly mean (11 years) at 05:00 (a-g) δUTCI AC→FB and (h-n) contribution of δUTCI AC→FB to ΔUTCI AC→FB for increases of (a, g) +0.5°C, (b, h) +1.0°C, (c, i) +1.5°C, (d, j) +2.0°C, (e, k) +2.5°C, and (f, l) +3.0°C. For WBGT see supplemental material. different strategies for residential Lyon in summer include 0.2°C-0.4°C from water aspersion and 0.4°C-0.7°C from vegetation (Morille and Musy 2017). Similarly, facade greening (roofs and walls) are estimated to be able decrease the August daytime maximum WBGT by 0.02°C-0.03°C, and the relocation of AC heat release from walls to roofs by 0.03°C-0.06°C for the 23 wards of Tokyo (Ohashi et al 2016). However, our estimated feedbacks would negate the mitigation benefits from these techniques in future climates, especially where AC use is high.

Future work
Our results the impact of AC on future temperatures suggest is of sufficient importance that future work is warranted: (1) Here heat stress metrics are calculated at 1 km scale but more detailed micro-scale variations (e.g. accounting for shadow patterns from building and vegetation such as by SOLWEIG Lindberg et al 2008) would allow human behaviour (e.g. movement) to be considered (e.g. Honjo et al 2018).
(2) Our estimates of the feedback on heat stress metrics may be low as a constant coefficient of performance (COP) is assumed. A variable COP would be more realistic and should be considered in future studies (e.g. (3) Our focus has been on building energy emissions from AC but Q F sources from traffic, cooling towers, nonwork day energy use variation, and electric and gas AC in office areas should all be considered.
(4) Analysis of other regions using the same methods to generalise the feedback impact, as the impacts may depend on climate, building type/materials, AC performance and human behaviours (e.g. how AC is used).
(5) The UTCI heat stress and physiological response is based on Europeans. Other regions and conditions need to be studied: e.g. Asian city residents.

Conclusions
Effects of GHG-induced global warming on heat stress are considered by analysing RCM (with urban canopy and building energy models) dynamically downscaled simulatons for current and six future climate scenarios (global warming: ΔT GW ). For the latter, CMIP5 global climate model (GCM) simulations with the highest IPCC greenhouse gas emissions scenario (RCP 8.5) are used. Two heat stress indices are calculated for Osaka during August, when air conditioning (AC) use (hence energy consumption) is greatest. From this we conclude: (i) Heat stress (e.g. UTCI) increases with ΔT GW and with AC feedback. At night, an urban heat stress island (i.e. higher UTCI in the urban area compared with the surroundings) is simulated in Osaka for the current and six future climates. In the current climate, only 1 h of no thermal stress occurs near 05:00, but this disappears with ΔT GW =+0.5°C and warmer climates. Moderate heat stress extends across the entire Osaka plain when ΔT GW =+1.5°C. People outside under these conditions begin to sweat, and their skin wetness increases.
(ii) Daytime UTCI tends to be greater inland than in coastal areas. An extreme heat stress area appears when ΔT GW =+2.0°C inland, affecting Kyoto and Nara. This extends over most of the plain when ΔT GW =+3.0°C. These are dangerous conditions for people outdoors, as they may experience large increases in sweating and human body core temperature, and lose the ability to shed heat unless they seek opportunities to reduce heat stress (e.g. shade outdoors, AC indoors).
(iii) The impact of AC-induced feedback on UTCI increases (δUTCI AC→FB ) roughly linearly with ΔT GW . At ΔT GW =+3.0°C, this reaches 0.6°C (12% of UTCI increase). This size is comparable to the suggested benefits of thermal mitigation techniques reported in the literature. Hence, the feedback is significant and could potentially cancel other mitigation benefits in the future, especially where AC use is large. This feedback must not be neglected in future urban climate projections.
(iv) UTCI and WBGT, two independent heat stress metrics, have similar diurnal variation and spatial patterns. As the latter is the official Japanese metric, it may be possible to roughly estimate diurnal variations in UTCI from existing maps of WBGT.