Predicting Changes in Population Exposure to Precipitation Extremes over Beijing–Tianjin–Hebei Urban Agglomeration with Regional Climate Model RegCM4 on a Convection-Permitting Scale

Abstract

In this study, we have investigated changes in precipitation extremes and the population’s exposure to these extremes during 2091–2099 in China’s Beijing–Tianjin–Hebei (JJJ) region relative to the historical period of 1991–1999. First, the regional climate model RegCM4, with a hydrostatic dynamic core, was run for east Asia, including China, at a 12 km resolution for 1990–1999 and 2090–2099. This model is forced by global climate model (GCM) MPI-ESM1.2-HR under the middle shared socioeconomic pathways (SSP245). The first year was used as a model spinup. Then, the 12 km results were used to force RegCM4 with a non-hydrostatic dynamic core (RegcM4-NH) at a 3 km convection-permitting scale over the JJJ region during the historical and future periods. Future precipitation extremes were predicted to increase over the whole of China and its four subregions, while decreases were predicted over the JJJ region. This may partly be caused by lower increases in specific humidity over the JJJ region. The percentage contributions of the three components of total population exposure, i.e., changes in exposure due to changes in the population, precipitation extremes and the joint impact of the population and extremes, were then analyzed. Changes in the population and wet extremes were closely related to changes in the total exposure over the JJJ region. The population is the dominant factor that most impacts the total exposure to dry extremes. Finally, changes in future population exposure to precipitation extremes per degree of warming were quantified for the JJJ region.

1. Introduction

Warmer climates increase the ability of the atmosphere to hold more water vapor, at a rate of around 6–7% per degree of warming, following the Clausius–Clapeyron (C-C) relationship [1,2]. This is expected to increase the occurrence of precipitation as well as its extremes at a super-C-C or sub-C-C scaling rate [3,4,5,6,7,8,9]. Thus, precipitation extremes are expected to become more frequent and more intense under warmer climates and to have more potential impacts on society, economics and natural ecosystems [3,10,11,12,13].

Compared to temperature, precipitation, as well as precipitation extremes, is difficult to realistically simulate due to its many impact factors and high spatiotemporal heterogeneity [14]. The global climate models (GCMs) used in the fifth and sixth phase of the Coupled Model Intercomparison Project (CMIP5 and CMIP6, respectively) [15,16] have been widely used to study current and future changes in global precipitation extremes [9,11,17]. They have also been adapted on regional scales in Africa [18,19], Asia [14,20,21,22,23], Europe [24], North America [25] and South America [26,27].

It remains challenging to project future precipitation extremes on the regional scale with a high resolution [28,29,30]. GCMs generally have relatively coarse spatial resolutions of around 100–200 km, which is not adequate to characterize the spatial distributions of precipitation extremes and their occurrence rates [28,31,32]. Climate models with high spatial resolutions can better simulate precipitation extremes as well as physical processes related to precipitation extremes, such as large-scale atmospheric rivers [5,33,34,35,36] and tropical cyclones [37,38]. Regional climate models (RCMs) describe dynamic and physical processes in more detail due to their high resolutions and can better reproduce spatiotemporal patterns of precipitation extremes over Asia [4,8,12,39], Europe [13,40], North America and South America [41]. GCMs are generally used to generate initial and boundary conditions (high temporal resolution of three-dimensional atmospheric temperature, specific humidity, atmospheric pressure, horizontal wind, etc.) to force RCM runs with high spatial resolutions for historical and future periods on a regional scale [42,43,44,45]. The regional climate model RegCM4, from the Abdus Salam International Center for Theoretical Physics [46], is widely used in climate studies [4,47,48,49,50,51,52,53]. The spatial resolutions of the majority of RegCM4 simulations are between 25 km and 60 km because the spatial resolution of the hydrostatic dynamic core is restricted to 10 km [54], physical processes are difficult to model and the GCMs used to generate initial and boundary conditions have relatively coarse spatial resolutions.

With increased computational ability, the spatial resolutions of RCMs have improved on the kilometer scale, i.e., the convection-permitting (CP) scale. This allows the characterization of spatial heterogeneities in more detail and explicit simulation of deep convection rather than the use of convection parameterization [55,56,57,58]. RCMs on the CP scale present obvious improvements in climate modeling [59,60,61,62,63] and generate more realistic precipitation extremes on the regional scale, including intensities, frequencies and spatiotemporal variabilities [28,64,65]. Lucas-Picher et al. [66] gave an overview of the latest developments and future directions of convection-permitting modeling with RCMs. Recently, a new version of RegCM-4.7, incorporating a non-hydrostatic dynamic core (RegCM4-NH) at a CP resolution [54], was developed to improve climate simulations [54,67,68,69].

The Beijing–Tianjin–Hebei (hereafter known as JJJ, since “Ji” is the abbreviation of Hebei province; Figure 1) region, a typical urban agglomeration in China, has encountered fast urbanization and rapid population growth in the last decades. This may affect the climate, and populations are at an increased risk of facing more frequent and intense climate extremes [70,71,72]. Besides the intensity and frequency of climate extremes, their effects also depend on how many people are subjected to them [31,73,74,75,76,77]. Increased population exposure to extremes, which is generally defined using extreme indices multiplied by the amount of people subjected to the extremes, is likely to negatively impact human lives [31,74,78,79].

In the present study, we have investigated changes in precipitation extremes and the population exposure to these extremes projected over the whole of China for 2091–2099 and in the JJJ region relative to the historical period of 1991–1999, i.e., the changes in extremes in one century. The rate of exposure to precipitation extremes per degree of warming was also quantified for the JJJ region. First, RegCM4 with a hydrostatic dynamic was run over East Asia at a 12 km resolution for the historical (1990–1999) and future periods (2090–2099) as usual [69], due to the expensive computing costs and large storage needed for regional climate modeling, especially on the CP scale. Initial and boundary conditions were generated with the Max Planck Institute Earth System Model (MPI-ESM1.2-HR), which uses an ECHAM6.3 atmospheric model coupled with an MPIOM1.62 ocean model in CMIP6 under the central shared socioeconomic pathways (SSP245) [80]. This model has been widely used because it results in more realistic climate modeling and is well-suited for prediction and mechanism studies [21,81]. Then, RegCM4-NH at a 3 km resolution was run for JJJ in the historical and future periods, forced by the former 12 km results from RegCM4 with the hydrostatic core. We have evaluated the performance of RegCM4 with the hydrostatic core in China and of RegCM4-NH over the JJJ region. Oleson et al. (2013) have shown that RegCM4-NH could simulate the spatial distribution of annual and summer precipitation in more detail over JJJ and that RegCM4-NH with the CLM4.5 land surface component [82] has shown the best performance in simulating precipitation extremes compared to other experimental setups. Here, we have investigated changes in future precipitation extremes predicted for 2091–2099 over four subregions of China (NW: Northwest China; NC: North China; SC: South China; and TP: Tibetan Plateau) and JJJ urban agglomeration, as well as their possible mechanisms, using RegCM4 with a hydrostatic core or RegCM4-NH (Figure 2). Finally, future changes in exposure to precipitation extremes, such as exposure change per degree of temperature increase, were investigated for JJJ, and the respective contributions of the changes in the extremes and the changes in the population to exposure changes were quantified.

2. Data and Model

2.1. Data

A spatial distribution of the global population at a resolution of 1 km from 2000 to 2100 under the middle SSP245 scenario, which was downscaled by Gao [83] from a population with a 1/8° spatial resolution [84] and has been widely used [31,73,85], was applied here. The population data were regridded to 0.1° × 0.1° (~10 km) in China and to 0.025° × 0.025° (~2.5 km) over JJJ, with distance weight. Because this dataset has begun in 2000 and provides data points for every 10 years, we used the spatial distributions of the population in 2000 and 2100 to represent the historical population during 1991–1999 and future population during 2091–2099, respectively.

The outputs of horizontal wind, the air temperature, specific humidity and the sea surface temperature with MPI-ESM1.2-HR during 1990–1999 and 2090–2099, under the SSP245 scenario, were used to generate the initial and boundary conditions (such as three-dimensional air temperature, water vapor, horizontal wind, two-dimensional surface temperature, etc.) for RegCM4, with hydrostatic and non-hydrostatic dynamic core runs. These outputs were at a temporal resolution of six hours and over 192 × 384 grid points, with a spatial resolution of about 0.9°. The daily precipitation from MPI-ESM1.2-HR was used to generate precipitation extremes to compare with those based on RegCM4.

2.2. Regional Climate Model Description and Experimental Design

The regional climate model RegCM4.7 from ICTP [46] was dynamically downscaled to East Asia, with the whole of China and the JJJ region included (Figure 1). RegCM4.7 can carry out simulations with a hydrostatic or a non-hydrostatic dynamic core [54]. In a similar experimental setup, RegCM4 with a hydrostatic dynamic core was used to dynamically downscale MPI-ESM1.2-HR under the SSP245 scenario in China, with a 12 km resolution of 600 × 450 grids for the historical period of 1990–1999 and the future period of 2090–2099 (RCM_CLM_12km). Due to expensive computing costs and the large storage required for the test and simulation results as well as the forcing data, only the middle SSP245 scenario, which is consistent with historical development rates [80], was adopted, as in a previous study [31], to force RegCM4 to study precipitation extremes. In future work, RegCM4 forced by different scenarios will be used to better investigate future precipitation extremes. The land surface component was adopted as the CLM4.5 scheme [82], which presented more spatial details of this surface. The Grell convection scheme [86] for land and ocean and the WSM5 microphysics scheme [87] were used as cumulus convection and explicit cloud microphysics simulations for RegCM4 with hydrostatic and non-hydrostatic cores, respectively, after careful sensitivity tests, since kilometer grid size is usually considered a gray-zone resolution and convection parametrization will still be needed [68]. The 12 km results of RegCM4 were then used to force RegCM4-NH in a one-nest way over the JJJ region at a 3 km resolution on a 300 × 300 grid (RCM_CLM_12km). The first year of each simulation was used as a model spinup, and only the last nine years were used for this analysis. RegCM4 with both hydrostatic and non-hydrostatic dynamics used the CLM4.5 land surface component, since RegCM4 with CLM4.5 performs better at simulating mean and extreme precipitation [82]. The outputs of RegCM_CLM_12km and RegCM_CLM_3 km were regridded to 0.1° × 0.1° (~10 km) in China and 0.025° × 0.025° (~2.5 km) over JJJ, with distance weight, to study the population exposure to precipitation extremes. Additionally, this study has mainly investigated the impacts of current and future CMIP6 forcing on precipitation extremes without land use changes as limitations, and the contributions of global atmospheric circulation and regional land use are expected to be revealed in the near future.

2.3. Precipitation Extreme Indices

Different definitions of precipitation extremes may affect the conclusions of climate change studies [88]. In this study, we have used extreme precipitation indices defined by the Expert Team on Climate Change Detection and Indices (ETCCDI) [89]. These indices have been widely used in studies on climate extremes [90,91,92,93,94]. Four precipitation extreme indices from the ETCCDI were used to study changes in future extremes: the annual total precipitation on days when the daily precipitation would exceed 95% percentile of the daily precipitation (R95p, unit: mm); the annual maximum 5-day consecutive precipitation (Rx5day, unit: mm); the annual maximum consecutive days with precipitation values greater than 1 mm (CWD; unit: days), and that of those less than 1 mm (CDD; unit: days).

3. Results

3.1. Changes in the Future Population in China and the JJJ Region

Figure 2 shows the spatial distribution of the population density in China in 2000 at a spatial resolution of 0.1° × 0.1° and in the JJJ region at a 0.025° × 0.025° resolution, as well as their predicted changes in 2100 relative to 2000. The gradient of the spatial distribution of the population density in China followed the spanning dry, semi-dry, semi-humid and wet complex climate types from the northwest to the southeast of China (Figure 2a). The majority of people live in Northern and Southern China, which belongs to the East Asian monsoon region [95]. By 2100, the population density in China is predicted to moderately decrease, especially in Northern and Southern China under the SSP245 scenario (Figure 2b) [31]. Chen, Guo, Wang, Cai, Wang and Wang [95] estimated that the population in China will increase to a peak, ranging from 1.44 to 1.48 billion around 2027–2034 m and then decrease continuously until 2100.

As a typical urban agglomeration of China, the JJJ region has a high population density due to fast urbanization (Figure 2c). The population is mainly concentrated in the downtown areas of Beijing, Tianjin and the large cities of Hebei (dark regions in Figure 2c). Similarly to the changes in the future population density in China, the population density over the JJJ region was predicted to decrease by 2100 (Figure 2d). The only exception was a moderate increase in the population density of the high-density downtown areas of JJJ, which means that more and more people will move to those areas in the future.

3.2. Changes in Future Precipitation Extremes

In a previous study, we evaluated the performances of RegCM4 with the hydrostatic dynamic core in China and the non-hydrostatic dynamic core over the JJJ region in simulations of historical mean and extreme precipitation during 1991–1999 [82]. High-resolution RCMs can characterize the spatial distribution of precipitation extremes in high detail and exhibit good performance in simulating precipitation extremes.

Figure 3 and Figure 4 show the spatial distributions of changes in the precipitation extreme indices in China, for the future period of 2091–2099, and the JJJ region, relative to the historical period of 1991–1999, using MPI-ESM1.2-HR, RCM_CLM_12km and RCM_CLM_3km. Here, the absolute changes in precipitation extremes were presented rather than the relative changes, since absolute changes in extremes directly contribute to changes in population exposure to extremes [48]. Reasonably simulating mean and extreme precipitation is a great challenge, as extremes in precipitation depend on many physical processes and have high spatiotemporal heterogeneity [14]. In spite of the fact that most areas of China are generally projected to have warmer climates by the end of the 21st century [4], heavy precipitation extremes R95p and Rx5day, simulated with both MPI-ESM1.2-HR and RCM_CLM_12km, have shown large spatial variations (Figure 3a–d). The future R95p extremes simulated with MPI-ESM1.2-HR and RCM_CLM_12km showed increases with rates of 3.54 and 13.62 mm, respectively, over the NW subregion; 15.14 and 28.08 mm, respectively, over the TP subregion; 26.77 and 31.45 mm, respectively, over the NC subregion; and 30.95 and 3.38 mm, respectively, over the SC subregion (Figure 3 and Figure 5a). RegCM4_CLM4_12km showed more spatial details of precipitation extremes due to its higher resolution compared to the MPI-ESM1.2-HR results. The R95p in the whole of China was predicted to increase by 12.62 and 19.90 mm with MPI-ESM1.2-HR and RCM_CLM4_12km, respectively (Figure 5a). Future changes in the Rx5day and R95p showed similar results, with increases over China as well as its four subregions, with both MPI-ESM1.2-HR and RCM_CLM_12km (Figure 3b,d). These increases in the R95p and Rx5day over the whole of China and its subregions are probably due to the warmer climate, which will increase the ability of the atmosphere to hold more water vapor following the C-C relationship [1,2] and the specific humidity in the atmosphere in general (Figure 6b).

RegCM4-NH produced more elaborate spatial distributions of the R95p and Rx5day over the JJJ region (Figure 3e,f). Due to a high spatiotemporal variability of the mean and extreme precipitations, changes in the future R95p and Rx5day over a small region showed different results than those of the larger area. R95p and Rx5day with RCM_CLM_3km over JJJ were found to moderately decrease by 56.71 and 48.29 mm, respectively, in spite of low increases in specific humidity (Figure 5d), with a larger decrease than simulated with RCM_CLM_12km, which gave decrease rates of 29.49 and 22.96 mm, respectively. The probable reason for this is the lower increases in future specific humidity over the JJJ region compared to other areas in China (Figure 6b,d), with slight increases in specific humidity and no obvious changes in horizontal wind at 500 hPa over JJJ in the summer (JJA) or winter (DJF) or the annual mean (Figure 6c–h). Land surface conditions such as land use and topography might greatly impact changes in precipitation extremes due to the high spatial resolution of 3 km used. Here, the mechanisms of the changes in precipitation extremes have been given in brief, and more detailed mechanism analysis will be carried out in the future.

Compared to R95p and Rx5day, changes in future consecutive wet days (CWDs) were not obvious, with increasing or decreasing rates ranging from −0.78 to 2.03 days in China and its four subregions with MPI-ESM1.2-HR and RCM_CLM_12km, respectively (Figure 5e). Similar results for CWDs over JJJ with RCM_CLM_3km showed slight decreases in the future and similar spatial patterns from those shown for R95p and Rx5day (Figure 5f). As a dry precipitation extreme index, consecutive dry days (CDDs) generally present negative change rates compared to R95p and Rx5day (Figure 4b,d,f). The CDDs estimated with both MPI-ESM1.2-HR and RCM_CLM_12km were found to decrease at rates of 2.83 and 1.39 days, respectively, over the NW subregion; 0.54 and 4.00 days, respectively, over the TP subregion; and 0.50 and 1.56 days, respectively, over the NC subregion and increase at rates of 0.23 and 3.89 days, respectively, over the SC subregion. RCM_CLM_3km showed a lower CDD decrease rate, of 0.78 days, compared to RCM_CLM_12km, which produced a CDD decrease rate of 4.47 days. RCM_CLM_3km presented decreased and increased CDDs over the north and south of the JJJ region, respectively (Figure 4h), which might be due to the different terrain in the mountain and plain regions (Figure 1b). Mountain areas might present more complex climates relative to plain regions [96], and more detailed analysis together with that of different land use will be performed in specific studies.

Therefore, precipitation extremes with higher spatial resolution modeling may result in different conclusions than those predicted by coarse models. The inclusion of physical processes and higher-resolution input data should also be developed to improve high-resolution climate models.

3.3. Changes in Population Exposure to Precipitation Extremes

After investigation of the changes in the future population and precipitation extremes in China and the JJJ region, the changes in the future population’s exposure to these precipitation extremes over the JJJ region were analyzed using RCM_CLM_3km. Population exposure to precipitation extremes is quantified using precipitation extreme indices multiplied by the amount of people exposed [31,74,78]. This is expressed as EXPO = POP × PE, where EXPO is the population exposure to precipitation extreme indices, POP is the population and PE represents extreme indices. Changes in population exposure are decomposed into three components: the impact of extreme precipitation, the impact of the population and the joint impact of the extremes and the population [79,97]. This is expressed as

$$∆\mathrm{E}\mathrm{X}\mathrm{P}\mathrm{O}=\mathrm{P}\mathrm{O}\mathrm{P}\times ∆\mathrm{P}\mathrm{E}+∆\mathrm{P}\mathrm{O}\mathrm{P}\times \mathrm{P}\mathrm{E}+∆\mathrm{P}\mathrm{O}\mathrm{P}\times ∆\mathrm{P}\mathrm{E},$$

(1)

where $∆\mathrm{E}\mathrm{X}\mathrm{P}\mathrm{O}$ represents changes in the population’s exposure to extreme precipitation; $\mathrm{P}\mathrm{O}\mathrm{P}$ and $∆\mathrm{P}\mathrm{O}\mathrm{P}$ are the historical population and future changes in the population, respectively; and $\mathrm{P}\mathrm{E}$ and $∆\mathrm{P}\mathrm{E}$ are the historical precipitation extreme index and future changes in precipitation extremes, respectively.

Figure 7, Figure 8, Figure 9 and Figure 10 show the population exposure to R95p during the historical period of 1991 to 1999; projected future changes in 2091–2099, relative to the historical period; and the percentage contributions from changes in precipitation extremes, changes in the population and combined changes in precipitation extremes and population. The spatial patterns of the historical population’s exposure to the R95p and Rx5day wet extreme indices in the JJJ region were generally dominated by the spatial distribution of the population density, i.e., more exposure to R95p and Rx5day over areas of JJJ with higher population densities (Figure 7a and Figure 8a). This is consistent with global results [31]. Because of a moderate decrease in the future population over most of the JJJ area except in some downtown regions (Figure 2), as well as less R95p and Rx5day over most of the JJJ region (Figure 3 and Figure 5b,d), changes in the future population’s exposure to R95p and Rx5day are expected to decrease over almost all of JJJ, especially over the south, with decreases of more than 10 × 10⁴ person mm (Figure 7b and Figure 8b). Changes in exposure to both R95p and Rx5day due to changes in the population and the extreme indices will present negative contributions to the total changes in the population’s exposure to these two indices over most of JJJ. These changes can range from about −60 to 20 × 10⁴ person mm and from −40 to 10 × 10⁴ person mm, except in some downtown regions with projected increasing populations for the future (Figure 7d,e and Figure 8d,e). The contributions of the changes in population exposure due to the combined impacts of the population and extreme indices were positive but had lower absolute values compared to the impacts due to changes in either the population or extreme indices alone, representing no more than 50,000 person mm over most of JJJ (Figure 7f and Figure 8f). The percentage of absolute contribution from each component of the changes in the population exposure in Figure 7c, Figure 8c, Figure 9c and Figure 10c is defined as

$$\mathrm{a}\mathrm{b}\mathrm{s}(\mathrm{P}\mathrm{O}\mathrm{P}\times ∆\mathrm{P}\mathrm{E})/\left(\mathrm{abs}\right(\mathrm{P}\mathrm{O}\mathrm{P}\times ∆\mathrm{P}\mathrm{E})+\mathrm{a}\mathrm{b}\mathrm{s}(∆\mathrm{P}\mathrm{O}\mathrm{P}\times \mathrm{P}\mathrm{E})+\mathrm{a}\mathrm{b}\mathrm{s}(∆\mathrm{P}\mathrm{O}\mathrm{P}\times ∆\mathrm{P}\mathrm{E}\left)\right)\times 100\mathrm{\%},$$

(2)

with the spatial average population exposure over the JJJ region, where $\mathrm{a}\mathrm{b}\mathrm{s}(\bullet )$ is the absolute value operator. Predicted changes in the future R95p and population over JJJ represented the dominant contributions to the total exposure changes in R95p, with percentages of 43.47% and 44.30%, respectively, while the combined impact of R95p and the population represented 12.23% (Figure 7c). The percentages of the three components of exposure to Rx5day over the JJJ region were 47.92%, 38.37% and 13.70%, respectively.

Similarly to the simulated exposure changes to R95p and Rx5day, the total population exposure to CWD decreased over most of JJJ but with relatively small rates of no more than 1 × 10⁴ person days, partly due to small changes in future CWDs (Figure 4e, Figure 9 and Figure 11c). We found negative contributions to changes in the total exposure due to changes in the future population and CWDs but positive contributions to the combined changes in the population and CWDs, with percentage contributions of 31.62%, 58.95% and 9.43%, respectively (Figure 9).

Unlike the total exposure to R95p, Rx5day and CWD wet precipitation extreme indices, the changes in the total population exposure to CDDs were projected to increase over the large downtown areas of JJJ but decrease over other parts of JJJ (Figure 10d). This is consistent with the spatial distribution of changes in the projected future population (Figure 2d). The total exposure to CDDs over the JJJ region ranged from about −5 to 5 × 10⁴ person days (Figure 11d). The percentage contribution of the population changes to the total exposure was 84.92%: largely dominant among the three components of the total population (Figure 10c).

Finally, we quantified the changes in the total population exposure to precipitation extremes with air surface temperature (TAS) increases over the JJJ region, which have been widely adopted in previous studies [4]. Changes in the population exposure to precipitation extremes per degree of warming were calculated through spatial mean changes in the precipitation extremes and the surface air temperature over the JJJ region, with a sample size of 36,098. The multi-year annual mean temperatures in 1991–1999 and 2091–2099, which were calculated with the daily temperature mean in each grid of the JJJ region, represent the climatological temperatures during the historical and future periods, respectively. The increase rate of the TAS over JJJ generally ranged from about 2.7 to 3.5° (Figure 11). Under warmer climates in the period of 2091–2099, the expected changes in the total population exposures to R95p, Rx5day, CWDs and CDDs showed decreases of 7.37 × 10⁴ person mm, 5.84 × 10⁴ person mm, 0.15 × 10⁴ person days and 0.64 ×10⁴ person days, respectively, and these changes are significant at the 0.05 level using a t-test with grid-point numbers of 12,925, 25,251, 32,178 and 27,451, respectively.

In summary, the changes in total population exposure to the four precipitation extreme indices in this study were found to have generally decreased over most areas of the JJJ region. Changes in the future population and the extreme indices R95p and Rx5day will contribute moderately to changes in the total exposure over JJJ, while the population will dominantly impact the total exposure to CDDs. In addition, expected changes in total exposure to precipitation extremes per degree of warming were quantified over JJJ during 2091–2099 and compared with those of the historical period of 1991–1999.

4. Discussion

The regional climate model on the convection-permitting scale presented a good performance in simulating the regional climate, especially precipitation extremes, which usually show high spatiotemporal variations [66,98]. Benefitting from their high spatial resolutions on the kilometer scale, RCMs on the CP scale could capture more details of regional climate conditions [56] and better simulate land–atmospheric interactions, especially over areas with frequent anthropogenic activities, such as urban areas [68]. The present study has firstly analyzed changes in precipitation extremes with RegCM4 on the CP scale over the JJJ urban agglomeration with fast urbanization and rapid population growth.

The risks of precipitation extremes are impacted by the extremes themselves as well as the populations involved [31,76]. This study has investigated changes in precipitation extremes over the JJJ region as well as population exposure to extremes by RegCM4 with a non-hydrostatic dynamic core and a fine spatial resolution of 3 km and revealed the possible reason of these changes from annual or seasonal global atmospheric circulation. As in the previous study, the changes in precipitation extremes presented large spatial variations under climate change [48,68] due to too many physical processes being involved [99]. This study has focused on the impacts of current and future GCM forcing to precipitation extremes without land use changes. Since precipitation extremes are also affected by regional land–atmospheric interactions, such as evapotranspiration and land use; the contributions of global circulation and regional land–atmospheric interactions should be deeply investigated, and the performance of dynamic downscaling in simulating precipitation extremes could be evaluated in more detail, such as with regard to added values [100,101].

To better investigate the population’s exposure to climate extremes, population policies should also be considered, as this might greatly affect population sizes and spatial distribution [95]. For example, the number of permanent residents in Beijing was restricted to 23 million in 2020 [102]. This study has compared the changes in the population exposure to precipitation extremes during the historical period from 1991 to 1999 and projected for the future period from 2091 to 2099, and different selections of historical and future periods might present different conclusions. In our future work, more recent historical populations will be adopted and precipitation extremes will also be simulated for the same time periods. These conclusions are model-dependent, and more simulations should be carried out to reduce uncertainties due to RCMs, GCMs and their different experimental setups. This could be carried out by using the multiple RCMs that participate in CORDEX [45,103] and with more GCM forcing in CMIP5 or CMIP6. Compound climate extremes usually result in greater impacts on human lives and society [104,105]. The population’s exposure to compound extremes should also be investigated on the convection-permitting scale in future studies.

5. Conclusions

Warmer climates increase the ability of the atmosphere to hold more water vapor [1] and impact the occurrence of precipitation as well as precipitation extremes [106]. In the present study, changes in precipitation extremes and the population exposure to these extremes have been projected for the future period of 2091–2099 relative to those for the historical period of 1991–1999 in in China and the JJJ urban agglomeration, with spatial resolutions of 12 km and 3 km, respectively. The percentage contributions of the three components of total exposure, precipitation extremes multiplied by changes in the population amount, changes in precipitation extremes multiplied by the population amount and the multiplication of the changes in precipitation extremes by the changes in the population amount, were then investigated [107]. Additionally, changes in the population exposure to precipitation extremes for each degree of local warming were calculated for the JJJ region.

First, RegCM4 with a hydrostatic dynamic core at a spatial resolution of 12km was run over East Asia, with the whole of China included, for the historical and future periods, supported by high-resolution MPI-ESM1.2-HR in CMIP6 under the middle SSP245 scenario [80]. Results above 12 km were then used to derive the initial and boundary conditions to force RegCM4-NH at a spatial resolution of 3 km over the JJJ region. The performances of RegCM4 with hydrostatic and non-hydrostatic dynamic cores in China and over the JJJ region have been investigated that RegCM4-NH could simulate more detailed spatial distribution of precipitation over the JJJ region but with large spatial variations, as has been mentioned in other studies [68].

The heavy precipitation extremes R95p and Rx5day are projected to increase in the majority of China while decreasing over the JJJ region, which is consistent with previous studies [48,108]. These changes might be caused by slight increases in specific humidity over the JJJ region in the future, with moderate increases in the rest of China. On the other hand, land surface conditions such as land use and topography profoundly impact precipitation extremes through land–atmospheric interactions. RegCM4-NH, with a higher spatial resolution of 3 km, might better simulate such local interactions than can RCMs with hydrostatic dynamic cores [54]. On the other hand, more precise descriptions of physical processes and basis data, with higher resolutions consistent with RegCM4-NH, should be developed to improve RegCM4-NH simulation further.

Finally, changes in population exposure to precipitation extremes were investigated over the JJJ region. The population densities in China and JJJ are expected to decrease in general by 2100, except in some downtown areas of JJJ, where more people are expected to migrate from rural to urban areas with already high population densities. The population exposure to precipitation extremes is projected to decrease over most areas of the JJJ region. The contributions of changes in precipitation extremes and the population amount to the total exposure were then investigated [106]. The future population amount and precipitation extremes R95p and Rx5day were projected to equally contribute to changes in the total exposure over the JJJ region, while changes in the population amount were the dominant factor that would impact the total exposure to CDDs. Additionally, changes in population exposure to precipitation extremes under each degree of local warming were quantified.

The present study has had some limitations. One is that only one GCM was used to force the RCM RegCM4, and the conclusion was model-dependent [109]. Multiple RCMs forced by several GCMs might be better to reduce model uncertainties of precipitation simulation. Another limitation is that this study investigated changes in precipitation extremes due to the differences in current and future CMIP6 forcing without a land use change effect. In our following work, the contributions of changes in GCM forcing and land use changes during a more recent historical period with longer time span (such as 1985–2014) and a future period will be analyzed in more detail, as will the contributions of atmospheric circulation and local land–atmospheric interaction.