Characteristics of population exposure to climate extremes from regional to global 1.5 °C and 2.0 °C warming in CMIP6 models

The intensities and occurrences of heat extremes are projected to increase in a warmer climate, and relevant policies have been established to address different warming levels. However, how climate extremes change at regional warming levels is not well-known because changes in temperature vary over different regions. This study investigated climate extremes and population exposure to these extremes at regional and global 1.5 °C or 2.0 °C warming over 58 reference regions with 16 Coupled Model Intercomparison Project, 6th phase models. The years of reaching local 1.5 °C or 2.0 °C warming occurred earlier than the timing of global warming over certain land areas, with more than 30 years advance in northern high latitude land areas. Heat extremes are projected to increase in all reference regions under regional and global 1.5 °C or 2.0 °C warming. Moving from regional to global 1.5 °C or 2.0 °C warming, heat extremes were found to increase over most land areas, especially over mid- and high-latitude areas. Population exposure to climate extremes increased over more than half the land regions under regional to global 1.5 °C or 2.0 °C warming. Changes in population exposure to absolute heat extremes were mainly generated by changes in population over about 34 land regions, whereas changes in population exposure to percentile-based heat extremes over more than 40 land regions were mostly due to changes in climate extremes. These results provided references to establish relevant strategies at regional scale to address possible risks related to climate extremes.


Introduction
As one of the most important climate hazards, heat extremes have a severe impact on society, natural ecosystems and human safety, and have received much attention (Russo et al 2019, Dong et al 2021, IPCC 2021, Wang and Yan 2021, Bartusek et al 2022, Wang and Wang 2023, Wu et al 2023, Yin et al 2023, Zhang and Boosa 2023, Zhou et al 2023, Zhang et al 2023c).Increasing heat extremes have led to deterioration in health and mortality (IPCC 2023).North America experienced exceptional heat extremes in the summer of 2021 with more than 200 heat-related deaths in Washington and Oregon (Bartusek et al 2022, WMO 2022, Dong et al 2023).In 2022, both China and Europe experienced severe, long-lasting heat extremes over large areas.As a result, about 15 000 people in Europe lost their lives (Ma and Yuan 2023, WMO 2023, Zhang et al 2023a).Many wildfires occurred during or after the record-breaking heat extremes in North America in 2021 and China in 2022, which brought about enormous economic loss and seriously damaged natural ecosystems (WMO 2022(WMO , 2023)).
Under warmer climate, the intensities and occurrences of heat extremes are projected to increase in general (King and Karoly 2017, Zhang et al 2021, Qin 2022, Kim et al 2023).The Paris Agreement aims to keep global warming well below 2.0 • C and ideally below 1.5 • C compared with preindustrial levels from 1850 to 1900 (UNFCCC 2015, Seneviratne et al 2018, Meinshausen et al 2022).At global warming greater than 1.5 • C, there is heightened chance of major climate tipping points being breached, such as rising sea level due to ice sheet collapse in the Antarctic, the Arctic, and Greenland and loss of mountain glaciers (Mckay et al 2022).
However, relevant policies established to address 1.5 • C or 2.0 • C warming cannot totally eliminate the risks associated with global warming, such as regional climate extremes have already exceeded the specific thresholds and might result in severe damage to the society and ecosystems, since changes in air-surface temperature (TAS) present large spatial variabilities (Seneviratne et al 2018).Therefore, climate extremes under regional warming should also be investigated over different regions of the world in addition to global warming as well as differences between regional and global warming.(Chen et al 2020, Qin 2022).However, how population exposure to extremes changes when regional temperature reaches 1.5 • C or 2.0 • C warming is not well-known since the timing at regional warming levels varies over different regions, i.e. behind or ahead of the timing at global warming levels.
Therefore, it is crucial to deeply understand differences in climate extremes and their population exposure under different regional and global warming levels to better mitigate and adapt to future risks related to climate extremes.Based on global climate models from the Coupled Model Intercomparison Project, 5th phase (CMIP5) (Taylor et al 2012), Harrington et al (2018) investigated how much warming was needed at each grid to meet the same changes in a heat extreme index TXx at 1.5 • C warming over region of interest.But they did not present the differences of changes in climate extremes at local and global warming levels which is helpful to establish relevant strategies at regional scale to face possible climate risks since changes in temperature shows great spatial variability.Furthermore, Harrington and Otto (2018) analyzed population exposure to heat extremes over the Eastern Africa and Southern Asia at 1.5 The following issues were revealed in this study: (1) the years when regional 1.5 • C or 2.0 • C warming was reached over 46 land and 12 ocean reference regions (Iturbide et al 2020, IPCC 2021), whether behind or ahead of global warming; (2) the differences in climate extremes (heat and wet extremes) under regional and global 1.5 • C or 2.0 • C warming, as well as in population exposure to climate extremes; and (3) the contributions of climate extremes and population size to population exposure to climate extremes.

Data
The global 0.125 • population in 2000 and during 2010-2100 at 10 year intervals was used to study population exposure to climate extremes (Jones and O'Neill 2016), which has been adopted in previous studies (Jones et al 2018, Lei et al 2022, Qin 2022).In addition, global 1 km populations in 1970, 1980, and 1990 from the Center for International Earth Science Information Network were also used because many regions first reached regional 1.5 • C or 2.0 • C warming before 2000 (CIESIN 2017).These population data in every decade were regridded to 1 • × 1 • spatial resolution with distance-weighted method after converting to population density and then linearly interpolated to annual population from 1970 to 2100.
The HadEX3 gridded climate extremes product from the Met Office Hadley Centre was used to provide observed extremes to evaluate the performance of the CMIP6 models in simulating climate extremes (Dunn et al 2020).It ranges from 1901 to 2018 with 1.25 • × 1.875 • spatial grid cells and includes 29 climate extreme indices by the Expert Team on Climate Change Detection and Indices (ETCCDI) (Frich et al 2002, Zhang et al 2011).
Daily precipitation (PR), maximum (TX) temperature, and minimum (TN) temperature from 16 CMIP6 models (Eyring et al 2016) under the SSP245 scenario (Fricko et al 2017) during the historical period from 1950 to 2014 and the future period from 2015 to 2100 were adopted to calculate climate extremes (table S1).To make them comparable, climate extremes by HadEX3 and CMIP6 models were both remapped to 1 The annual TASs during 1850-2100 from the 16 CMIP6 models were used to derive the years when regional and global 1.5 • C or 2.0 • C warming levels were first reached after distance-weighted regridding to 1 • × 1 • spatial resolution and 20 year window smoothing as shown in section 2.3.

Random forest regression
Because time series of climate extremes usually present large perturbations (Qin 2022), the random forest regression method, which is a widely used machine learning method, was used to reveal the trends of climate extremes with climate change.Random forest is nonparametric and can capture linear and nonlinear relationships of related data sets (Breiman et al 2003, Reichstein et al 2019).

Identifications of regional and global 1.5 • C or 2.0 • C warming levels
Similar methods to those used in the Intergovernmental Panel on Climate Change (IPCC) Sixth Assessment Report (AR6) (IPCC 2021) were used to identify the year when each region first reached regional and global warming levels.TAS was averaged with latitude weighted at global scale or over the 58 IPCC reference regions (figure 1(a)) for every CMIP6 model during 1850-2100 with the mean TAS during 1850-1900 for pre-industrial conditions removed (Hawkins et al 2017).The TAS time series were then smoothed using a 20 year moving average as in IPCC AR6 (IPCC 2021) and other studies (Seneviratne and Hauser 2020, Wu et al 2021, Qin 2022).We also investigated the effects of different time smoothing windows such as 10 year and 20 year window and found small differences in the timing when reaching 1.5 • C or 2.0 • C warming levels (figure S1).The year 'n' when the temperature first attained the given warming level at global scale or over any of the 58 reference regions was identified as the year when the related warming level was reached, and the time window [n − 9, n + 10] was assumed to be the period at that warming level.In addition, warming levels for each 1 • × 1 • grid cell at global scale were also calculated through the TAS time series of that grid cell by the same method.

Climate extreme indices
The climate extreme indices used in this study were suggested by the ETCCDI (Frich et al 2002, Zhang et al 2011), including daytime heat extremes: the maximum of daily TX (TXx) and the percentage of days with TX greater than 90% percentile (TX90p), nighttime heat extremes: the maximum of daily TN (TNx) and the percentage of days with TN > 90% percentile (TN90p), and wet extreme of total precipitation with PR > 95% percentile (R95p).The base period for calculating percentile-based thresholds was from 1981 to 2010.

Results
3.1.Regional and global 1.5 • C or 2.0 • C warming Before investigating climate extremes, the spatial variabilities of the regional and global warming figures are summarized.Figures 1(b) and (c) show the differences between the year when local 1.5 • C or 2.0 • C warming was first reached over each 1 • × 1 • grid cell and the year when global 1.5 • C or 2.0 • C warming was first reached, as well as TAS minus 1.5 • C or 2.0 over each grid cell.The global temperature is expected to reach a 1.5 • C or 2.0 • C increase in around 2030 and 2049, respectively (the two vertical black lines in figure 2) based on the 16 CMIP6 models.In general, the years when local 1.5 • C or 2.0 • C warming occur are earlier than global warming over land, with more than 30 years' advance in northern highlatitude land areas for 1.5 • C warming, and more than 40 years ahead for global 2.0 • C warming.Therefore, it can be stated that land will warm faster, with higher warming levels compared with global warming.On the other hand, the years when the temperature over ocean regions reaches local 1.5 • C or 2.0 • C warming lag behind global warming, especially for the southern Pacific Ocean (SPO) (figure 1(a)).

Climate extremes at different regional and global warming levels
Compared with the historical period, wet extreme (R95p) at regional 1.5 • C warming was projected to increase over almost all land regions and many ocean regions except for the southern Indian Ocean, the equatorial and southern Atlantic Ocean, and the SPO (figures 3(y) and 4).Moving from regional to global 1.5 • C warming, heat extremes were found to increase over most land areas (figures 3(c), (i), (o), (u) and 4), especially over mid-and high-latitude areas such as North America (NWN, NEN, WNA, SNA, and ENA), Europe (NEU, WCE, and EEU), and northern Asia (RAR, WSB, ESB, ECA, and Eastern Asia (EAS)).Changes in R95p showed similar spatial distributions of heat extremes, i.e. moderate increases over northern mid-and high-latitude land areas, which might be generated by a warmer climate following the Clausius-Clapeyron relationship (O'Gorman and Schneider 2009) (figures 3(A), and 4).Similar results were also found when moving from regional to global 2.0 • C warming, but with stronger rates of increase over land regions (figures 3(d), (j), (p), (v), (B) and 4).

Population exposure to climate extremes at different regional and global warming levels
Population changes at different local and global warming levels were investigated before examining population exposure to climate extremes.The  As in previous studies (Jones et al 2018, Liu et al 2021, Qin 2022), population exposure to climate extremes and their changes can be defined as: where EP is population exposure to climate extreme index E with an involved population size of P and ∆EP, ∆E, and ∆P are the changes in these quantities.The spatial distributions of historical population exposure to climate extremes are determined both by the population distribution and by the intensity of the extremes (figure S4).The 16 CMIP6 models present the spatial patterns of historical exposure to climate extremes with positive biases of heat extremes and negative biases of precipitation extremes over the SAS and EAS regions of high population density (figure S4), as shown in a previous study (Qin 2022).Because of increases in both heat extremes and population size at local and global 1.5 or 2.0 • C warming relative to the historical period, population exposure to heat extremes showed moderate increases over central Africa (WAF, NEAF, SEAF), EAS, and SAS (figures 5(a), (b), (g), (h), (m), (n), (s), (t) and S5) (Iyakaremye et al 2021), as shown in previous studies (Sun et al 2022, Ullah et al 2022).For example, increases in TXx at regional 1.5 • C warming were greater than 30 × 10 6 person • C in the WAF and SEAF regions and more than 26 and 118 × 10 6 person • C in the EAS and SAS regions (figures 5(a) and S5).Changes in precipitation extremes showed similar spatial patterns to those of heat extremes (figures 5(y), (z) and S5).

Contributions of population and climate extremes to exposure to climate extremes
Population exposure to different climate extremes exhibits different change patterns over different regions under different warming levels.The relative contributions of the factors related to population exposure to extremes need to be identified (Chen et al 2020).
Figure 6 shows differences in population exposure to climate extremes between regional and global 1.5 • C or 2.0 • C warming in the 46 land reference regions used in IPCC AR6, as well as the contributions of the three components of changes in exposure.Among the 44 populated regions among the 46 land reference regions, population exposure to heat and wet extremes was found to increase over about 28 regions due to regional and global 1.5 • C warming.The number of regions with positive changes in population exposure to climate extremes ranged from 24 to 30 between regional and global 2.0 • C warming (figure 6).Changes in population exposure to absolute heat extremes (TXx and TNx) were predominantly controlled by ∆P × E, i.e. changes in population multiplied by heat extremes, over about 34 land regions when moving from regional to global 1.5 • C and 2.0 • C warming.On the contrary, changes in population exposure to percentile-based threshold heat extremes (TX90p and TN90p) over more than 40 land regions were mostly generated by P × ∆E, which is the changes in the extremes multiplied by the population size involved (figure 6).Besides changes in climate extremes and population size varying over different regions and the uncertainties due to multiple CMIP6 ensembles, another possible reason for this result is that changes in absolute heat extremes usually involve a few degrees of warming, whereas changes in percentile-based heat extremes are close to or greater than 10 (figure 3).In addition, there were about 30 land regions with changes in population exposure to wet extreme, which were mainly generated by changes in population multiplied by the extremes.
This paper gives an overview of contributions to changes in population exposure to five climate extremes over 44 land regions with human populations living under regional to global 1.5 • C or 2.0 • C warming levels.Due to the spatiotemporal variabilities of different climate extremes and the different mechanisms of these extremes, deeper analyses should be performed for specific regions as well as for related physical processes to reveal the possible reasons behind these changes.
Numerous studies have focused on evaluation of historical CMIP6 multi-model simulations of climate extremes and have found that CMIP6 models could reproduce spatiotemporal historical climate extremes in general (Dong and Dong 2021, Faye and Akinsanola 2021, Zhao et al 2021, Qin 2022, Ali

Figure 1 .
Figure 1.Global elevation and the land reference regions (a) used in this study from IPCC AR6, with the years behind or ahead of global 1.5 • C (b) or 2.0 • C (c) warming and the temperature differences (d), (e).
et al 2023).Compared with HadEX3 extreme indices during the historical period from 1981 to 2000, the ensemble of 16 CMIP6 models could present the spatial patterns of climate extremes in comparison with HadEX3 extreme indices (figure S2).A thorough evaluation of these 16 CMIP6 models has been performed by Qin (2022).
Figure 2 presents anomalies of daytime (TXx and TX90p) and nighttime (TNx and TN90p) heat indices and heavy precipitation (R95p) compared with historical period during 1981-2000 by the multiple CMIP6 model ensemble median over the 46 land (warm lines) and 12 ocean (cold lines) reference regions (Iturbide et al 2020) used in IPCC AR6 (IPCC 2021).A random forest regression scheme with a maximum depth of four was also applied to these time series to exhibit the trends of climate extreme anomalies more obviously (Qin and Shi 2021).Heat extremes are projected to increase markedly from 2015 to 2100 under the SSP245 pathway over all land and ocean reference regions (figures 2(a)-(d)) (Qin 2022).

PFigure 2 .
Figure 2. Time series of climate extreme anomalies from historical period during 1981-2000 over 46 land (lines with warm colors) and 12 ocean (lines with cold colors) reference regions from IPCC AR6, with their ensemble (black line) by the 16 CMIP6 models' ensemble median.The thick colored or black lines in each figure are regressions with the random forest method to time series of climate extremes, and the two vertical lines are the years of global 1.5 • C or 2.0 • C warming.
Under the joint impacts of changes in climate extremes and population size, changes in population exposure to climate extremes exhibited wide spatial variations when moving from local to global 1.5 • C or 2.0 • C warming (figures 5(c), (d), (i), (j), (o), (p), (u), (v), (A), (B) and S5).Because local 1.5 • C or 2.0 • C warming over most areas of the SAS region occurred behind global warming, small population size and climate extremes resulted in negative changes in population exposure to extremes between glob1.5 and reg1.5 and between glob2.0 and reg2.0,respectively.

Figure 6 .
Figure 6.Changes in population exposure to climate extremes over 46 land reference regions from regional to global 1.5 • C or 2.0 • C warming (a); the number of regions with increases (b) or decreases (c) of exposure changes, as well as the percentage contributions of the three components.
The disaster risks by climate extremes are affected both by climate extremes and population amount involved, i.e. population exposure to climate extremes (Jones et al 2018, Zhang et al 2018, 2023b, Chen and Sun 2019, 2021, Li et al 2020, 2022, Liu et al 2021, Thiery et al 2021).Numerous studies have investigated population exposure to various climate extremes, such as precipitation (Jones et al 2018, Shen et al 2022), flooding (Smith et al 2019), heat (Tuholske et al 2021, Ascencio et al 2023, Zhang et al 2023b), drought (Chen and Sun 2019), and compound extremes (Liu et al 2021, Zhang et al 2022, Wang et al 2023).Under 1.5 • C or 2.0 • C global warming, changes in population exposure to climate extremes have been investigated regionally (King et al 2018, Wu et al 2021) and globally • × 1 • resolution with distanceweighted method as in previous studies (Kim et al 2020, Iyakaremye et al 2021, Wu et al 2022, Paik et al 2023), since spatial resolutions of majority of the 16 models were close to 1 • .