The recent surge in global temperatures, with multiple all-time monthly records getting broken on a recurrent basis since mid-2023 and widespread heatwaves all over the world, has sparked several new discussions and studies on extreme heat and its causes(1–5). Urbanization further amplifies local temperatures, exacerbating heat impacts on a disproportionate number of people due to the high population densities in cities(6–8). Consequently, there have also been multiple recent assessments and deliberations on the sources and consequences of urban heat from both the public health and energy demand perspectives(9–12).
Since cities are highly heterogeneous, their spatial variabilities modulate the intra-urban distribution of heat. It is somewhat standard in scientific studies, media commentaries, and news stories to discuss intra-urban heat exposure in terms of population-level disparities(13–18). However, almost all multi-city assessments that have provided strong evidence of pervasive income-based disparities in heat hazard and vulnerability focus on U.S. cities(19–27). The handful of studies on cities outside the U.S. have found less clear intra-urban disparities in heat hazard and exposure and its causes(28–32). The availability of nationwide fine-scale census estimates, common in the U.S. and other countries in the Global North, makes it relatively easy to conduct studies on intra-urban environmental disparities for these regions. In contrast, similar consistent datasets that can resolve the income and wealth distributions within cities are rare for the Global South(33–36). Since urban population distributions result from present cultural norms as well as historical policies(37–39), which can be distinct for different countries and regions, it is unknown whether the typical patterns of within-city heat hazard found in the U.S. context will translate to the many rapidly urbanizing and densely populated countries in the Global South.
Here, we use a suite of spatially continuous global datasets to examine distributions of heat within Global South cities. The results show that, unlike U.S. cities, where lower income neighborhoods usually showing higher heat hazards, Global South cities generally show higher heat hazard in wealthier neighborhoods. However, the direction of intra-urban vegetation disparities, which modulate heat disparities, is more ambiguous and depends on vegetation category and metric of greenness. This opposite direction of observed intra-urban heat disparity by income in the Global South is due to the completely different distributions of population groups within these cities compared to the typical patterns seen for U.S. cities. The study shows that income-based inequality in heat hazard is not a universal outcome of urban development but lies at the nexus of the physical and socio-demographic aspects of urbanization, the latter molded by distinct cultural histories and locational preferences within cities.
Disparities in urban heat hazard
We generate a global dataset of urban clusters (n = 2440; cities henceforth) based on the Global Human Settlement Layer(40) and combine that with the Relative Wealth Index (RWI) dataset(41) to estimate potential disparities in heat hazard for Global South cities. To get more robust statistical associations, we only choose the clusters that cover at least 10 RWI grids. The final 924 clusters used for the Global South, housing over 1.1 billion people, are then further grouped into sub-regions (Fig. 1a) and we estimate various measures of temperature across data sources and seasons for all the RWI grids within these clusters. We focus on the 10-year average maximum summer air temperature (Tmax henceforth) for the main analysis.
Most cities in the Global South show positive associations between RWI and Tmax i.e. urban heat hazard is less in poorer neighborhoods (Figs. 1b, 1d), which we call a pro-poor distribution since higher temperature is usually an environmental burden during summer. Of the 924 cities, 72.6% (671 clusters) show this positive association, with 40.2% (n = 270) of these associations being statistically significant (p < 0.05; Fig. 1b). Among the sub-regions, this positive association is seen everywhere other than in Sub-Saharan Africa (Fig. 1c). This is probably related to the unique urban warming signal for arid cities(42). This pro-poor distribution of urban heat hazard is seen for various sources of temperature data (Fig. 1d), including satellite-derived land surface temperature (LST), for both daytime and nighttime, and for annual average in addition to the summer values described earlier. These results are in stark contrast to heat hazard distributions for U.S. cities (n = 492), where it is warmer in poorer neighborhoods (negative associations between income and Tmax) in over 88% of cases (Fig. S1a).
Inequalities in urban tree cover
Urban vegetation can reduce local temperatures, and the disparity in heat hazard in U.S. cities usually parallels disparities in urban vegetation(21, 24, 25), with wealthier neighborhoods generally having more tree cover than poorer ones. To examine if this pattern of urban development is cross-cultural, we estimate fraction of tree cover from a global 10 m land cover dataset(43) for all the cities. Associations between tree cover and wealth is more complicated in the Global South, with East Asia & Pacific, Sub-Saharan Africa, and Latin America & Caribbean all showing generally pro-poor distributions i.e. higher tree fraction in poorer neighborhoods (Fig. 2b). The differences across regions suggest that greener neighborhoods may not always get a real estate premium in the Global South, while this is quite common across U.S. cities (Fig. S1b). Overall, slightly over half (52.4%) of Global South cities show this pro-poor distribution of tree fraction (Fig. 2a).
That one can simultaneously have pro-poor distributions of heat hazard, but pro-wealthy distributions of tree fraction (for instance, see South Asia) is because tree cover is only a subset of total vegetation and local temperatures also depend on other factors. We check consistencies of these distributions across different greenness metrics and for different subtypes of vegetation, including both tree cover and grass (Fig. S1a). When we consider the sum of grass and tree fraction, with grass fraction including both parks and various informal non-tree green spaces, the percentage of cities with pro-poor distribution goes up to 65.5% (from 52.4% for only tree fraction; Fig. 2a). The occurrence of pro-poor distributions increases further to around 81.1% when a bulk proxy of photosynthetically active vegetation (Enhanced Vegetation Index) is used. In line with cautionary discussions in past studies that mainly focused on single metrics of greenness(36, 39, 44), this demonstrates that the revealed preference, as measured by the bulk outcome based on multi-city evidence, of wealthier populations for urban green space in the Global South depends strongly on the type of vegetation, with tree cover showing the most ambiguity in direction.
Population and heat distributions
Temperature distributions within cities are strongly controlled by local land use land cover, from cooler than average microclimate of urban parks to local hotspots over impervious surfaces with higher anthropogenic activities, such as in industrial and commercial areas and over parking lots and other built-up structures(45). There are many combinations of these surfaces within cities due to their morphological and biophysical heterogeneity. While the association between local microclimate and surface characteristics is constrained by physical processes, the distribution of urban populations, especially of different income groups, is a function of cultural and political history of the city or region in question. For U.S. cities, historical policies, like redlining, and population redistribution trends, such as urban/white flight, influence these distributions(37, 38). A combination of these physical and socioeconomic distributions result in the income-based disparities in both tree cover and heat hazard in cities(19, 22, 46).
Countries in the Global South have their own distinct (separate from the U.S. and separate from each other) cultural and policy influences on urban population distributions(33, 39, 47). To better understand these distributions, we examine wealth and heat hazard as functions of distance from the centroid of each urban area (Fig. 3). In parallel, we do a similar analysis for all census tracts for each urbanized area in the U.S (Fig. 3h). Since cities can be of quite different sizes with various wealth/income ranges, we scale all variables to lie within 0 and 1 for each city. For both the Global South and the U.S., Tmax generally peaks close to the centroid (Figs. 3a to 3h), albeit much closer for the U.S. (Fig. 3h) than for the Global South cities (Fig. 3a), and then decreases as we move towards the edge of the urban area. However, for the wealth/income distributions, we see clear differences, with the scaled income rising as one moves to the edge of the city in the U.S., but generally following the distribution of scaled Tmax in the Global South. The former (Fig. 3h) is expected based on wealthier urban residents moving to the suburbs in the U.S., reflecting neighborhood and housing (“white picket fence”) preferences (38, 48). The latter (Fig. 3a) is why we see more pro-poor distribution in urban heat hazard in the Global South. While there are differences by region, the general trends are consistently in stark contrast to the distributions seen for the U.S.
A couple of examples are shown for both the Global South and the U.S. to illustrate the differences in the spatial distributions of wealth/income and Tmax (Fig. 4). For both Kolkata in India (Figs. 4a, 4c) and Jakarta in Indonesia (Fig. 4b, 4d), greater RWI values are found in the middle of the urban area, where the Tmax is also generally highest. On the other hand, for Denver and Baltimore (Figs. 4e, 4f, 4g, 4h), median income is higher at the edge of the urban area, where Tmax is lower. Note that we use urban clusters (cities) generated from population-based estimates of urbanization for these analyses, in line with multiple similar studies in the U.S. using census-defined urbanized areas(20, 22, 24). We replicate all these results by generating clusters representing physical urbanization (see Methods), though this reduces our sample size significantly (Fig. S2a). However, all the results are consistent for this second set of urban clusters (Figs S2, S3, S4), but with somewhat weaker sensitivity due to the smaller sample size and since population-based thresholds of urbanization no not map linearly to physical urbanization cross-culturally. While examining the spatial variability of all individual urban clusters is outside the scope of this study, the Tmax and RWI distributions for all the cities considered here can be explored through this web application: https://ee-tc25.projects.earthengine.app/view/globalsouthheat
Implications for intra-urban heat risk
Through a comprehensive multi-city assessment, we demonstrate generally pro-poor distributions of urban heat hazard for Global South cities (Fig. 1), which is opposite of the results seen for U.S. cities(20, 22–24). Global South cities also show more variability in these associations, with ‘only’ 72.6% of them showing pro-poor distribution of heat hazard compared to 88% of U.S. cities showing pro-wealthy distributions of the same. This is expected given the wider range of historical backgrounds and distinct cultures for Global South cities. Our results suggest that income-based disparities in heat hazard being disproportionately higher in poorer neighborhoods is not a universal consequence of urbanization. Rather, U.S. urban population distributions diverge from those in the Global South, making results from U.S. studies not necessarily generalizable everywhere. As such, a more nuanced discussion about urban heat, one that does not implicitly assume the U.S. urban environmental outcomes to be the norm, an ongoing concern in several scientific fields involving human/social components(49), is needed.
There is also a need for more careful framing around urban heat risk, with relevance for developing local-scale heat mitigation and adaptation policies. For this, intra-urban variability of heat hazard is more informative than comparing city-averaged values between Global North and Global South cities since urban climate policies are not decided between countries, but enacted by local governments(50). Beyond heat hazard, our results also extend to heat exposure, since the wealthier areas are also more population dense (Fig. S6), with population-weighted urban heat exposure increasing with higher RWI in the Global South (Fig. S7). This is a simplistic yet common approach for representing heat exposure in the literature(9, 13, 22), with urban residents assumed to be exposed to outdoor conditions during the warmest times of the day. In reality, one would expect a large fraction of them to be indoors during these periods. Since poorer residents are more likely to be employed in blue-collar industries with greater likelihood of outdoor exposure, using a simple population-weighted estimate by neighborhood or grid would underestimate heat exposure for these sub-populations. Moreover, many urban residents in the Global South live in slums or other informal settlements, where the coupling between outdoor and indoor condition is stronger(51) and there is little option for air conditioning. This access to air conditioning, which normally leads to almost complete decoupling between indoor and outdoor environments, contributes to “vulnerability”, the final component of environmental “risk” other than “hazard” and “exposure”(35). Even though the Global South still has extremely low air conditioning penetration(52), wealthier urban residents are still more likely to have air conditioning. Greater air conditioning penetration in richer neighborhoods has been seen in the U.S.(27) and is expected to be cross-culturally true, exacerbating impacts of heat exposure on poorer urban populations. The distinction between overall heat risk and heat hazard is critical for making decision on heat mitigation versus adaptation in the Global South. If an equitable solution to urban heat is a policy priority and poorer neighborhoods are already cooler, with possibly more vegetation (than in wealthier neighborhoods), adding more vegetation, which has been a common urban heat mitigation recommendation in the US(24), may not have the same potency, and would primarily address outdoor heat hazard. Given the stronger coupling between outdoor and indoor conditions in informal and non-air-conditioned dwellings, strategies that lead to equitable adaptation, such as providing financial incentives to poorer and more vulnerable communities to install air conditioners or improve building insulation, may be more effective for reducing overall heat risk.
We are reasonably confident that the RWI dataset can capture at least the direction of the income/wealth distributions within these cities (Fig. S8). Moreover, the patterns found here are generally consistent with the handful of previous studies on heat and vegetation disparity for Global South cities (see Methods). However, given the coarse nature of the dataset, we cannot resolve important fine-scale variability and expect errors for individual cities. Therefore, we recommend more studies outside the Global North to better understand variations in intra-urban heat and vegetation distributions and particularly their relation to income and other socioeconomic characteristics, which can be distinct across countries, regions, and cities. Given the dominant narrative about urban heat disparities driven by U.S.-centric evidence and patterns(13, 16, 18, 20), it is possible that results showing higher heat hazard in wealthier urban neighborhoods are systematically not published since they are not considered as relevant. We hope that our results also encourage these kinds of studies so as to avoid sampling bias in the scientific literature and lead to a more complete understanding of urban climate and its many complexities across the globe.