Introduction

The escalating frequent, intensity and duration of heatwaves resulting from climate warming pose significant threats to both global ecosystems and human societies, impeding sustainable development1,2,3,4. Exposure to extreme heat presents a profound risk to human health, escalating morbidity and mortality rates5,6,7. Global cities accommodate 55% of the world’s population, with an estimated 68% residing in urban areas by 20508. The urban environments and the distinct urban climates, typified by the urban heat island9,10,11, amplify the risk of heat exposure for city dwellers. Recent studies indicate an approximately 200% increase in the global urban population exposed to extreme heat from 1983 to 201612. Urgently, there is need to accurately assess population exposure to extreme heat on a global scale, formulate tailored mitigation strategies for individual cities and advance the overall livability of urban areas.

Thermal environment and population distribution constitute critical elements in evaluating the risk of extreme heat exposure among populations13. In the context of the Clinton Risk Triangle framework (Hazard-Exposure-Vulnerability)14, higher population density inherently translates to elevated “exposure”15. The existence extreme heat environments signifies the “hazard” inherent in heat risk assessment15. Typically, studies have used daytime heatwaves or nighttime heatwaves to characterize urban thermal environments and thus explore heat exposure12,16,17. However, daytime and nighttime heatwaves do not usually occur independently of each other; extreme daytime and nighttime temperatures occur simultaneously to form compound heatwaves18,19. Continuous high temperatures during the day and night aggravate the human body and pose a greater threat to human health than independent daytime heatwaves and independent nighttime heatwaves18. Therefore, we must be concerned about the population exposure to compound heatwaves20. Besides the focus on heatwave types, diverse heatwave indicators are also used to assess heat exposure. Commonly, heatwave frequency (HWF) serves as a key indicator for gauging heat exposure risk13,17. However, the HWF may not fully characterize urban heat hazards, so, in some cases, heatwave intensity has been utilized to develop comprehensive heat exposure indicators21,22.

Urbanization, characterized by the transformation of natural surfaces into impervious areas and release of anthropogenic heat23, has changed the surface energy balance24. These have given rise to higher temperatures in urban areas compared to the rural surrounding, known as the urban heat island effect11. The synergistic effect of heatwaves and urban heat islands will further exacerbate thermal threats in cities25,26. Urban heat, resulting from the interplay of climate change and urbanization (specifically, urban heat islands), plays a pivotal role in shaping heat exposure12. While previous studies have assessed the collective impact of urban warming on heat exposure12, the specific influence of urbanization-induced warming on heat exposure remains inadequately explored and is often overlooked in heat exposure assessments. Accurate prediction of urban climate stands as a fundamental prerequisite for evaluating urban heat exposure12. However, relying solely on global climate model data for assessing historical and future population exposure to heatwaves has limitations, as it often lacks representation of urbanization effects, leading to an underestimation of urban heatwave risks27. Some studies have sought to characterize the impact of total urban warming on urban heat exposure using reanalysis data12. Nevertheless, the coarse spatial resolution of both global climate model data and reanalysis data does not distinguish between urban and rural areas and therefore it fails to independently discern urbanization-induced warming. Although meteorological station data can shed light on the influence of urbanization on heatwaves28,29, its limited spatial distribution restricts global urban assessments. In this context, satellite-based surface temperature data offering high spatial and temporal resolution, emerge as a valuable tool for capturing temperature variations attributable to urbanization effects10,30. Land surface temperatures and their inverse counterparts, near-surface air temperatures, find widespread application in heat risk assessments15,31,32,33. For example, inequalities in urban heat exposure32,34 and mitigation strategies31 were effectively assessed based on land surface temperature. Estoque et al. notably developed a heat hazard index rooted in land surface temperatures to evaluate heat-related health risks in Philippine cities15. Near-surface air temperatures, derived from remotely sensed temperature, have also been employed to establish global urban heat and cold rankings35 and assess the interannual variation of canopy urban heat island33. Moreover, while prior studies have underscored the unequal exposure of global urban populations to extreme heat risk17,22,36, and emphasized the higher heat exposure trends in the global South cities compared to global North cities17, the influence of urbanization-induced warming on global urban heat exposure inequality has remained inadequately explored. And, its manifestation at the suburban scale remains unclear. Addressing this gap is of urgent importance, as understanding the heat exposure inequalities among global cities is pivotal for formulation adaptation and mitigation strategies that prioritize environmental justice on a global scale.

Since compound heatwaves are more hazardous to human health, we focus on investigating urban population exposure to compound heatwaves with the aim of assessing the impact of urbanization-induced warming on heat exposure and on the inequality of heat exposure between cities in the global South and North. Our study is based on a near-surface air temperature dataset derived from remotely sensed surface temperature, enabling the incorporation of urbanization effects37. The heat exposure index is developed by coupling the population density and the heat index. Due to the diversity of heatwave characteristics, three heat indices are used, namely HWF, integration of HWF, heatwave magnitude (HWM) and heatwave duration (HWD), and cumulative heat (CH). We find that the urbanization-induced increase in compound heatwaves is observed in more than 90% of cities globally, and that the amplification effect is stronger in high urbanization areas (5.89 × 10–3) than in low urbanization areas (2.46 × 10–3) (Fig. 1). The results underscore the stronger urbanization-induced warming in the global North cities than that in the global South cities (6.97 × 10–3 vs. 2.78 × 10–3, Fig. 1). Furthermore, we compare the difference in heat exposure without considering urbanization-induced warming (hypothetical scenario) versus considering urbanization-induced warming (real scenario). Heat exposures in the global South cities are generally higher than those in the global North cities under the hypothetical scenarios (Fig. 2 and Supplementary Fig. 4). One key finding of our analysis is that in real scenarios, the difference in heat exposure between cities in the global North and South is typically narrowed, due to stronger urban warming effects in the global North cities (Fig. 2 and Supplementary Fig. 4). Finally, the urban-rural differences in heat exposure are attributed to the independent role of urban warming, the independent role of population growth, and the interaction of the two. We find that the contribution of urbanization-induced warming is stronger in the global North cities, while the independent role of population is stronger in the global South cities (Fig. 3 and Supplementary Fig. 12). These findings emphasize the complexity of urban heat exposure dynamics and highlight the importance of considering urbanization-induced warming.

Fig. 1: Mean urban-rural heatwave differences from 2003 to 2019.
figure 1

Spatial distribution of urban-rural heatwave differences in urban areas (a), high urbanization areas (c), and low urbanization areas (e). Urban-rural heatwave differences in Global North and Global South cities in urban areas (b), high urbanization areas (d), and low urbanization areas (f). ‘GNc’ represents the Global North cities and ‘GSc’ represents the Global South cities. Box plots represent the interquartile range (IQR) as the box, median as a horizontal line within the box, mean as a point within the box, and 1.5× IQR as the whiskers. Outliers are omitted for clarity. Mann–Whitney U tests were conducted to determine significance between the global North and South cities. ‘***’ indicates significant differences at the 0.001 levels.

Full size image
Fig. 2: Differences in heat exposure index (HEI) between real and hypothetical scenarios.
figure 2

a Spatial distribution. b Differences in HEI between the global North and South. Heat exposure levels of global North and South cities are represented in red bar graph and blue bar graph, respectively, when considering urbanization-induced warming (real scenario). Heat exposure levels of global North and South cities without considering urbanization-induced warming (hypothetical scenario) are shown in gray bar graph. Error bars indicate standard deviation. The red and black dots indicate the average HEI difference between the global North and global South cities under the real and hypothetical scenarios, respectively. Mann–Whitney U tests were performed to determine significance between global North and South cities. ‘***’ indicates significant differences at the 0.001 levels. c Latitudinal characteristics of HEI differences. The solid line and shadow represent mean and standard deviation, respectively.

Full size image
Fig. 3: The independent contribution of urbanization-induced warming to heat exposure.
figure 3

Contribution of urbanization-induced warming to heat exposure index in urban areas (a, b), high urbanization areas (c, d), and low urbanization areas (e, f). ‘GNc’ represents the Global North cities and ‘GSc’ represents the Global South cities. Box plots represent the interquartile range (IQR) as the box, median as a horizontal line within the box, mean as a point within the box, and 1.5× IQR as the whiskers. Outliers are omitted for clarity.

Full size image

Results

Global urban population exposure to compound heatwaves

We conducted a comprehensive assessment of heat exposure in 1028 cities globally. The results revealed that more than half the cities globally pose a low exposure risk (53.9%). Conversely, a low percentage of cities faced a high exposure risk (6.0%) and a moderate exposure risk (40.2%) (Fig. 4a, b). We further compared the heat exposure between the high urbanization areas and the low urbanization areas and found that the former had a higher heat exposure (Fig. 4c, e). The percentage of highly exposed cities in high urbanization areas was 8.9%, compared with 1.4% in low urbanization areas (Fig. 4d, f). In time, there was a significant increase in the global urban population exposure to the compound heatwave. The average global urban heat exposure risk increased by 11.8% in 2020 compared to 2005 (Supplementary Fig. 1a). The increase in the heat exposure risk from 2005 to 2020 in the global average high urbanization areas (13.0%) was larger than that in the low urbanization areas (8.7%) (Supplementary Fig. 1b, c).

Fig. 4: Global urban population exposure to compound heatwaves.
figure 4

Spatial distribution of heat exposure in urban areas (a), high urbanization areas (c), and low urbanization areas (e). Heat exposure levels are categorized into 3 classes using the Jenks Break method. Percentage distribution of different heat exposure levels in urban areas (b), high urbanization areas (d), and low urbanization areas (f).

Full size image

Inequitable urbanization-induced warming in the global North and South cities

Focusing on the impact of urbanization-induced warming on heat exposure, we observed widespread amplification of compound heatwaves in urban environments (Fig. 1). Among the 1028 cities studied, a significant majority experienced amplified compound heatwave (94.5%). Further analysis at the suburban scale demonstrated a greater amplification effect of urbanization on heatwaves in high urbanization areas compared to low urbanization areas (Fig. 1c, e). In regions characterized by high urbanization, the amplification effect was 5.89 × 10–3. Conversely, in low urbanization areas, the amplification effect was comparatively lower (2.46 × 10–3), indicating the pivotal role of urbanization in exacerbating compound heatwave levels and subsequently amplifying urban heat exposure risk. Significantly differing levels of amplification in urbanization-induced compound heatwaves were observed between cities in the global North and South, with the effect being notably stronger in the global North (6.97 × 10–3 vs. 2.78 × 10–3, Fig. 1b). These disparities were evident in both high- and low-urbanization areas, although the amplification effect of urbanization-induced compound heatwaves was more pronounced in areas characterized by high urbanization compared to those with low urbanization (Fig. 1d, f). We tested the effects of different temperature thresholds and different heatwave indicators on urban-rural heatwave differences. For three temperature thresholds (90%, 95% and 98%) and three heatwave metrics (see Methods), the more intense amplification of compound heatwaves caused by urbanization in Global North cities was observed (Supplementary Figs. 2, 3).

Impact of urbanization-induced warming on heat exposure inequalities in the global North and South cities

Cities in the Global South face a higher risk of heat exposure. Specifically, among the 46.2% of cities categorized as high- and moderate-exposure, only 16.5% were in the global North, while a significant majority, 29.7%, were situated in the global South (Fig. 4b). Similarly, in high (low) urbanization areas, of the 53.8% (20.0%) of high- and moderate-exposure cities, merely 20.1% (6.6%) were in the global North, further accentuating the heightened vulnerability of cities in the global South, accounting for 33.7% (13.4%) (Fig. 4d, e).

The exploration of inequality between cities in the global North and South under hypothetical and realistic scenarios revealed compelling insights. In the hypothetical scenario, wherein urbanization effects were not considered, heat exposure levels were significantly higher in global South cities compared to their global North counterparts (Fig. 2). We also tested heat indicators defined by 95% and 98% temperature thresholds and found similar global North-South heat exposure differences under hypothetical scenarios (Supplementary Fig. 4). However, when accounting for urbanization-induced warming (the real scenario), the inequality between the two regions narrowed due to the stronger urban warming effect in the global North cities (Fig. 2 and Supplementary Fig. 4). For example, the difference in average heat exposure between cities in the global North and South narrowed from –2.42 × 10–3 to –1.98 × 10–3 in urban areas (Fig. 2). We further tested the impact of different heat indexes and temperature thresholds on the heat exposure (Supplementary Fig. 5). We also found a narrowing of difference in heat exposure between cities in the global North and South due to urbanization-induced warming, especially when defining the heat index by HWF (Supplementary Fig. 5). By observing the changes in the hypothetical and real scenarios at different times, we obtained consistent results (Supplementary Fig. 6). Notably, while urbanization-induced warming has narrowed the disparities between cities in the global South and North, the disparities have continued to widen over time (Supplementary Fig. 6).

The contribution of urbanization-induced warming to heat exposure in the global North and South cities

Stronger urban heat exposure was attributed to urban warming and urban population growth than in rural areas. Here we decomposed the increased urban heat exposure into the independent contribution of urban warming, the independent contribution of increased urban population, and the interaction of the two (see Methods). The assessment revealed that urbanization-induced warming contributes significantly to heat exposure, constituting 24.2% (Fig. 3a). This effect was more pronounced in areas characterized by low urbanization (34.2%) compared to those with high urbanization (25.8%) (Fig. 3c, e). Interestingly, the contribution of urbanization-induced warming to heat exposure displayed a notable disparity between global North and South cities. The contribution was significantly stronger in the global North (31.2%) compared to the global South (18.0%) (Fig. 3b). This discrepancy was consistent across both high and low urbanization areas, emphasizing the heightened impact of urbanization-induced warming on heat exposure in cities of the global North (Fig. 3d, f).

Discussion

Compound heatwaves are more hazardous than independent heatwaves under the combined influence of extreme daytime heat and extreme nighttime heat18,20. The stronger trend of compound heatwaves compared to independent heatwaves was observed, especially in HWM and CH (Supplementary Figs. 9, 10). For example, in urban areas, the increasing trend in CH from compound heatwaves was 3 times the trend from independent daytime heatwaves and 1 time the trend from independent nighttime heatwaves, respectively (Supplementary Fig. 10d). Urban heat islands intensify urban heat challenges, amplifying both daytime and nighttime extreme heat28,29,38. These heightened heatwaves in urban areas escalate the risk of morbidity and mortality among urban residents39. That is, it is urgent to understand the global urban populations exposure to compound heatwaves in the context of more intense and hazardous compound heatwave events. The conventional approach of employing global climate model data for assessing heat exposure often neglects urban representation, potentially skewing the assessment of urban heat exposure. The limitations of these models, including low spatial resolution, further hinder the specific consideration of urbanization-induced warming effects on heat exposure. Land surface temperatures obtained by remote sensing play an important role in the assessment of global urban warming10,30. Based on the satellite-derived near-surface air temperature data, we observed that urbanization had a much greater impact on compound heatwaves than on independent heatwaves (Supplementary Figs. 7, 8, 9, 10). For example, the urban HWF trend in compound heatwaves was about twice the rural HWF trend, while the urban HWF trend in independent heatwaves was slightly higher than the rural HWF trend (Supplementary Fig. 7a, d). Attention was therefore given to the impact of increased compound heatwaves due to urbanization on urban heat exposure.

The increase in compound heatwaves due to urbanization was observed in most cities globally (Fig. 1). Importantly, the impact of global urbanization on compound heatwaves displays spatial heterogeneous. The amplifying effect of urbanization on compound heatwaves was more pronounced in the global North cities than in the global South cities (Fig. 1). Further studies showed that heat exposure was generally stronger in the global South cities than in the global North cities under the hypothetical scenarios (Fig. 2). While the rural areas in the global North experience more severe heat extremes, the substantial populations in urban areas of the global South led to higher levels of overall heat exposure in this region (Supplementary Fig. 11). And the contribution of population to urban-rural differences in heat exposure was more pronounced in the global South cities (Supplementary Fig. 12). However, due to more intense urban warming in the global North, the disparity in heat exposure between the global North and South in the real scenario narrowed compared to hypothetical scenarios (Fig. 2). Instead, our focus was not on urban heat exposure, but rather on the overestimation of the difference in urban heat exposure between the global South and North caused by urbanization-induced warming. These highlight the complexity of urban heat exposure, where different components can manifest differently and contribute to the overall risk. And these underscores the need for targeted interventions and strategies tailored to the unique challenges pose by urban heat exposure in both the global North and South, considering their specific urbanization patterns, climatic conditions, and population demographics.

Notably, most of the studies focused on temporal variations in heat exposure12,17,20, emphasizing stronger trends in the global South cities17. However, it’s worth noting that our study’s time frame (2003–2019) was relatively short, emphasizing the need for caution when interpreting temporal trends. Typically, studies analyzing heatwave trends span at least 30 years40. Due to the limited time series, the spatial pattern of heatwaves by averaging data from 2003 to 2019 was concerned. Of course, we also counted the heatwaves trend in global cities from 2003 to 2019. The trend in HWF in the global South cities was stronger than that in the global North cities (Supplementary Fig. 7), similar to previous studies17. However, we found that while the CH trend was stronger in the global South cities during independent heatwaves, it was stronger in the global North cities during compound heatwaves (Supplementary Fig. 10). These further highlights the importance of heatwave indicators and heatwave types in assessing heat exposure.

In our exploration of the effects of urban compound heatwaves on heat exposure, it’s essential to acknowledge that we focused on near-surface air temperatures, which was based on land surface temperature data and weather station data. It was important to recognized that weather stations were not evenly distributed in space, which can bias global temperature projections. Here, the effects of urbanization-induced warming on heat exposure was not a comparison between individual cities, but a comparison between cities in the global North and South, and this regional averaging reduced uncertainty. Moreover, cross-validation with in-situ near-surface air temperature measurements and land surface temperature41 showed the reliability of the results (Supplementary Discussion).

It’s crucial to recognize that these temperature measurements do not fully capture the complexity of the urban heat environments. Besides temperature, humidity and wind speed are important environmental factors that characterize urban heat stress12,42. Hot and humid weather can impede the body’s ability to maintain a safe core temperature43,44. Our study focused only on urban-rural differences in compound heatwaves and neglected the role of other environmental factors, which may bias the understanding of urban-rural heat stress. This is because the low humidity in the city can relieve heat stress42. In subsequent studies we should consider environmental factors such as temperature, humidity, and wind speed to construct a heat index and evaluate its urban-rural differences. Furthermore, the urban population density was used to construct the heat index and did not take into account the “vulnerability” of the population. Urban thermal environments tend to have a greater impact on older populations22 and should be considered in subsequent studies. Finally, the amplifying effect of urbanization on compound heatwaves was explored. However, the synergistic effect of heatwaves and urban heat islands in urban environments has been widely assessed25,26, and whether this synergistic effect exacerbates urban heat exposure should be further investigated.

Past research revealed disparities in future scenarios regarding global urban heat exposure22,36. On one hand, the urban expansion in the future is expected to intensify both daytime and nighttime urban heat islands, significantly heightening global urban heat risk (mainly in the tropical Global South45). Additionally, future urban population growth is projected to be most pronounced in developing regions of Asia and Africa. Furthermore, the global demographic is shifting towards an older age structure, with a growing proportion of individuals aged 65 years and above. Notably, a majority of this elderly population will reside in low- and middle-income countries22. The compounding influences of urbanization and demographic shifts have placed cities in the global South under substantial pressure regarding heat exposure challenges46. This predicament may further exacerbate existing inequalities in heat exposure between cities in the global North and global South. Therefore, considering urbanization-induced warming in the assessment of urban heat exposure and accurately evaluating global urban heat exposure in the context of future urbanization and population growth are imperative. Such an approach will enable the development of equitable and effective mitigation strategies to address the impending challenges posed by urban heat.

Methods

Delineation of urban and rural areas

Based on the 2018 Global Urban Boundaries (GUB) dataset47, we selected 1028 urban areas larger than 100 km2. To ensure that the size of the urban and rural areas is essentially the same, we created dynamic buffers for each city boundary. The buffer distance is48:

$${D}_{{\rm{buffer}}}=(\sqrt{2}-1)\sqrt{\frac{S}{{\rm{\pi }}}}$$
(1)

where S is the size of the urban area. Based on the International Geosphere-Biosphere Programme (IGBP) global land cover classification system in the MODIS (MCD12Q1) annual land cover product49, water bodies, wetlands and ice pixels were excluded. To avoid the effect of elevation on temperature, based on a digital elevation model (DEM, GTOPO3050), we excluded pixels in rural areas that were 50 m above or below the average urban elevation. Moreover, to evaluate heat exposure at the sub-city scale, we used the 2018 global artificial impervious area (GAIA) dataset51 to classify urban areas into high urbanization areas and low urbanization areas. The spatial resolution of the GAIA dataset is 30 m. To match the spatial resolution of the remotely sensed temperature data (1 km), we solved for the percentage of GAIA city pixels in each near-surface air temperature pixel. A pixel was defined as a high urbanization area if the percentage exceeded 0.5, and a low urbanization area if the percentage of GAIA was greater than 0 and did not exceed 0.5.

Calculation of urban heatwaves

Here, the heatwaves were calculated based on the daily maximum and daily minimum temperatures of the near-surface air temperature dataset37, which has a spatial resolution of 1 km and a time scale of 2003-2020. The fine spatial resolution and wide global distribution of the dataset allows us to better study urban-scale heatwaves compared to climate model data, reanalysis data and meteorological station data. We extracted the daily average maximum and minimum temperature time series for rural, urban, high urbanization and low urbanization areas. The missing values were replaced in the time series with the mean of the two days before and after.

A heatwave is defined as a high temperature event that exceeds temperature thresholds for several consecutive days52,53. There are no standardized criteria for defining heatwaves, and the diversity of criteria is mainly in the heat benchmarks, heatwave thresholds, heatwave durations, etc.54,55. Here, we extracted three different types of heatwaves, i.e., compound heatwaves, independent daytime heatwaves, and independent nighttime heatwaves, for the extended summer season (Northern Hemisphere: May to September, Southern Hemisphere: November to March) in global cities from 2003 to 2019. First, hot day (night), i.e., day (night) temperature above the threshold, were identified based on the average maximum (minimum) temperatures in the city. Based on the temperatures in the rural area, we defined the temperature thresholds for the city using a relative threshold. That is, for a given day, a day was defined as a hot day (night) if the temperature exceeds 90% of the maximum (minimum) temperatures on that day and on the 7 days before and after from 2003 to 2019. An independent daytime heatwave was defined as at least three consecutive days with hot days and no hot nights in the process. An independent nighttime heatwave was defined as at least three consecutive days with hot nights and no hot days in the process. A compound heatwave was defined as at least 3 consecutive days with both hot days and hot nights. We further defined heatwaves using 95% and 98% temperature thresholds to test the effect of different heatwave definitions on the results.

We used heatwave frequency (HWF), averaged heatwave magnitude (HWM), average heatwave duration (HWD), and cumulative heat (CH) to characterize frequency, intensity, and duration of heatwaves. HWF is defined as the total number of days for all heatwave events in a year. HWD indicates the average duration of all heatwave events in a year. The HWM represents the average magnitude of all heatwave events in a year, with the event magnitude being the average excess heat (Eq. (2)). CH is defined as the total excess cumulative heat during heatwave in a year (Eq. (3))40.

$${\rm{HWM}}=\frac{{\sum }_{i}^{N}\frac{{\sum }_{j}^{D}({T}_{i,j}-{T}_{o,i,j})}{D}}{N}$$
(2)
$${\rm{CH}}=\mathop{\sum }\limits_{i=1}^{N}\mathop{\sum }\limits_{j=1}^{D}({T}_{i,j}-{T}_{o,i,j})$$
(3)

where, \(N\) denotes the total number of heatwave events in a year. \(D\) indicates the number of days in a heatwave event. \({T}_{i,j}\) denotes the temperature of the heatwave day and \({T}_{o,i,j}\) denotes the threshold temperature.

Heat exposure assessment

Heat exposure index

Heat exposure risk indicates the spatial proximity of the population to the heat hazard56. The heat exposure index (HEI) is expressed as:

$${\rm{HEI}}={\rm{HW}}\times {\rm{Pop}}$$
(4)

where, HW denotes the heat hazard and Pop denotes the population. HWF is often used as a heat hazard to characterize the population exposure to the heat risk12,17. However, HWF does not complete the representation of heat hazard, and there have been studies coupling frequency, intensity, and other characteristics21,22. Here, we used three indices to describe heat hazards. The first was to use the commonly used HWF (Eq. (5)). The second coupled the frequency, intensity and duration of heatwaves (Eq. (6)). The third was the use of cumulative heat, which is usually used as a comprehensive heatwave indicator (Eq. (7)).

$${\rm{HW}}\,{\rm{index}}\,1={\rm{HWF}}$$
(5)
$${\rm{HW}}\,{\rm{index}}\,2={\rm{HWF}}\times {\rm{HWM}}\times {\rm{HWD}}$$
(6)
$${\rm{HW}}\,{\rm{index}}\,3={\rm{CH}}$$
(7)

The population density, a commonly employed metric in urban heat risk assessment15,57, was used to characterize urban population exposure. Population data are from the Global Population Grid dataset (GPWV41158).

We normalized the features with the equation59:

$$X^{\prime} =0.1+\frac{X-{X}_{{\rm{MIN}}}}{{X}_{{\rm{MAX}}}-{X}_{{\rm{MIN}}}}\times (0.9-0.1)$$
(8)

where X′, X, XMAX, and XMIN are the normalized data, the original data, the maximum value in the original data, and the minimum value in the original data, respectively. Moreover, We used the Jenks Natural Break method60 to classify urban heat exposure. The Natural Break method has the characteristics of large differences between groups and small differences within groups, and is often used to classify data. Specifically, the heat exposure index of urban area, high urbanization area and low urbanization area is divided into three categories: high heat exposure cities, moderate heat exposure cities, and low heat exposure cities. The Jenks Natural Break method is implemented based on python’s ‘jenkspy’ libraries.

Impacts of urbanization-induced warming to heat exposure

We calculated HEI for rurual areas, urban areas, high urbanization areas and low urbanization areas. To analyze the effects of urbanization-induced warming on HEI, we assumed that if there were no urbanization-induced warming, heat hazards in urban environments would be consistent with the countryside, and the HEI could be expressed as:

$${{\rm{HEI}}}_{{\rm{counter}}}={{\rm{HW}}}_{{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{urban}}}$$
(9)

where,\(\,{\text{HW}}_{\text{rural}}\) denote the heat hazards in rural areas, and Popurban denotes the population density in urban areas. However, due to urbanization, the actual heat exposure is expressed as:

$${{\rm{HEI}}}_{{\rm{urban}}}={{\rm{HW}}}_{{\rm{urban}}}\times {{\rm{Pop}}}_{{\rm{urban}}}$$
(10)

where, \({\text{HW}}_{\text{urban}}\) denote the heat hazards in urban areas.

Furthermore, we decomposed the urban-rural difference in HEI (\({\text{HEI}}_{\text{urban}-\text{rural}}\)) into three components, namely the independent impact of urbanization-induced warming, the independent impact of population growth, and the interaction impact of the two.

$$\begin{array}{l}{{\rm{HEI}}}_{{\rm{urban}}-{\rm{rural}}}={{\rm{HW}}}_{{\rm{urban}}-{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{rural}}}+{{\rm{HW}}}_{{\rm{rural}}}\\\qquad\qquad\qquad\qquad\times {{\rm{Pop}}}_{{\rm{urban}}-{\rm{rural}}}+{{\rm{HW}}}_{{\rm{urban}}-{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{urban}}-{\rm{rural}}}\end{array}$$
(11)

The relative contribution of each component can be expressed as:

$${{\rm{RC}}}_{{\rm{HW}}}=\frac{{{\rm{HW}}}_{{\rm{urban}}-{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{rural}}}}{{{\rm{HEI}}}_{{\rm{urban}}-{\rm{rural}}}}$$
(12)
$${{\rm{RC}}}_{{\rm{Pop}}}=\frac{{{\rm{HW}}}_{{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{urban}}-{\rm{rural}}}}{{{\rm{HEI}}}_{{\rm{urban}}-{\rm{rural}}}}$$
(13)
$${{\rm{RC}}}_{{\rm{int}}}=\frac{{{\rm{HW}}}_{{\rm{urban}}-{\rm{rural}}}\times {{\rm{Pop}}}_{{\rm{urban}}-{\rm{rural}}}}{{{\rm{HEI}}}_{{\rm{urban}}-{\rm{rural}}}}$$
(14)

where, \({\text{RC}}_{\text{HW}}\), \({\text{RC}}_{\text{Pop}}\), and \({\text{RC}}_{\mathrm{int}}\) represent the independent contribution of urbanization-induced warming, the independent contribution of population growth, and the interaction of the two, respectively.

Inequality analysis of heat exposure between Global North cities and Global South cities

Cities located in the Global North exhibit higher levels of economic development and urbanization compared to those in the Global South. While the categorization of the global North and South varies slightly from study to study, in general the global South is generally characterized as a vulnerable developing country. We divided the global North and South cities with reference to previous studies61. We focused on inequality between cities in the Global North and the Global South. The Mann–Whitney U-test (two-sided) was used to test the differences in the statistics of cities in the global South and the global North. The Mann–Whitney U test is implemented based on python’s ‘scipy’ libraries.