Will population exposure to heat extremes intensify over Southeast Asia in a warmer world?

The daily maximum and daily minimum temperatures in this study are derived from 20 Coupled Model Intercomparison Project Phase 6 (CMIP6) ensembles (supplementary table S1 available online at stacks.iop.org/ERL/17/044006/mmedia, Eyring et al 2016). The future changes in temperature extremes are projected under two Shared Socioeconomic Pathway (SSP) scenarios (SSP2-4.5 and SSP5-8.5, O'Neill et al 2016). They represent the medium part and high end of the ranges of future forcing pathways. In order to evaluate the historical runs of CMIP6 during the period of 1985–2014, the observational datasets of SA-OBSv2.0 covering the period of 1981–2017 derived from Southeast Asian Climate Assessment & Dataset (SACA&D) is selected in this study (Van Den Besselaar et al 2017). It has undergone accurate quality control and is currently one of the high–quality available gridded observational dataset for SEA, which has been widely used to evaluate regional and global climate simulations (Van Der Schrier et al 2016, Van Den Besselaar et al 2017, Ge et al 2019a, Dong et al 2021).

The domain of SEA is located from 10° S to 23° N and 95° E to 140° E and divides into five subregions as shown in figure S1: the Indochina Peninsula (ICP; 6° N–23° N, 95° E–110° E), Kalimantan (KAL; 4° S–6° N, 109° E–118° E), Sumatra (SUM; 8° S–6° N, 95° E–108° E), Philippines (PH; 5° N–20° N, 118° E–130° E), and Sulawesi (SUL; 6° S–3° N, 118° E–126° E).

Table 1 shows the four recommended indices from the Expert Team on Climate Change Detection and Indices (ETCCDI), which are used to estimate the temperature extremes over SEA (You et al 2011, Sillmann et al 2013b, Sheikh et al 2015). Further information could be found at the ETCCDI website, http://etccdi.pacificclimate.org/indices.shtml.

Based on assumptions that the population was predominantly influenced by fertility, income, and political change, a new spatial population downscaling model dataset is provided by the National Center for Atmospheric Research and City University of New York Institute for Demographic Research (NCAR-CIDR). Taking into account the uneven economic development of different regions in SEA, we used population projections under SSP3, which is the pathway that faces great challenges of climate change adaptation and mitigation (Jones and O'Neill 2016). Additionally, an event is here defined as a heat extreme when the daily maximum temperature exceeded the threshold corresponding to the 90th percentile of the baseline, in accordance with the TX90p index recommended by the ETCCDI (Batibeniz et al 2020, Iyakaremye et al 2021).

The 'time sampling' approach was employed to define the GWLs, since this could eliminate interannual variability and reduce uncertainties from the different models (Vautard et al 2014, King et al 2017, Ge et al 2019b). The pre–industrial reference period is defined as years of 1881–1910, which is available from all CMIP6 historical simulations. In order to get stable climatology states, the 30 year averaged window of the running mean which reached 2.0 °C and 3.0 °C higher than the pre–industrial baseline is identified as the corresponding GWL. According to this approach, seven models under SSP2-4.5 and SSP5-8.5 scenarios reached the 2.0 °C and 3.0 °C thresholds, which together comprise 14 ensemble members for each GWL. The spread of the years covered by the participating outputs is shown in figure 1.

Zoom In Zoom Out Reset image size

Here, the relative root mean squared error (RMSE') is applied to evaluate the empirical reliability and the robustness of the model performance (Sillmann et al 2013a, Zhou et al 2014, Zhu et al 2020a)

$$\begin{equation}{\text{RMSE}}{^{^{\prime}}} = \frac{{{\text{RMSE}} - {\text{RMS}}{{\text{E}}_{{\text{Median}}}}}}{{{\text{RMS}}{{\text{E}}_{{\text{Median}}}}}}.\end{equation} \tag{ 1 }$$

where ${\text{RMS}}{{\text{E}}_{{\text{Median}}}}$ is the ensemble median of RMSEs for all selected models. In general, a better (worse) performance than the majority of the model was represented as negative (positive) RMSE'.

To assess the interannual variability of the models relative to the observations, the interannual variability skill score (IVS, Gleckler et al 2008, Yang et al 2021) is used and calculated as

$$\begin{equation}{\text{IVS}} = {\left( {\frac{{{\text{SDm}}}}{{{\text{SDo}}}} - \frac{{{\text{SDo}}}}{{{\text{SDm}}}}} \right)^2}\end{equation} \tag{ 2 }$$

where SDm and SDo are the standard deviations of the model simulations and observations, respectively. IVS is a symmetric variability statistic to quantitatively estimate the interannual variation, with smaller value indicates better agreement with the observational data. The CMIP6 model outputs are interpolated onto a common horizontal resolution of 1° × 1° by bilinear interpolation to keep consistency with the SA-OBSv2.0 observations.

Figure 2 presents the statistical metrics of the temperature extreme simulations in selected models relative to the SA-OBSv2.0 observations. It can be seen that the models vary considerably in simulating temperature extremes. Most of the models show relatively better modeling performance, especially EC-Earth3 and IPSL-CM6A-LR. Their correlations range among 0.5–0.9 and the ratios of standard deviations (RSD) are generally close to 1 (see text S1 in the supplementary material for equations). However, relatively 'weak' performance occurs in simulation from CMCC-CM2-SR5, with positive RMSE' values for all indices except TX90p. Overall, lighter colors represent a better model performance. The ensemble medians with lighter colors are superior to each single model in each respect, the IVS values are below 1.5 for all indices (figure 2, last row). This suggests that the ensemble median can reasonably represent the results of the future projections, which reduce the uncertainties of the participating models to a great extent and is therefore determined to represent the projected changes in the following study.

Zoom In Zoom Out Reset image size

Population exposure to heat extremes is calculated by multiplying the projected frequency of warm days (TX90p) in each grid point and the projected population for each corresponding point (Liu et al 2017b, Chen and Sun 2019). Here, we estimate the population exposure change following the approach of Jones et al (2015), and typically decomposed into three terms (equation (3))

$$\begin{equation}\Delta {\text{E }} = {\text{ }}{{\text{P}}_1} \times \Delta {\text{C }} + {\text{ }}{{\text{C}}_1} \times \Delta {\text{P }} + \Delta {\text{C}} \times \Delta {\text{P}}.\end{equation} \tag{ 3 }$$

where ΔE is the total population exposure changes, C₁ and P₁ are the heat extremes and population of the baseline, respectively. ΔC and ΔP are the changes in heat extremes and the population in the warmer world relative to the base time. Hence, P₁ × ΔC, C₁ × ΔP and ΔC × ΔP represent the climate effect, population effect and their interaction effect on exposure change.

Finally, the avoided impacts (AI) of exposure change due to the lower warming are calculated as follows:

$$\begin{equation}{\text{AI}} = \frac{{{C_{3.0}} - {C_{2.0}}}}{{{C_{3.0}}}} \times 100{{\% }}\end{equation} \tag{ 4 }$$

where C_2.0 and C_3.0 are the changes of heat extremes (represented by TX90p) and population exposure under the 2.0 °C and 3.0 °C warmer climates, respectively. Here, the two-tailed Student's t-test is employed to verify all statistical significances at the 95% confidence level.

Future changes in temperature extremes over SEA at 2.0 °C and 3.0 °C GWLs are presented in figure 3, respectively. Temperature extremes are projected to show a general increase over SEA at different GWLs, except for daily temperature range (DTR). These projected changes are substantially greater in some regions over SEA, with disproportionate spatial magnitudes and extents. It is noteworthy that more significant temperature extremes can be found at the 3.0 °C GWL than at the 2.0 °C GWL. The three indices of warm spell duration index (WSDI), TX90p and TN90p (representing the warm spell duration index, number of warm days and number of warm nights, respectively; see table 1) are projected to intensify significantly over Maritime Continent at both GWLs. They also share similar spatial distributions, which show that the increases over the Indochina Peninsula (ICP) are smaller than those over the Maritime Continent. Future changes in WSDI at both GWLs indicate longer warm spell duration would be existed over SEA. Over the Maritime Continent, a prominent amplification of WSDI by 43 days would exist at the 2.0 °C GWL, which would further increase by 18 days and especially even reach 90 days in some areas at the 3.0 °C GWL. For TX90p, the largest increase occurs across the PH, and with an increase of approximately 40 days in response to a warmer future. Higher TX90p and TN90p imply that intensified temperature extremes would occur more frequently and also that the change in nighttime warming will slightly exceed that of daytime. The increases in warm nights could be attributed to the high temperature and wet climate at nighttime, which tend to pose more heat stress (Li et al 2018, Liao et al 2018, Ullah et al 2019). An increasing probability of adverse impacts exert on public health, especially for those countries with dense coastal population. These indices of relative thresholds can capture warm weather under various climatological conditions (Radinović and Ćurić 2012, Sillmann et al 2013b).

Zoom In Zoom Out Reset image size

On the other hand, the projections show anomalous decreases of DTR that occur in the central areas of the ICP at both GWLs, which could be mainly attributed to inconsistent rapid increases in maximum and minimum temperatures (Horton 1995, Caesar et al 2006, Martínez et al 2010, Yao et al 2013). This also suggests that the decreased DTR are closely associated with the increasing number of warm days (TX90p) and warm nights (TN90p), confirming a potential risk of aggravated heat extremes under the acceleration of global warming. This result is also consistent with our conclusions from CORDEX model ensembles (Zhu et al 2020a).

Changes in temperature extremes and population combine to drive exposure. In this section, we estimate the future changes in population exposure over SEA and the subregions at two GWLs, as shown in figure 4. Generally, changes in projected exposure show a distinctly great increase in different GWLs. Notably, the ensemble medians of the total projected exposure changes (ΔE) exhibit a particularly large increase at the 3.0 °C GWL, suggesting a higher population exposure will be established at future warmer world. For the SEA region as a whole (figure 4(f)), the median of the total exposure to heat extremes (TX90p) increases by 205% under the 2.0 °C GWL while that value rises to 337% under the 3.0 °C GWL. For the subregions, striking increases in exposure are projected with magnitudes of 549% and 1094% over the PH at the 2.0 °C and 3.0 °C GWLs, respectively. Over KAL, projected exposure changes increase by 316% (513%) at the 2.0 °C GWL (3.0 °C GWL). The total exposure changes become moderate in the ICP compared with other regions. If temperatures reach 3.0 °C of warming, the ICP has an increase of 202% in exposure changes. Although the increasing magnitude is smaller than in other regions, the absolute change under the two GWLs is still dramatic, which emphasizes the necessity of limiting global warming. In general, these results imply that the risks of population exposure would be further exacerbated over the entire SEA under the higher warming scenarios.

Zoom In Zoom Out Reset image size

To quantify the relative importance of the different drivers to population exposure, we further used equation (3) to estimate the total exposure change and its components (including the climate change component, P₁ × ΔC, the population change component, C₁ × ΔP, and the interaction component, ΔC × ΔP) for different regions at 2.0 °C and 3.0 °C GWLs (figure 5). Under the 3.0 °C GWL, the tendency of exposure change is drastically increased compared with the 2.0 °C GWL, reaching more than 1200 million person-days over SEA, equivalent to a threefold increase relative to the reference period of 1985–2014. Overall, the increase in population exposure over SEA is largely attributed to the climate change component, which accounting for 590 million person-days, while the population and their interaction effects contributed to an increase of 177 million person-days and 434 million person-days, respectively.

Zoom In Zoom Out Reset image size

At the regional scale, the climate change component also has more far–reaching influence on the increase in exposure change, as substantially greater exposure always contributes by climate change than population change, particular in the ICP and SUM. For example, the increased exposure in SUM is primarily from the contribution of climate effect, which accounts for 66 million person-days, followed by the attributable portion from interaction effects for 33 million person-days. The population effect can be observed a relatively small proportion in increased exposure at a warmer climate, accounting for approximately 22 million person-days. In other words, 54% of the increase is induced by the climate change component.

In contrast, the intensified heat extremes over the KAL have relatively moderate impact on local exposure as a result of the low population density. The exposure increases by 30 million person-days in the future (at 3.0 °C GWL), exceeding fivefold with respect to the current climate. However, greater impacts occur over PH from the interaction component than other two components. The increase in exposure goes from around 253 million person-days at 2.0 °C GWL to 470 million person-days at 3.0 °C GWL. This is largely due to the widespread increase in heat extremes coinciding with higher population density. Under such warming, an increase caused by the interaction component strengthens over the PH, to 102 million person-days and 267 million person-days, respectively, under the two scenarios.

In brief, for both SEA as a whole and the subregions, increased population exposure is to a great extent associated with climate change in the future. Therefore, the explicit target of restricting climate warming is undoubtedly more beneficial than other climate events and avoided impacts and climate actions should be encouraged to promote such restricting regardless of the mitigation and adaptation strategies in this region.

The future projection of the CMIP6 ensembles in climate simulating indicate that the occurrences of heat extremes are expected to intensify at a rate that is unprecedented in the majority regions over SEA as global warming continuous (Kim et al 2020, Chen et al 2020b). If the Paris Agreement goal of limiting the rise below 2.0 °C is met, some temperature extreme events will not be tested. We therefore estimate the avoided impacts of heat extremes and population exposure in response to 1.0 °C limited warming (figure 6). Our results show that nearly all regions would see considerable avoided impacts, although the magnitudes differ among regions. For the whole of SEA, the increase of heat extremes (TX90p, figure 6(a)) can be avoided by approximately 33% in the situation of the 1.0 °C limiting warming, and the increase of population exposure is also avoided by 39% when compared with that under the 3.0 °C GWL. It is notable that the most pronounced changes can be found in the regions of the ICP and PH, where limited 1.0 °C warming will help to avoid increase in heat extremes by at least 40%, and the increase of population exposure can be avoided by 50% in the future. The change tendencies of heat extremes and population exposure are consistent, which also illustrates that climate change is the essential role in driving the increased population exposure. Over the regions of SUL, population exposure presents a moderate reduction in response to the 1.0 °C limited warming, of approximately 29%. Generally, population exposure would be drastically lowered across SEA when GMST increase is controlled below 2.0 °C.

Zoom In Zoom Out Reset image size

Surprisingly, regarding the decomposed components of exposure, the avoided impacts of population exposure caused by the interaction component over SEA are higher than for the single climate effect (figure 6(b)). The increase of exposure in PH caused by the interaction component (ΔC × ΔP) would be significantly decreased by 62%, while only 38% by the climate component (P₁ × ΔC). Over SEA, the interaction component avoids the impacts by more than 50%. These results indicate that the combined effects of increased population concurrent with climate change would induce more risks of population exposure in a warmer future. More efforts to control rapid global warming and increasing populations would lead to more impacts being avoided in the future.