Evaluation and Projection of Population Exposure to Temperature Extremes over the Beijing−Tianjin−Hebei Region Using a High-Resolution Regional Climate Model RegCM4 Ensemble

Global surface air temperature (Tas) in 2011−20 has increased by about 1.1°C relative to the pre-industrial period of 1850−1900, due to human activities and natural variability. Global warming is predicted to continue in the future under intermediate and high emission scenarios of greenhouse gases (IPCC, 2021; WMO, 2023). Anthropogenic activities cause increases in temperature extremes over larger areas and for longer durations (Lee et al., 2023; Wang and Wang, 2023). Temperature extremes present profound risks to human health and agriculture (IPCC, 2021; Liao et al., 2021; Qin, 2022; Batibeniz et al., 2023; da Silva et al., 2023b; Liu et al., 2023; Si et al., 2023; Wang et al., 2023; Xu et al., 2023; Zhou et al., 2023). In 2021, China experienced its highest temperature in March, May, and September since the beginning of the record in 1961 (WMO, 2022). The heatwave that occurred in 2021 in western North America was characterized by a wide spatial area, long duration, and high intensity. It was twice as severe as any other heat wave that has ever occurred over that region (Lo et al., 2023). These temperature extremes are related to surface air temperature anomalies which are mainly caused by air advection, adiabatic warming, and diabatic heating (Röthlisberger and Papritz, 2023). They are also synergistically impacted by sea surface temperature anomalies of the North Atlantic and the Northwest Pacific (Wang et al., 2023).

In the context of a warming climate, temperature extremes are expected to increase in frequency and intensity as well as in duration, thereby increasing the risks to human safety, society, and ecosystems (IPCC, 2021; Liao et al., 2021; Kim et al., 2023; Zhou et al., 2023). Younger people will encounter more extremes during their lifetime, experiencing an estimated 30 extreme heat waves over the lifetime for persons born in 2020, which lies in contrast to 4 heat waves for persons born in 1960 (Thiery et al., 2021). Increasing temperature extremes threaten land vertebrates under future intermediate (SSP245) (Fricko et al., 2017) and high emission scenarios (SSP585) (Murali et al., 2023). Global climate models (GCMs) in the Coupled Model Intercomparison Project phase 6 (CMIP6) (Eyring et al., 2016) under different Shared Socioeconomic Pathways (SSPs), are widely used to investigate historical and future climate extremes (Das et al., 2023; Delgado-Torres et al., 2023) as well as population exposure to climate extremes (Iyakaremye et al., 2021; Wu et al., 2021; Qin, 2022; Sun et al., 2022; Ullah et al., 2022).

Due to their relatively coarse spatial resolution, GCMs cannot represent complex topography and some detailed physical processes. Regional climate models are usually adopted to capture higher spatial characteristics of climate variables as well as climate extremes (Gao et al., 2017; Gao, 2020; Wu et al., 2021; Tang et al., 2022). GCMs can provide initial and boundary conditions to regional climate models (RCMs), which are then run over specific regions. (Di Virgilio et al. (2022) carefully chose CMIP6 GCMs with good computing performance to dynamically downscale RCMs over the Coordinated Regional Downscaling Experiment (CORDEX) regions and provided higher resolution output for regional climate studies (Diez-Sierra et al., 2022). RCMs support higher spatial resolutions at the convection-permitting (CP) scale, i.e., ~3-km spatial resolution (Coppola et al., 2020; Gutowski et al., 2020; Lucas-Picher et al., 2021; Nguyen-Xuan et al., 2022; Takayabu et al., 2022). Thus, RCMs at the CP scale can simulate regional climate factors with higher spatial resolution, which allows us to better understand related mechanisms and predict future regional climate as well as climate extremes (da Silva et al., 2023b).

As one of the most widely used RCMs, RegCM4-7 by the Abdus Salam International Center for Theoretical Physics (Giorgi et al., 2012) is used to evaluate current and future regional climate extremes. It includes an option to choose a hydrostatic or non-hydrostatic dynamic core, at spatial resolutions of more than 10 km or less than 5 km, respectively (Shahi et al., 2021; Nguyen-Xuan et al., 2022; da Silva et al., 2023b; da Silva et al., 2023a). The RegCM4 with a non-hydrostatic core (RegCM4-NH) showed better performance in simulating precipitation extremes at urban scales (Qin et al., 2023b), but its ability to simulate temperature extremes is not well known.

The size of the population encountering climate extremes should also be considered to evaluate the impacts of climate extremes on human lives and economics since climate risks depend on the size of the population exposed as well as climate change (Jones et al., 2018; Gilabert et al., 2021; IPCC, 2021; Liu et al., 2021; Thiery et al., 2021; Tuholske et al., 2021; Ullah et al., 2022). The Beijing−Tianjin−Hebei (JJJ) region, located in northern China, is undergoing rapid urbanization and has seen increasing trends of severe heat extremes during the historical period (Si et al., 2023), exacerbated by the urban heat island effect (Cui et al., 2023). Rapid urbanization in the JJJ region has been occurring since 1978, significantly increasing heat extremes there (Liao et al., 2021). Based on the projections of CMIP6 models, the JJJ region is expected to face more heat-related risks in the middle of the 21^st century (Zhang et al., 2021). Because of the relatively small area of the urban region and the coarse spatial resolution of CMIP6 GCMs, RCMs have been widely used to investigate regional climate at urban or urban agglomeration scales (Langendijk et al., 2021; Nguyen-Xuan et al., 2022; Lemonsu et al., 2023).

In the present study, we investigated historical temperature extremes during 1991−99, and estimated future changes during 2091−99 relative to the historical period, for both China and the JJJ region. Changes in future population exposure to temperature extremes over the JJJ region were also investigated. The RCM results were simulated by RegCM4 with a hydrostatic and non-hydrostatic dynamic core in China and the JJJ region with spatial resolutions of 12 and 3 km, respectively, during the historical period from 1990−99 and the future period 2090−99. The CMIP6 MPI-ESM1.2-HR (Müller et al., 2018) from the Max Planck Institute under the SSP245 scenario was used to force RegCM4 with a hydrostatic core in China and the outputs of RegCM4 were used to force RegCM4-NH over the JJJ region. We have previously used this approach to evaluate the performance of RegCM-NH in simulating precipitation extremes and future changes (Qin et al., 2023b; Qin et al., 2023a). Two land surface components within RegCM4, BATS (Dickinson et al., 1993) and CLM4.5, were adopted to partially reduce the uncertainties due to land processes (Oleson et al., 2013). After evaluating the performance of RegCM4 with different land surface components in simulating mean and extreme temperatures in China as well as the JJJ region, the ensemble of these two results was used to investigate the future changes in temperature extremes as well as exposure of the population to these extremes.

The remainder of this paper is organized as follows. Section 2 describes the specifics of data and model configuration and section 3 presents the main results, including evaluation of RegCM4 in simulating temperature extremes and projection of future population exposure to these extremes. Finally, we gives a discussion and conclusion in section 4.

The observed daily maximum and minimum temperatures (OBS) from more than 2480 meteorological stations by the China Meteorological Administration (CMA) were used to calculate observed temperature extremes (Wu et al., 2022). Figure 1a shows the locations of the stations and the climatic summer temperature. This dataset started in 1951 and has been widely used in climate studies (Zhang et al., 2018; Qin and Shi, 2022).

Figure 1.  Spatial distribution of climatological temperature in summer (JJA) during the historical period (1991–99) by (a) OBS, (b) MPI_ESM1.2-HR, (c) RegCM4 with hydrostatic dynamic core downscaling in China at a 12-km resolution: RCM_BATS_12km, (d) RCM_CLM_12km, and (e) RegCM4 with a non-hydrostatic core run over the JJJ region at a 3-km resolution: RCM_BATS_3km, and (f) RCM_CLM_3km. The four subregions of China are also shown in a–d: Northwest China (NW), North China (NC), South China (SC), and the Tibetan Plateau (TP) based on elevation and climate types.

The global population at a 1-km resolution by Gao. (2017), obtained by downscaling the spatial resolution to 0.125° (Jones and O'Neill, 2016), was used to investigate changes in exposure to temperature extremes (Zhang et al., 2021; Qin et al., 2023a). For this study, population data in 2000 and 2100 were used to represent the size of the population during historical and future periods, respectively. To correspond with the spatial resolution of temperature extremes by RegCM-NH over the JJJ region, the population data was regridded to a 0.025° spatial resolution.

The daily maximum and minimum temperatures from MPI-ESM1.2-HR (Müller et al., 2018) under the SSP245 scenario in the historical (1991−99) and future (2091−99) periods were used to generate the temperature extremes used to compare with the corresponding extremes simulated by RegCM4 and the OBS.

A brief description of the RCM RegCM4.7 with a hydrostatic and non-hydrostatic dynamic core is provided here, additional details can be found in Qin et al. (2023a). The ICTP RegCM4.7 (Giorgi et al., 2012) with a hydrostatic core used to dynamically downscale CMIP6 MPI-ESM1.2-HR under a middle scenario of the SSP245 in China at a 12-km spatial resolution during 1990−99 and 2090−99. The results of the first year were used for model spinup and not analyzed. The RegCM4 with a non-hydrostatic core was then nested in one way by the above RegCM4 run at a 12-km resolution to perform simulations over the JJJ region at a 3-km resolution. Both the BATS and CLM4.5 land surface components were used to run all the simulations in China and the JJJ region to reduce uncertainties due to land features, because RegCM4 with different land features performs differently over different areas. Two new types of land use have been added to BATS in RegCM4 to describe urban and rural regions (Giorgi et al., 2012). Three urban land units were adopted to represent the urban region in CLM, which was based on an urban canyon, such as street width and building height, and a 1-D heat conductive equation was used to calculate the surface fluxes (Oleson et al., 2013). The ensemble of RegCM4 with BATS and CLM4.5 land features was used to investigate temperature extremes as well as exposure to these extremes. All results of RegCM4 with a hydrostatic and an embedded non-hydrostatic core were regridded to a 0.1° and 0.025° spatial resolution in China and the JJJ region, and are referred to as RCM_EN_12km and RCM_EN_3km, respectively.

The temperature extreme indices used in this study were from the Expert Team on Climate Change Detection and Indices (ETCCDI) (Zhang et al., 2011). These indices were used to investigate temperature extremes both on global and regional scales (Iyakaremye et al., 2021; Hu and Sun, 2022; Lagos-Zúñiga et al., 2022; Singh et al., 2022; Imran et al., 2023). The three ETCCDI temperature extremes were 1) the annual maximum values of maximum daily temperature (TXx, units: °C), 2) the number of days with a daily maximum temperature greater than 25°C (SU, units: days), and 3) the number of days of daily minimum temperature less than 0°C (FD, units: days).

First, we evaluated the performance of RegCM4 in simulating mean and seasonal temperatures during the historical period. Figure 1 shows the spatial distributions of summer (JJA) temperature during 1991−99 by OBS, MPI-ESM1.2-HR, RegCM4 with a hydrostatic dynamic core using a 12-km resolution in China with BATS (RCM_BATS_12km) or CLM4.5 (RCM_CLM_12km) land surface component, and RegCM4-NH with a 3-km resolution over the JJJ region (RCM_BATS_3km and RCM_CLM_3km). The MPI-ESM1.2-HR, RCM_BATS_12km, and RCM_CLM_12km better captured spatial patterns of JJA temperature in China compared with observations, showing higher JJA temperatures over the South China (SC) subregion and lower temperatures over the Tibetan Plateau (TP) subregion (Figs. 1a–d), which is consistent with a previous study (Nguyen-Xuan et al., 2022). In general, annual, summer, and winter temperatures by MPI-ESM1.2-HR, RCM_BAS_12km, and RCM_CLM_12km were highly correlated with OBS in China and its four subregions, with most correlation coefficients greater than 0.7 (Figs. 2a–e). Compared with OBS, the JJA temperatures over the TP subregion with high elevation and complex topography by RCM_BATS_12km and RCM_CLM_12km presented large cold biases, consistent with previous studies (Gu et al., 2012; Gao, 2020; Gu et al., 2020). Compared with the summer temperature by MPI-ESM1.2-HR with a coarse resolution of about 1° in China, summer (JJA) temperature in China by RCM_BATS_12km and RCM_CLM_12km showed more spatial details (Figs. 1a–d). The performance of RCM_BATS_12km and RCM_CLM_12km in simulating summer temperature varied over different subregions and RCM_CLM_12km showed large deviations over the TP subregion.

Figure 2.  Taylor diagrams of spatial annual, summer (JJA), and winter (DJF) temperature in China and its four sub-regions (a–e) by MPI-ESM1.2-HR, RCM_BATS_12km, RCM_CLM_12km, and over the JJJ region by two additional simulations: RCM_BATS_3km and RCM_CLM_3km. All simulations were bilinearly interpolated to the observed stations to calculate the correlations and standard deviations.

Compared to MPI-ESM1.2-HR, RCM_BATS_12km and RCM_CLM_12km, and RCM_BATS_3km and RCM_CLM_3km showed similar summer temperatures as well spatial patterns over the JJJ region (Fig. 1). The JJA temperature by RegCM4-NH with a 3-km resolution over the JJJ region was highly correlated with OBS, with correlation coefficients exceeding 0.9 (Fig. 2f). In general, RCMs with a 3-km resolution presented similar spatial patterns of temperature to those by RCMs with a 12-km resolution, which might be related to the relatively homogeneous topography and land-surface conditions of the JJJ region.

To detail the present temperature extremes over the JJJ region, Fig. 3 shows the spatial distributions of temperature extremes during the historical period over the JJJ region, noting that the results for the entirety of China are not shown here. Due to the coarse resolution of MPI-ESM1.2-HR, only a few grids of simulation results were in the JJJ region which might bring large spatial uncertainties. In general, all models could reproduce the spatial patterns of temperature extremes by OBS over the JJJ region, i.e., higher values of heat extremes (TXx and SU) in the southern JJJ region and lower values in the northern JJJ region, with negative patterns of the cold extreme index FD. Figure 4 shows the Taylor diagrams for the three indices of temperature extremes in China and its four subregions by MPI-ESM1.2-HR, RCM_BATS_12km and RCM_CLM_12km, and over the JJJ region by two additional simulations RCM_BATS_3km and RCM_CLM_3km. Compared with MPI-ESM1.2-HR, temperature extremes by RCM_BATS_12km and RCM_CLM_12km had higher correlations with OBS over the NW, NC, and SC subregions (Figs. 4b, d, e) but showed larger systematic biases over some stations (Figs. 3d–f and 4). RCM_BATS_12km and RCM_CLM_12km performed differently for different indices of temperature extremes over different subregions. Thus, the ensemble of RCM_BATS_12km and RCM_CLM_12km in China and the JJJ region, and the ensemble of RCM_BATS_3km and RCM_CLM_3km over the JJJ region were used to partially reduce uncertainties due to their different treatments of the land surface.

Figure 3.  Spatial distribution of temperature extremes during the historical period from 1991 to 1999 by OBS (a–c) , and absolute (TXx) or relative (SU and FD) biases by MPI-ESM1.2-HR (d–f), RCM_BATS_12km and RCM_CLM_12km ensemble mean (g−i), RCM_BATS_3km and RCM_CLM_3km ensemble mean (j–l), respectively. All simulations were bilinearly interpolated to the observed stations to make them comparable.

Figure 4.  Same as in Fig. 2, but for the three temperature extremes: annual maximum daily temperature (TXx), summer days (SU), and frost days (FD).

As mentioned in the previous study (Qin et al., 2023a), most of the population over the JJJ region lived in downtown areas with high population densities of more than 1200 persons per km² (Fig. 5a). The population density in 2100 is projected to decrease over most of the JJJ region by more than 80 persons per km² in the majority of the southern JJJ region, while in downtown areas it is expected to increase by more than 200 persons per km² (Fig. 5b) (Zhang et al., 2021).

Figure 5.  Spatial patterns of population density in 2000 over the Beijing−Tianjin−Hebei (JJJ) region with a 0.025° spatial resolution (a) and their changes in 2100 (b).

Figure 6 shows the spatial patterns of changes in three temperature extreme indices (TXx, SU, and FD) from 2091 to 2099 in China and the JJJ region relative to the historical period from 1991 to 1999 by the MPI-ESM1.2-HR, RCM_BATS_12km and RCM_CLM_12km ensemble mean, RCM_BATS_3km, and RCM_CLM_3km ensemble mean. In the context of a warming world (IPCC, 2021) (Zhang et al., 2023), the temperature extreme indices TXx and SU were projected to increase almost everywhere in China (Sun et al., 2019). The increase in TXx by MPI-ESM1.2-HR during 2091−99 ranges from 2.72°C −3.82°C relative to the historical period over four subregions of China, and those by RCM_BATS_12km and RCM_CLM_12km ranged from 2.28°C −2.85°C, and 2.52°C −3.51°C, respectively (Fig. 7a).

Figure 6.  Spatial patterns of changes in temperature extremes (TXx, SU, and FD) from 2091 to 2099 in the JJJ region relative to the historical period from 1991 to 1999 by the MPI-ESM1.2-HR (a–c), RCM_BATS_12km and RCM_CLM_12km ensemble mean (d–f), and RCM_BATS_3km and RCM_CLM_3km ensemble mean (g–i). The spatial distributions of extremes by MPI-ESM1.2-HR (a–c) were not masked in the JJJ region due to its coarse spatial resolution.

Figure 7.  Violin plots of future changes in temperature extremes in 2091−2099 relative to 1991−1999 in China and the four subregions by MPI-ESM1.2-HR, RCM_BAT_12km, and RCM_CLM_12km, and over the JJJ region by an additional two simulations of RCM_BATS_3km and RCM_CLM_3km. These results were consistent with previous studies in that TXx values over urban regions were projected to increase at the end of the 21st century relative to the historical period (Liao et al., 2021).

Larger increases of TXx in the future were predicted over the southern JJJ region compared to the northern JJJ region by MPI-ESM1.2-HR, RCM_ENS_12km, and RCM_ENS_3km (Fig. 6). MPI-ESM1.2-HR presented a maximum increasing amplitude of TXx by 3.77°C over the JJJ region while TXx values by the four RegCM4 runs increased by 2.32°C−3.33°C. The spatial spreads of TXx by MPI-ESM1.2-HR over the JJJ region were narrower than those by RegCM4 (Fig. 7b). This may result from the coarse spatial resolution of MPI-ESM1.2-HR (Fig. 6a). The results from RegCM4-NH with a 3-km resolution RCM_ENS_3km presented similar changes in TXx as those by RegCM4 with a hydrostatic core at a 12-km resolution RCM_ENS_12km despite more spatial details of temperature extremes given by the RegCM4-NH runs (Figs. 6d, g).

The number of summer days (SU) was also found to increase in 2091–99 relative to 1991−99 in the entirety of China, and its three subregions NW, NC, and SC, except for the TP subregion where no obvious changes in SU were found by MPI-ESM1.2-HR, RCM_BATS_12km, and RCM_CLM_12km (Figs. 6b, e, h, and 7e). The number of SU days was projected to increase by 26−47 days, 19−20 days, and 22−27 days over the above three subregions by MPI-ESM1.2-HR, RCM_BATS_12km, and RCM_CLM_12km, respectively, and 19−35 days over JJJ subregions by two additional RegCM4-NH results (Fig. 7f). These results agree with previous studies that global warming intensified SU over the majority of areas in China (Lu et al., 2019; Zhao et al., 2021). As a cold temperature index, the number of frost days FD was projected to moderately decrease in the future (Sun et al., 2019), by –32 to –13 days in all regions by all the simulations (Figs. 6c, f, i, and 7g, h) consistent with a previous study (Gu et al., 2012).

In summary, in a warming climate, the temperature extremes TXx and SU in China and the JJJ region were projected to increase by the end of the 21st century, while FD generally decreased.

To evaluate the impacts of temperature extremes on human safety and economics under climate change, the total population involved in temperature extremes should also be mentioned, since climate risks depend on climate changes as well as the population exposed to them (Jones et al., 2018; Qin, 2022; Ullah et al., 2022). The population exposed to temperature extremes in urban areas is impacted by the urban heat island effect, climate change, as well as the size of the population subjected to climate extremes (Tuholske et al., 2021). Population exposure is usually defined as climate extremes multiplied by population size over each spatial grid for each simulation (Jones et al., 2018; Qin, 2022; Zhang et al., 2023), i.e., EX = PO × TE, where EX is exposure to temperature extremes, TE is temperature extremes and PO is population size exposed to temperature extremes. Changes in exposure, EX, were expressed as $ \Delta \mathrm{E}\mathrm{X}=\mathrm{P}\mathrm{O}\times \Delta \mathrm{T}\mathrm{E}+\Delta \mathrm{P}\mathrm{O}\times \mathrm{T}\mathrm{E}+\Delta \mathrm{P}\mathrm{O}\times \Delta \mathrm{T}\mathrm{E} $, where $ \Delta \mathrm{P}\mathrm{O} $ represents changes in population size and $ \Delta \mathrm{T}\mathrm{E} $ changes in temperature extremes over each grid cell. Thus, $ \Delta \mathrm{E}\mathrm{X} $ was divided into three components: changes in exposure due to changes in climate extremes $ \mathrm{P}\mathrm{O}\times \Delta \mathrm{T}\mathrm{E} $, population changes $ \Delta \mathrm{P}\mathrm{O}\times \mathrm{T}\mathrm{E} $ as well as the combined effects of changes in extremes and population $ \Delta \mathrm{P}\mathrm{O}\times \Delta \mathrm{T}\mathrm{E} $.

Figures 8–10 show exposure to temperature extremes (TXx, SU, and FD) over the JJJ region by RCM_EN_3km in 1991−99 and their changes in 2091−99, as well as the contributions of the three components of exposure changes. Because temperature extremes present relatively small spreads over the JJJ region, spatial patterns of exposure to temperature extremes in the historical period as well as their changes were generally consistent with the spatial distribution of population density (Figs. 8a, b, and 9a, b, and 10a, b). As a result, future exposure to TXx over the JJJ region obviously decreased by more than 1 × 10⁴ persons °C over many parts of the JJJ region, except in the downtown areas, with increased population, where exposure to TXx moderately increased by more than 1 × 10⁴ persons °C (Fig. 8b). This implies that the downtown areas of the JJJ region might face more climate risks in the future under increases of both temperature extremes and population; relevant policies should be presented to meet these challenges, such as policies that guide people not to excessively gather in the downtown areas, etc. The percentage contributions of exposure changes are defined as the ratios of the absolute value of each component contributing to exposure change to the sum of the absolute values of all the components (Qin et al., 2023a). Changes in future population contributed the most to exposure to TXx with a ratio of 75.23%, compared to 19.26% due to changes in TXx and 5.51% due to combined extreme climate and population changes (Figs. 8c–f). These results were similar to previous results showing that changes in exposure to heat extremes were primarily due to population changes rather than the direct effects of climate change (Tuholske et al., 2021).

Figure 8.  Population exposure to the temperature extreme index TXx (units: 10⁴ persons °C) over the JJJ region by (a) RCM_EN_3km in 1991−99, (b) their changes in 2091−99, and (c) the contributions of the three components: (d) changes in TXx, (e) population, and (f) both TXx and population.

Figure 10.  Same as in Fig. 8, but for FD.

Figure 9.  Same as in Fig. 8, but for SU.

Similar to the heat extreme index, TXx, exposure to the two duration extreme indices SU and FD during 2091−99 was expected to decrease over most of the JJJ region except for some downtown areas (Figs. 9b and 10b), mainly due to decreased population with contributions of 52.85% and 72.54%, respectively (Figs. 9c and 10c). The increase in SU and decrease in FD play minor roles in changes in exposure (Figs. 9 and 10).

In summary, even though population changes greatly contributed to total population exposure to temperature extremes, changes in exposure to temperature extremes were quantified with surface air temperature (Tas) over the JJJ regions due to its direct impacts on temperature extremes. As the Tas over the JJJ region increased by around 2.7°C −3.5°C, changes in exposure to the temperature extremes TXx, SU, and FD were projected to decrease by –0.31, –0.48, and –1.84 × 10⁴ persons per degree warming of the JJJ region, respectively (Fig. 11).

Figure 11.  Joint plots of changes in exposure to the three temperature extremes: TXx, SU, and FD with temperature over the JJJ region in 2091−99 relative to 1991−99 by RCM_EN_3km.

In this study, we evaluated the performance of RCM RegCM4 with a hydrostatic or non-hydrostatic dynamic core in simulating seasonal temperature as well as temperature extremes during 1991−99 with a 12-km resolution in China and a 3-km resolution over the JJJ region. Changes in temperature extremes and population exposure to these extremes during 2091−99 were then projected for China and the JJJ region. Finally, we quantified changes in exposure to these extremes associated with temperature increases over the JJJ region.

In general, the RegCM4 model, with either a hydrostatic or non-hydrostatic core, better simulated spatial distributions of seasonal temperature and temperature extremes than the GCM MPI-ESM1 in China and the JJJ region compared to observations, with a high spatial correlation coefficient of more than 0.8 for most extremes over most subregions and the JJJ region. Because the performance of RegCM4-NH with BATS or CLM4.5 land surface component varied for different extremes over the JJJ region, the ensemble of these two simulations was adopted to investigate the interaction between population extremes and temperature extremes. Under a warming climate, the temperature extremes TXx and SU were found to increase in China and the JJJ region at the end of the 21^st century while FD generally decreased, which is in agreement with previous studies (Sun et al., 2019). Population exposure to TXx, SU, and FD was found to decrease over most areas of the JJJ region, mainly due to population declines, except in downtown areas where population increases are expected at the end of the 21^st century. Finally, we quantified changes in exposure to temperature extremes with Tas increases over the JJJ region.

Although this study investigated temperature extremes over an urban agglomeration with a high-resolution RCM RegCM4-NH at a 3-km spatial resolution, there were still many uncertainties. For example, abrupt changes in land cover and land use due to rapid urbanization will require higher spatial resolution simulations (Cui et al., 2023). How different land surface schemes, land–atmosphere energetic interactions, and water availability will affect climate simulations as well as climate extremes should be deeply and systematically investigated in future work. Although RCMs at convection-permitting scales usually led to more reliable simulations of regional climate with high spatial resolution, it might be better to investigate temperature extremes after the systematic biases of extremes were corrected over each grid cell. Additionally, since heat processes and mechanisms vary over the rural and urban areas (Liao et al., 2021; Shi et al., 2021), the differences in temperature extremes between rural and urban areas of the JJJ region should be systematically investigated, as well as their possible mechanisms. Furthermore, changes in future population sizes also varied over the rural and urban areas, which were projected to decrease over most areas of the JJJ region despite a projected increase over its core areas in this study. Therefore, population exposure to temperature extremes should also be separately investigated in rural and urban areas (Zhang et al., 2023). In addition, the age structure, morbidity, and life expectancy of the population should also be taken into consideration since climate extremes might present different risks among different social cohorts (Thiery et al., 2021). Finally, this study investigated temperature extreme indices separately while compound extremes, such as heat-humidity and heat-drought, known to have even more severe effects on human lives and ecosystems, are worthy of investigation (Seo and Ha, 2022; He et al., 2023).