Future increase in elderly heat-related mortality of a rapidly growing Asian megacity

Urban dwellers are at risk of heat-related mortality in the onset of climate change. In this study, future changes in heat-related mortality of elderly citizens were estimated while considering the combined effects of spatially-varying megacity’s population growth, urbanization, and climate change. The target area is the Jakarta metropolitan area of Indonesia, a rapidly developing tropical country. 1.2 × 1.2 km2 daily maximum temperatures were acquired from weather model outputs for the August months from 2006 to 2015 (present 2010s) and 2046 to 2055 (future 2050s considering pseudo-global warming of RCP2.6 and RCP8.5). The weather model considers population-induced spatial changes in urban morphology and anthropogenic heating distribution. Present and future heat-related mortality was mapped out based on the simulated daily maximum temperatures. The August total number of heat-related elderly deaths in Jakarta will drastically increase by 12~15 times in the 2050s compared to 2010s because of population aging and rising daytime temperatures under “compact city” and “business-as-usual” scenarios. Meanwhile, mitigating climate change (RCP 2.6) could reduce the August elderly mortality count by up to 17.34%. The downwind areas of the densest city core and the coastal areas of Jakarta should be avoided by elderly citizens during the daytime.

Comparison between simulated (nearest grid) and observed 3-hour near-surface temperatures before (Left) and after adjustment or bias-correction (Right). The colored lines correspond to a least-squares regression fitting between the horizontal and vertical values. The observation stations are denoted TPR, CGK, and KMY.

Figure 1b
Histogram of maximum daily temperatures throughout the simulated domain before (unadjusted) and after (adjusted) bias-correction for the August months of 2006 (2046) to 2015 (2055). All cases shown.

Figure 1c
August-average of daily mean (a,b,c) and daily (d,e,f) maximum unadjusted nearsurface temperature at urban grids of Jakarta metropolitan area for PRESENT (a,d), RCP26CC (b,e), and RCP85BaU (c,f) cases. (This figure is the same as Fig. 4 of the manuscript but using modeled temperature without bias-correction.

Detailed explanation and assumptions of the derivation of relative risks (RR) and heat-related mortality
In the manuscript, the derivation and definition of heat-related mortality relative risk RR mentioned in the manuscript are explained further in this section.
According to the textbook, "A Dictionary of Epidemiology (6 th edition) by Miguel Porta, Oxford University Press", relative risk (RR) is defined by the ratio of two risks, usually of exposed and not exposed. Mathematically, RR is obtained by dividing the incidence rate (i.e. probability for a specific incident to happen) of an exposed population with the incidence rate of an unexposed population. In terms of mortality, the specific incident is crude death or deaths caused by all causes. Since the concern of this study is to estimate heat-related mortality counts or heat-related mortality RR, the parameter to which the population is exposed refers to ambient temperature.
The value of heat-related mortality RR for a certain location depends on the difference between the location's ambient temperature and a certain temperature threshold. This threshold is referred to as optimum temperature (OT) in the manuscript and varies with geography or background climate. Multiple heat-related studies (stated in the manuscript) have confirmed that mortality increases when ambient temperatures exceed OT. RR is equivalent to 1.0 when the ambient temperature is equivalent to OT and RR increases depending on how much the ambient temperature exceeds OT. Eq. 4 in the manuscript is an example of how heat-related mortality RR for senior citizens (age 65 years and above) is quantified as a function of daytime maximum of ambient temperature and OT.

Fig. 2a
Age-specific death rate (ASDR) for senior citizens age 65 years and above. Data derived from Global Health Observatory data repository maintained by the World Health Organization and World Population Prospects by the United Nations. Accessed: January, 2020 After determining the heat-related mortality RR (eq. 4), the number of heat-related deaths of senior citizens is estimated. Initially, demographic databases (e.g. World Population Prospects by the United Nations, Global Health Observatory data repository by the World Health Organization) are inspected to obtain age-specific death counts in Indonesia (Table 1). Age-specific death rate (ASDR), defined as the number of deaths for a specific age group divided by the total population of the same age group, was estimated for senior citizens in Indonesia (population count with age greater or equal to 65). Fig. 2a shows a time-series of the ASDR of senior citizens estimated from the ASDR and population by age datasets for the years 2000 to 2016. The average and standard deviation for the period are 0.0709 and ± 0.000750, respectively. Given no clear trend and significant changes in the ASDR throughout this period, the average ASDR is then assumed for the present and future cases in the study. ASDR is also the crude death rate cdr used in the study. where exposure refers to ambient temperature values exceeding the OT. Combining eq. 2a and 2b, an expression for Z can be derived as follows, Using eq. 2c, the earlier derived cdr and eq. 4 in the manuscript, 3.86 out of 100 senior citizens are at risk of dying when a day experiences a maximum temperature of 33°. Given that Pop and cdr are mid-year estimates, D for the August month while utilizing the mean of daily maximum temperature can be estimated by multiplying eq. 2c by the weighted proportion of the number of days of August to the year. The resulting D for August month will be, where DAugust and RRAugust corresponds to the heat-related death toll and relative risk of senior citizens (age 65 and above) for the month of August, respectively. RRAugust can be estimated by substituting the August mean of daily maximum temperature to eq. 4 of the manuscript.

Wind rose plots of daily mean winds
Wind field plays a significant role in the transport of scalars within the atmosphere. The afternoon winds simulated for all August months in Jakarta tend to be north-easterly. This typical circulation is influenced by the monsoonal background of the region and sea-breeze flow. In the manuscript, winds are used to mainly attribute the relatively higher heat-risk locations at the southwest area downwind from the city center of Jakarta. Windrose plots are displayed in Fig. 3a to prove the typical northwesterly flow in the afternoon of August.

Climatic and human development index comparisons between Jakarta, Indonesia and Ho Chi Minh City, Vietnam
In this work, the heat-related mortality relative risks (RR) of Ho Chi Minh City, Vietnam was used as a proxy for the target study, Jakarta, Indonesia. The reasons for this are the lack of RR studies in Jakarta and the proximity between Vietnam and Indonesia in terms of geography (continental-scale), climate, and human development conditions. Both cities are located in Southeast Asia with tropical climates, experiencing both a dry season and wet season (for Jakarta, this is called monsoonal), with no cold season. The annual mean (max) temperature for Jakarta and Ho Chi Minh City are 27.7°C (31.8°C) and 27.7°C (32.2°C), respectively. Although Ho Chi Minh City tends to have a slightly higher daytime maximum temperature than Jakarta, both cities' daytime temperature maximums between 30°C to 35°C throughout the year. For the month of August, both cities experience almost similar values of air temperature. For monthly and annual statistics of air temperature, refer to Table  4a. Given their monthly temperature variations, the continentality type of both cities are the same, hyperoceanic. The human development index (HDI) is the parameter used in this study to measure the similarities of the citizens in both Jakarta and Ho Chi Minh City in terms of life expectancy, ability to acquire knowledge, and the ability to achieve a decent standard of living. From the Human Development Report Office of the United Nations Development Programme (UNDP), HDI is defined as a composite index focusing on three basic dimensions of human development: the ability to lead a long and healthy life, measured by life expectancy at birth; the ability to acquire knowledge, measured by mean years of schooling and expected years of schooling; and the ability to achieve a decent standard of living, measured by gross national income per capita (UNDP, 2018). HDI ranges from 0.0 to 1.0, with 1.0 having the highest state of human development.
According to the 2018 statistical update of UNDP, Vietnam and Indonesia are both classified as countries with medium human development. In terms of the recent ranking, both countries share the 116 th global rank with an equivalent value of 0.694 as of 2017. Comparing HDI of the past years (Table 4b) also reveal minimal differences. Although it is ideal that a unique RR function is to be used for Jakarta, these functions are difficult to determine under the scope of this study. Meanwhile, an RR function was recently estimated for Ho Chi Minh City, which shows similarity with Jakarta in terms of climatic and human development backgrounds as shown above. Thus, for the time being until an RR becomes available for Jakarta under different research, the RR function of Ho Chi Minh City is applied for Jakarta.

Urban parameters and population datasets used in the study
In the weather modeling of Jakarta by Darmanto et al. (2019), detailed grid-scale distributions of urban parameters including building morphological statistics and anthropogenic heat emission were used as surface boundary inputs. This was estimated for all cases; which means PRESENT, RCP26CC, and RCP85BaU have different urban parametric distributions (see Fig. 1 in Darmanto et al, 2019).
Interestingly, the preparation of their urban parametric distribution requires a detailed estimation of the population density according to each scenario, be it present, or from the shared socio-economic pathway (SSP) scenario (see Supplementary A4 of Darmanto et al, 2019). In other words, the simulation cases, PRESENT (which utilizes population density from LandScan TM ), RCP26CC, and RCP85BaU, were also founded upon the spatial differences in population density. Fig. 5a which was taken from the supplementary data by Darmanto et al.