Deadly Heat Exposure in an Urbanized World

Climate-change exposes an increasing share of the world population to potentially lethal heat, a threat accentuated by rapid urbanization. Here, we project occurrence of future deadly heat for urban agglomerations around the world until 2080 by using CMIP6 climate model projections of temperature and relative humidity, urbanization prospects and GDP projections from the SSP scenarios. We show that while nearly all regions within latitudes 35°S 45°N experience an increase in days of deadly heat, Sub-Saharan Africa and Southeastern Asia are particularly exposed, a trend exacerbated by rapid urbanization. By 2080, between 2.3 (59%) (SSP1-2.6) and 3.0 (75%) (SSP5-8.5) billion urban dwellers will experience more than 30 annual days of deadly heat, including 477 (66%) 546 (77%) million in Sub-Saharan Africa and 988 (93%) 993 million (94%) in South and South-Eastern Asia. The exposure to heat is highly unequal, with some of the poorest regions affected the most. Our results imply that jointly mitigating climate change, planning for well-ventilated cities, and combating poverty to enable economic access to air conditioning is required to avert a global-scale humanitarian crisis.

stantial spread 20 . 48 The urban focus matters threefold. 49 First, more than two thirds of human- 50 ity is expected to live in cities by 2050, 51 with most urbanization expected to hap- 52 pen in the Global South 21    which increasingly often will be deadly.

185
In the low emissions scenario (SSP1-2.6), 186 warming is projected to level off in the sec-

Competing Interests
The authors declare that there are no competing interests.

Author Contributions
SL and FC conceptualized the study. CM contributed the code for determining deadly heat from climate data and assisted with its implementation as well as interpretation of the results. DR contributed to the discussion on distributional aspects. SL wrote the computer code, carried out the analyses and produced the figures. SL wrote the first draft of the manuscript, which was substantially reviewed and revised by all authors.

Data Availability
The code used for producing the research outcomes and figures in this article can be shared in an GitHub repository upon request. The climate model output used is available via the ISIMIP programme.  We use a support vector model on tas and hurs for determining heat anomalies as presented in 3 . We use a 95% margin for the SVM to select the heat anomaly as lethal.
For each city, number of deadly days are summarized per year, and the outputs reported in this paper are the 10-year rolling mean of deadly days.
We use the IPCC AR6 language for depicting multi-model uncertainty: very likely (90% -100%) and likely (66% -100%) for the multi-model central range. These are applied to the rolling 10-year mean number of deadly days.

Population Predictions
Population predictions for individual cities over several decades are to be taken with a grain of salt. Reasons are the uncertainty in growth rates, owing to both their economic and social drivers, and possible inhibitors such as climate change, lack of land mass or other local resource constraints. We here (Tab. 3) compare our results for urban population numbers with the outcome of a more refined analysis which population size for the 100 largest cities under different assumptions for urban growth rate 51 . The projections are largely in the same order of magnitude, even though some exceptions exist (Table 3) which are a result of the different methodologies: While we here assume a constant distribution among cities in one country, we do not use the urban growth rate of individual cities as done in 51 , which leads to an overestimation of the growth of large cities in our methodology.

Caveats
Further methodological challenges include the challenge to forecast individual city growth and the large, but difficult to model, contribution of the urban heat island effect on temperature.A practical solution to account for the urban heat island effect direct in climate models has recently been presented using an urban climate emulator that has originalle been included in the CESM2 climate model 25 .
The urban population projections presented here do not account for differential growth rates of different cities, which exist without doubt. Different assumptions for how the observed urban growth may continue in future has been investigated 51 , and a further way to include precise population numbers could be through the use of spatially explicit population forecasts 52 .   51 . The WUP population numbers are dveloped in that study using the urban growth rates from WUP, and there is no direct analogue in this study.