Rising temperatures, rising disparities: The future of heat exposure in the U.S. by 2050

Abstract

Future global warming will affect nearly every aspect of human society and daily life. Yet, little is known about disparities in exposure to future heatwaves across different social groups. This study provides a fine-grained, spatially explicit prediction of heat-affected populations in the continental United States by 2050, considering four combined scenarios of shared socioeconomic pathways and representative concentration pathways. Incorporating projected changes in socio-demographic and economic compositions, we find that global warming is likely to exacerbate disparities in heat exposure, as measured by the Gini coefficient. The poverty group, residing in 52 % of continental states, is projected to face the greatest impact, followed by Black and Asian populations. Additionally, residents of densely populated metropolitan areas are expected to experience greater inequalities in heat exposure compared to those in smaller metropolitan regions. These findings underscore the regions where heat interventions are most urgently needed and emphasize the importance of prioritizing social equity—aligning with the United Nations’ Sustainable Development Goals and their guiding principle of leaving no one behind.

1. Introduction

Global warming, one of the greatest challenges of the 21st century, threatens nearly every aspect of daily life (Trenberth et al., 2014). In the summer of 2022, record-breaking high temperatures swept across countries in Europe, Asia, and North America (Morita, 2022), with the Northern Hemisphere’s peak reaching 53.6 °C in Shush, Iran, on August 9, 2022. Although this was not the highest temperature ever recorded, the past eight years have been the warmest on record globally, and such extreme conditions are projected to become the norm by 2050 (Battisti and Naylor, 2009). The impacts of rising temperatures are not evenly distributed: marginalized groups—including communities living in poverty, people of color, women, and Indigenous populations—face disproportionate risks. Adaptation to global warming is complex and context-dependent, occurring at both individual and collective levels. It is shaped not only by the degree of exposure to heat but also by socioeconomic capacity, coping strategies, and perceptions of heat-related risks (Diffenbaugh and Burke, 2019). In alignment with the core principle of the United Nations’ 2030 Agenda for Sustainable Development—leaving no one behind (UN, 2016)—it is critical to prioritize social equality and justice in climate change policies to safeguard the most vulnerable populations.

While a growing body of literature has examined retrospective patterns of global warming and projected future heat trends (Perkins-Kirkpatrick and Lewis, 2020; Oliver et al., 2018; Cramer et al., 2018; Chen et al., 2022), far less attention has been devoted to predicting how future warming will shape disparities in heat exposure across different social groups and between urban and rural areas. In this context, social disparity—often measured by the Gini coefficient—refers to the unequal distribution of environmental stressors such as heatwaves, resulting in disproportionate risks across socioeconomic classes, racial and ethnic groups, and genders. A consistent finding across global studies (Deilami et al., 2018) is that historically marginalized communities, often defined by lower income and minority status, bear a greater burden of urban heat. These inequities are frequently rooted in historical planning practices, such as redlining in the U.S., which left affected neighborhoods with limited green infrastructure and more heat-absorbing surfaces. The consequences are wide-ranging, including elevated public health risks (e.g., heat stroke, respiratory illnesses), higher energy demand, strain on infrastructure, and significant economic productivity losses (Heaviside et al., 2017). To analyze these dynamics, recent research increasingly integrates satellite-based thermal remote sensing with ground-based observations and advanced analytics to capture the complexity of urban thermal landscapes (Saaroni et al., 2018). Addressing urban heat disparities requires integrated planning approaches that prioritize environmental justice, data-driven interventions, and public health preparedness to foster resilient and equitable cities. Incorporating social disparity and justice into climate change research is therefore essential for developing solutions that not only reduce risks but also confront historical injustices and promote fairer distribution of resources and benefits (Islam and Winkel, 2017).

As such, two essential questions urgently need to be addressed: 1) how many and where populations will be exposed to future global warming? and 2) how will future global warming change the social (in)equality of heat-affected populations across different social groups (e.g., minority and people in poverty)? The answers to these two questions would provide spatially explicit evidence to guide governments at various levels in avoiding climate actions that exacerbate existing heat-exposed disparities and instead contribute to a more just and sustainable future for all.

Our study aims to address these knowledge deficits by predicting populations exposed to increasing summer heat in the continental US by 2050 at a fine-grained level, as well as estimating the heat-exposed disparity across social groups, including Black, Hispanic, Asian, White, Native American, and individuals in poverty. Drawing on the predicted heat imageries from the Coupled Model Inter-comparison Project Phase 6 (CMIP6, 2022) and future population landscape (Gao and Pesaresi, 2021), we delineate the locale and number of heat-affected populations under the shared socioeconomic pathways (Meinshausen et al., 2020) and the representative concentration pathways (Van Vuuren et al., 2011). Further based on the US decennial census and American Community Survey (ACS) data in 1980, 1990, 2000, 2010, and 2019 (Logan et al., 2014, 2016), we employed a second-order polynomial regression model (see Methods) to predict the proportion of the social groups (i.e., the Asian, Black, Hispanic, White, Native American, and people in poverty) over the total population and examine the change of heat-exposed disparity through the Gini coefficient. Our key findings show that people living in densely populated large metropolitan areas face greater disparities in terms of heat exposure compared to those residing in counterpart areas. Global warming by 2050 will potentially exacerbate heat-exposed disparity, measured by the Gini coefficient—the poverty group, residing in 52 % of the total continental states, will experience the greatest impact, followed by the Black and Asian groups. Such a rising heat-exposed disparity by 2050 is attributed to both the increasing variations in predicted heat exposure across social groups and the increasing concentration of vulnerable groups within the heat-affected areas. We make contributions to the literature by providing detailed future projections of populations exposed to heat, introducing the measurement of social disparities in heat exposure, using multiple climate and socioeconomic scenarios to capture a range of possible outcomes, and offering actionable insights to prevent further marginalization of vulnerable groups. The research design of our study can be applied to predict the impact of other environmental stressors on populations and their subgroups, and to evaluate social disparities and environmental justice based on different modelling scenarios.

2. Literature review

2.1. Exposure risk assessment

Exposure risk assessments have been extended to multiple natural disasters such as floods, wildfires, hurricanes, and extreme heat, particularly concerning their impacts on vulnerable populations (Ward et al., 2020). Traditional assessments often relied on static hazard maps and generalized population data, limiting their capacity to capture the dynamic and disproportionate effects on marginalized groups. Contemporary approaches increasingly leverage high-resolution spatial data, remote sensing, real-time monitoring, and machine learning models to improve both spatial and temporal accuracy (Paltrinieri et al., 2019). A key development is the integration of social vulnerability indices with geospatial hazard models. These composite indices combine demographic, socioeconomic, and health data to identify populations more susceptible to harm, such as low-income communities, the elderly, people with disabilities, and ethnic minorities (Jurgilevich et al., 2017). When layered with hazard exposure data—such as flood depth, fire perimeters, or wind speed—these models provide nuanced insights into who is at risk and to what extent. Another important shift is the adoption of dynamic risk assessment frameworks that account for changing environmental conditions and human behavior over time (Aerts et al., 2018). For instance, mobile phone data and crowd-sourced information are now being used to track evacuation patterns or identify underserved areas in disaster response efforts. Climate change has also prompted the development of forward-looking risk assessments, which model future scenarios under different emissions and land-use pathways. This is particularly important for long-term planning and resilience building in regions where historical data may no longer reflect current or future risks.

One common development pathway that has been widely adopted in the ongoing Intergovernmental Panel on Climate Change (IPCC) assessment of global climate change is the shared socioeconomic pathways (SSP) framework (Samir and Lutz, 2017). It describes how global society, demographics and economics will change in the coming century concerning policy assumptions and the socio-economic narrative (e.g., energy demand and supply, and technological change) (Meinshausen et al., 2020). The framework we utilize offers a comprehensive approach, enabling us to conduct multi-scenario predictions that are unified and comparable. By considering various potential pathways and uncertainties associated with future socio-economic factors, we can achieve a more holistic understanding. Specifically, there are five scenarios within this framework (Samir and Lutz, 2017). SSP1 represents a sustainable pathway that prioritizes people in an environmentally friendly direction. SSP2 falls between SSP1 and SSP3, representing a middle pathway. SSP3 depicts a regional rivalry pathway characterized by limited global cooperation. SSP4 represents a divided pathway where inequality and stratification increase both within and across countries. Ultimately, SSP5 represents a fossil-fueled development pathway in which the global economy experiences rapid growth yet also faces significant challenges in mitigating climate change. Our study will follow the SSP framework to predict the future heat and heat-exposed disparity across social groups and urban-rural spaces.

2.2. Heat exposure disparity

Heat disparity represents a critical environmental and social challenge globally. This phenomenon, where urban areas are significantly warmer than surrounding rural regions, is caused by the modification of land surfaces, the design of the built environment, and waste heat from human activities (Deilami et al., 2018). Empirical studies across diverse continents underscore the pervasive nature of this disparity. In Europe, cities like London and Paris exhibit temperature differentials of up to 10 °C at night during extreme heat events, contributing to significant premature deaths across multiple European cities (Chakraborty et al., 2019). Asian cities, particularly in East Asia, are on average 1.6–2.0 °C warmer, with some areas in Indonesia and the Philippines experiencing temperatures up to 5.9 °C higher, leading to heat-related deaths in the East Asia and Pacific region (Chua et al., 2022; Damiati et al., 2016). African cities, undergoing rapid urbanization, show urban zones 3–4 °C warmer due to impervious surfaces (Li et al., 2021), despite often having significant green spaces. In the Americas, U.S. cities can be 1 °F–7 °F hotter during the day, with studies in Philadelphia and Tempe, Arizona, revealing clear socioeconomic discrepancies in heat exposure (Habeeb et al., 2015). The consistent empirical evidence from diverse regions underscores the critical role of green infrastructure in global heat mitigation. The above multi-national studies consistently highlight vegetation and green spaces as primary cooling mechanisms. This establishes green infrastructure not merely as an aesthetic amenity but as a fundamental, globally recognized, and empirically proven tool for urban heat mitigation and climate adaptation. Its strategic implementation, particularly in vulnerable areas, is essential for public health and environmental justice.

Furthermore, heat exposure varies significantly across different social groups, with notable disparities influenced by factors such as income level, age, and race, largely due to historical patterns of residential segregation that have concentrated minority populations in these hotter, less-resourced neighborhoods (Renteria et al., 2022). These communities often have fewer trees, parks, and other cooling infrastructures. Additionally, individuals in certain occupations with a higher representation of minority groups, such as agriculture and construction, experience prolonged and intense heat exposure. This intersection of age, income, and race creates a compounding effect, where the most marginalized groups bear the greatest burden of rising temperatures and heat-related illnesses (Gronlund, 2014). More specifically, low-income neighborhoods typically have less vegetation and more impervious surfaces, which exacerbate the urban heat island effect, leading to greater heat exposure compared to wealthier areas (Alizadeh et al., 2022; Benz and Burney, 2021). In the United States, a substantial study found that 72 % of urban areas exhibit an income gradient where poorer neighborhoods face higher surface temperatures (Benz and Burney, 2021). Older adults, particularly those aged 65 and above, are more vulnerable to heat-related health risks due to declining physiological resilience, exacerbating the impact of heat stress on this demographic (Lo et al., 2019). Racial disparities are also present, as communities of color often reside in regions with higher heat exposure, which compounds social inequalities (Khatana et al., 2022; Rastogi et al., 2023). The health impacts of extreme heat manifest disproportionately, with lower socioeconomic, racial, and age groups bearing a higher share of health risks, underscoring the need for targeted public health interventions (Kunda et al., 2024).

2.3. Prediction of heat exposure

The prediction of heat patterns in our study employed the framework known as the representative concentration pathway (RCP) (Van Vuuren et al., 2011), which was adopted as a trajectory for greenhouse gas concentrations by the IPCC Fifth Assessment Report in 2014, encompassing four distinct pathways designed for climate modelling and research purposes (IPCC, 2014). These pathways depict various potential climate futures, each contingent on the level of greenhouse gas emissions in the coming years. The RCP framework comprises four scenarios (Van Vuuren et al., 2011), offering plausible future scenarios of human-induced forcing across a spectrum from low to high emissions. Specifically, RCP2.6 represents a scenario with a peak radiative forcing of around 3 W/m² by the mid-century, followed by a decline to 2.6 W/m² by 2100. RCP4.5 represents a moderate scenario characterized by stabilization (without overshoot) of radiative forcing at 4.5 W/m² post-2100, with emissions peaking around 2040 and subsequently declining. RCP6.0 describes a scenario with stabilization (without overshoot) of radiative forcing at 6 W/m² post-2100. Finally, RCP8.5 represents the highest baseline emissions scenario, projecting a rise in radiative forcing to 8.5 W/m² throughout the 21st century, as emissions continue to increase. Since the original RCP pathways proposed by the IPCC Fifth Assessment Report are being considered together with SSPs, our study integrated SSPs with RCPs as four combined SSP-RCP scenarios to align the predicted population growth and predicted heatwave—SSP1-RCP2.6 (sustainability-low emission), SSP2-RCP4.5 (middle of the road-moderate emission), SSP3-RCP6.0 (regional rivalry-moderate emission), and SSP5-RCP8.5 (fossil-fueled development-high emission) and the Climate Analytics Group. These projections were derived from the general circulation model as part of the Coupled Model Intercomparison Project Phase 6 (CMIP6, 2022). The NEX-DCP30 dataset contains monthly temperature predictions from 1950 to 2099, covering the conterminous United States. According to CMIP6, the dataset provides temperature predictions for four representative concentration pathways (RCPs) representing different greenhouse gas emissions scenarios: RCP2.6 (active mitigation), RCP4.5 and RCP6.0 (intermediate emissions), and RCP8.5 (high emissions).

3. Materials and methods

3.1. Predicted heat data

We obtained projected heat data from the NASA Earth Exchange (NEX) Downscaled Climate Projections dataset (NEX-DCP30), developed by NASA and the Climate Analytics Group. These projections are derived from general circulation models (GCMs) under the Coupled Model Intercomparison Project Phase 6 (CMIP6, 2022). The NEX-DCP30 dataset provides monthly temperature predictions for the conterminous United States from 1950 to 2099. CMIP6 incorporates approximately 100 distinct climate models developed by different modeling groups worldwide. The dataset includes projections under four representative concentration pathways (RCPs), representing a range of greenhouse gas emission scenarios: RCP2.6 (active mitigation), RCP4.5 and RCP6.0 (intermediate emissions), and RCP8.5 (high emissions). Following common practice in climate studies (e.g., Tebaldi et al., 2021; Babaousmail et al., 2021; Kim et al., 2020; Chen et al., 2020; Sellar et al., 2020), we calculated ensemble averages for each RCP scenario to capture the full spectrum of possible future climate outcomes. The CMIP6 dataset provides temperature predictions at a spatial resolution of approximately 25 km (0.25°). To align with sociodemographic data from the U.S. Census, we aggregated the raw raster data to the census tract level by calculating mean values within each tract, enabling integration of climate projections with population and socioeconomic characteristics.

3.2. Predicted population data

We obtained projected population data from the global 1-km downscaled population base year and projection grids for 2000–2100, developed under the Shared Socioeconomic Pathways (SSPs) framework. This dataset, provided by the Socioeconomic Data and Applications Center (SEDAC) at Columbia University (Gao and Pesaresi, 2021), offers global urban, rural, and total population estimates at 1-km resolution (approximately 30 arc-seconds) across five SSP scenarios: SSP1 (sustainability), SSP2 (middle of the road), SSP3 (regional rivalry), SSP4 (inequality), and SSP5 (fossil-fueled development). Following O’Neill et al. (2016), we integrated these SSPs with Representative Concentration Pathway (RCP) scenarios (Supplementary Fig. S6), excluding SSP4 since it is not directly paired with any RCP. This integration yielded four combined SSP–RCP scenarios that jointly account for socioeconomic trajectories and greenhouse gas emission pathways: SSP1–RCP2.6 (sustainability–low emissions), SSP2–RCP4.5 (middle of the road–moderate emissions), SSP3–RCP6.0 (regional rivalry–moderate emissions), and SSP5–RCP8.5 (fossil-fueled development–high emissions).

3.3. US demographic and socioeconomic data

We obtained the census tract-level data on populations in poverty and populations of different ethnic and racial groups (i.e., Non-Hispanic White, Non-Hispanic Black, Asian and Pacific Islander, Hispanic, and Native American) for 1980, 1990, 2000, 2010, and 2019 from the longitudinal tract database (LTDB) (Logan et al., 2014, 2016). The LTDB provided interpolations for those five decades of tract-level data to match the 2010 tract boundaries, enabling tract-level analysis of the evolution of demographic and socioeconomic characteristics over time and the prediction of these characteristics in future decades.

3.4. Defining heat-affected population

We integrated the predicted heat data with population projections within a consistent temporal framework. The raster-based heat dataset provides monthly temperature estimates for each grid cell, whereas the population dataset reports projected counts at decadal intervals (2030, 2040, and 2050). To align these datasets, we selected the summer months (June–August) as the temporal extent. Annual summer temperature values were calculated as the average of the three monthly temperatures to correspond with the decadal population estimates. Following previous studies (Merbitz et al., 2012; Vogel et al., 2020), we identified heat hotspots as locations exceeding the 80th percentile of temperatures within the same census tract. Using this threshold, we generated a binary raster layer where cells with values of 1 represented heat-affected locations (hotspots) and cells with values of 0 represented non-heat-affected areas. This classification was applied separately to daily minimum and daily maximum temperatures, and a location was considered heat-affected only if it exceeded the 80th percentile threshold for both. The resulting binary heat layer was then overlaid with the population projections to estimate the number of people residing in heat-affected areas. These population counts were further aggregated to multiple administrative levels (tract, county, and state) using the Zonal Statistics tool in ArcGIS Pro 3.2.

3.5. Predicting sociodemographic composition

We rasterized the acquired LTDB data into 1-km grid cells, ensuring compatibility with the pre-existing temperature and population aggregate data. Taking into account the limited availability of ground-truth census records (i.e., in 1980, 1990, 2000, 2010, and 2019), it is largely possible that the prediction of future sociodemographic compositions based on a singular grid cell and its historical records in five census years would generate uncertainty and bias to some degree. Instead, we employed a neighborhood-aware forecasting method. To project the measure for a single central pixel, we employed a second-order polynomial regression model (Brusco et al., 2009), which incorporated historical records of the central pixel from the years 1980, 1990, 2000, 2010, and 2019 (Supplementary Fig. S7) along with its neighboring grid cells. Considering a grid at the location (i, j), its regression follows Eq. (1):

$$Y_{\left(i,j\right)}={\alpha }_{\left(i,j\right)}+\sum _{k=1}^p{\beta }_{\left(i,j\right)k}{x}_{\left(i,j\right)k}+\sum _{k=1}^p{\gamma }_{\left(i,j\right)k}{x}_{\left(i,j\right)}^2+{\epsilon }_{\left(i,j\right)}$$ (1)

where $x_{\left(i,j\right)k}$ is the observed value; ${\beta }_{\left(i,j\right)k}$ and ${\gamma }_{\left(i,j\right)k}$ are the first-order and second-order parameters, respectively; ${\alpha }_{\left(i,j\right)}$ is the constant; and ${\epsilon }_{\left(i,j\right)}$ is the error term for grid $\left(i,j\right)$, which is assumed to follow a normal distribution with zero mean and constant variance ${\epsilon }_{\left(i,j\right)}\sim N\left(0,{\sigma }^2\right)$. In this study, we implemented a 5 × 5 neighboring window. Thus, the grids that participate in the regression include grids from locations at $\left(i-2,j-2\right)$ to $\left(i+2,j+2\right)$, with a total of p = 25.

3.6. Measure of heat-exposed disparity

We employed the Getis-Ord Gi hotspot analysis (Songchitruksa and Zeng, 2010), conducted in ArcGIS Pro 3.2, to measure the spatial disparity of the heat-affected population. It identifies statistically significant spatial clusters of the large number of heat-affected populations as hot spots colored in red and conversely the small number of heat-affected populations as cold spots colored in blue (e.g., Fig. 1e), based on the predefined confidence levels (i.e., p-value set to be 0.1, 0.05 and 0.01) and accordingly the confidence level bin fields (i.e.,±3 bins reflect statistical significance with a 99 percent confidence level;±2 bins reflect a 95 percent confidence level;±1 bins reflect a 90 percent confidence level; and the clustering for features in bin 0 is not statistically significant). We then employed the Gini coefficients (Lerman, R. I., & Yitzhaki, 1984), the most commonly used index, to measure heat-exposed disparity among different social groups. The value of the Gini coefficient ranges from 0 (indicating the least disparity) to 1 (indicating the greatest disparity). The measure of Gini coefficients was applied to different spatial scales, including at the state and metropolitan level.

Fig. 1.

Predicted heat-affected population at the state and county level. (a) the average increase rate for the heat-affected population from 2030 to 2050 in each state based on each of the four SSP scenarios; (b, c, d) the estimated number of heat-affected populations by 2050 averaged across four SSP scenarios at the state, county, and tract level, respectively (the maps for each SSP scenario at the county and tract level are provided in Supplementary Figs. S1 and S2); (e) hotspots (high concentrations) of the heat-affected population at the county level to indicate the spatial inequality in heat exposure (the hotspot maps for each SSP scenario at the county level are provided in Supplementary Fig. S2).

4. Results

4.1. Predicted heat-affected population at various levels

The diverged landscape of heat-affected populations can be observed at the state level (Fig. 1b), at the county level (Fig. 1d), and at the tract level (Fig. 1c). At the state level (Fig. 1b), the number of heat-affected populations is highest in the state of New York (around 20 million by 2050 based on the average of four SSP scenarios), followed by Pennsylvania (+9.9 million), Michigan (+6.6 million), Arizona (+5 million), and Ohio (+4.5 million); however, the increase rates of heat-affected population are varying across states and SSP scenarios (Fig. 1a). For example, such increase rates are obviously large for the SSP5 scenario in Florida, Louisiana, and Delaware, whilst they are relatively minor for other SSP scenarios for the majority of states. At the county level, the hot spots (high concentrations) of heat-affected populations (Fig. 1e) appear predominantly on the western and eastern coasts, covering the vast land in California, Oregon, New York, and Pennsylvania.

The number of predicted heat-affected populations across metropolitan areas is shown in Fig. 2a. The metropolitan area with the most heat-affected population by 2050, averaged across four SSP scenarios, is Detroit (+2.3 million), followed by Minneapolis (+2.2 million), Portland (+1.5 million), Seattle (+1.4 million), and Pittsburgh (+1.2 million). Among the top 30 most populated metropolitan areas (Fig. 2b), the increased rate of heat-affected populations from 2030 to 2050 varies across different metropolitan cities. However, it is relatively consistent for the SSP1 scenario, and more varied for the SSP2 scenario. Such an increase rate is the highest in Orlando under the SSP2 scenario (53.1 %), followed by St. Louis under the SSP5 scenario (38.4 %), and Boston under the SSP3 scenario (36.8 %).

Fig. 2.

Predicted heat-affected population in the top 30 most populated metropolitan areas. (a) the estimated number of heat-affected populations by 2050 at the metropolitan level, averaged across four SSP scenarios, with the increased number of heat-affected populations from 2030 to 2050 shown as the vertical bar; (b) the increased rate of heat-affected populations from 2030 to 2050 in the top30 most populated metropolitan areas, ranked by the number of populations in 2022 in a descending order.

4.2. Heat-exposed disparity across social groups

After breaking down the heat-affected populations into different social groups, we would be able to examine the spatial variations of heat-affected populations across social groups (Fig. 3a), as well as the change of heat-exposed disparity across social groups (Fig. 3b). The Gini coefficients for the Black groups living in each of 361 metropolitan areas in 2030, 2040, and 2050 are plotted in Fig. 3b; the percentages of Black people living in all metropolitan areas (Fig. 4; Table 1) who will encounter increased heat-exposed disparity are 68.1 % in 2030–2040 and 66.8 % in 2040–2050, whilst such percentages for them living in the top 30 most populated metropolitan areas are much higher, 83.3 % and 80.0 %, respectively. For populations in poverty, 75.3 % and 74.8 % of them living in all 361 metropolitan areas encounter increased heat-exposed disparity in heat exposure in 2030–2040 and 2040–2050, respectively; such percentages in the top 30 most populated metropolitan areas increase to 83.3 % and 76.7 %. It indicates that populations in poverty living in more populated large metropolitan areas encounter an increasing level of heat-exposed disparity compared to all metropolitan areas as a whole. Such an observation also applies to Asian, White, Black, and Native American groups. In addition, among all metropolitan areas, populations in poverty experience more significant inequality in heat exposure, followed by the Asian, Black, Hispanic, White, and Native American populations, regardless of the time period. However, such an order changes to the Black, Poverty, Asian, White, and Native American among the top 30 most populated metropolitan areas (Supplementary Fig. S4), reflecting that future global warming would increase heat-exposed disparity more significantly for populations in poverty, and the Black and Asian groups.

Fig. 3.

Heat-exposed disparity among different social groups from 2030 to 2050 at the metropolitan and tract level. (a) the spatial patterns of the estimated number of heat-affected populations for the Asian, Black, Hispanic, Native, and poverty groups at the tract level; (b) the Gini coefficients for the Black group, as an example, in each of 361 metropolitan areas in 2030, 2040 and 2050, ordered by the Gini coefficient in 2030. The visibility of x points here was set to be 60 % to reduce the crowdedness and fuzzy overlapping of points.

Fig. 4.

The increase in Gini coefficients for the Black group from 2030 to 2050. The statistics for other social groups are provided in Supplementary Fig. S4.

Table 1.

The percentage of certain social groups encountering increased heat-exposed disparity in a 10-year period.

	Among all metropolitan areas (%)		Among the top 30 metropolitan areas (%)
	2030–2040	2040–2050	2030–2040	2040–2050
Asian	72.6	65.4	76.7	80.0
White	63.4	70.4	70.0	80.0
Hispanic	67.0	71.5	66.7	56.7
Black	68.1	66.8	83.3	80.0
Native	42.1	35.7	43.3	36.7
Poverty	75.3	74.8	83.3	76.7

We further measured the increase in heat-exposed disparity across social groups at the state level (Fig. 5). If the relative measure of Gini coefficients (e.g., green line as the ratio of Gini values in 2050 over 2030) is in line with the ring of 1.0 (shown in the legend), it indicates there is no substantial increase in heat-exposed disparity from 2030 to 2050; otherwise, if it goes beyond the ring of 1.0 and towards the outside edge, it indicates the increase in social disparity. We can observe that there is a continuous increase in heat-exposed disparity from 2030 to 2050 in all 47 states examined; such an increase appears most obviously within the poverty group in 23 states (52 % of the total), followed by the Black group in seven states (16 %) and the Asian group in five states (11 %). In addition, such an increase in social equality in heat exposure appears more obvious on the eastern coast of the US (e.g., Rhode Island, Delaware, Vermont, and Washington D.C.). The largest increase in heat-exposed disparity for the Black and Native American group is observed in Washington D.C. up to 2050; whilst such an increase for the Asian group is observed in Rhode Island up to 2050, and such an increase for the Hispanic group is observed in Vermont up to 2040 and in Delaware up to 2050.

Fig. 5.

Change in the heat-exposed disparity among different social groups from 2030 to 2050 at the state level. The figure showing the raw measure of Gini coefficients is provided in Supplementary Fig. S5.

5. Discussion

5.1. Key findings

Our study provides fine-grained predictions of heat-affected populations in the continental United States by 2050 under four SSP–RCP scenarios. We further examine disparities in heat exposure across social groups, including Black, Hispanic, Asian, White, Native American, and people in poverty. The results show that the largest increase in heat-affected populations will occur along the western coast, the Great Lakes region, and New England, with particularly notable growth in metropolitan areas such as Detroit, Minneapolis, and Portland. Our findings indicate that by 2050, global warming will exacerbate disparities in heat exposure among all social groups. The poverty group, residing in 52 % of continental states, is projected to experience the greatest impact, followed by Black and Asian populations. In addition, residents of densely populated metropolitan areas will face greater disparities in heat exposure compared to those in other metropolitan regions. Together, these findings provide a spatially explicit projection of future heat-affected populations, identifying priority areas for targeted heat interventions and equitable resource distribution. This evidence can help guide efforts to address the needs of ethnically diverse populations with multicultural backgrounds and varying levels of vulnerability.

The increasing disparity in heat exposure, as projected in our study, is primarily driven by a confluence of two critical factors: the increasing variations of predicted heat exposure across different social groups and the increasing concentration of vulnerable populations within areas most affected by heat. First, the uneven distribution of heat itself is intensifying across social groups. Global warming is not affecting all areas or populations equally. Our study indicates that by 2050, certain demographic segments, particularly the poverty group, followed by Black and Asian communities, are projected to experience the most significant impacts of heat. This suggests that the physical manifestation of heat, exacerbated by climate change, is becoming more pronounced in the neighborhoods where these groups predominantly reside. This is often a legacy of historical planning practices, such as redlining in the U.S., which led to less green infrastructure and more heat-absorbing surfaces in historically disadvantaged areas. As global temperatures rise, pre-existing thermal inequalities are amplified, resulting in greater temperature differentials between affluent and marginalized neighborhoods within the same urban areas.

Second, there is a discernible increase in the concentration of vulnerable groups within these heat-affected areas. The study highlights that populations in poverty, as well as Black and Asian communities, are increasingly facing higher levels of heat-exposed disparity, particularly in densely populated large metropolitan areas. This indicates a demographic trend where those most susceptible to the adverse effects of heat are either already concentrated in, or are increasingly moving into, urban zones that are becoming hotter. This concentration can be due to various socioeconomic factors, including housing affordability, limited mobility, and historical patterns of residential segregation. The combination of rising temperatures and the clustering of vulnerable populations in areas with less adaptive capacity (e.g., fewer green spaces, more impervious surfaces, limited access to cooling resources) creates a compounding effect that exacerbates the overall heat disparity. This dynamic means that the communities least equipped to cope with extreme heat are the ones increasingly bearing the brunt.

5.2. What was added to the literature

The IPCC’s reports have substantially advanced understanding of the complex interactions between climate change and social vulnerabilities. Building on this foundation, existing literature has linked climate change, heat exposure, and social disparities through the IPCC’s quantitative assessment framework (Cheng et al., 2021; Huang et al., 2011; Mitchell and Chakraborty, 2015; Renteria et al., 2022; Wilson et al., 2013), highlighting socioeconomic factors such as income, race, and ethnicity that shape differential vulnerability to heat-related risks. However, few studies have generated fine-grained predictions of future heat-affected populations or examined the resulting disparities with sufficient detail to inform ethnicity-specific interventions and spatial planning strategies. Our research addresses this gap by providing high-resolution projections of heat-exposed populations across multiple SSP–RCP scenarios, thereby strengthening the robustness of outcomes and offering a range of possible futures for consideration. Moreover, this work advances the U.S. Environmental Protection Agency’s (USEPA) demographic and spatial allocation models, which are used to produce the Integrated Climate and Land Use Scenarios (ICLUS) Version 2 (USEPA, 2017). A critical limitation of the ICLUS framework has been its omission of disparities in heat exposure among social groups. By integrating measures of inequality using the Gini index, our study delivers a more comprehensive and equitable assessment of climate impacts, thereby enhancing the evidence base for urban planning, public health, and policy interventions.

Our research framework can be integrated into existing models and scenarios, such as the USEPA’s Integrated Climate and Land Use Scenarios (ICLUS), to strengthen their capacity for assessing future heat impacts. By incorporating our projections of heat-affected populations and applying the Gini index, these models can more effectively capture disparities in exposure and support decision-making that prioritizes equality. Embedding the Gini metric into heat mitigation planning highlights the need to allocate resources and interventions in a way that reduces social inequities and ensures equal access to protective measures, thereby helping to prevent the further marginalization of vulnerable communities and advancing social justice in planning processes. In addition, our framework enhances the spatial resolution and granularity of existing models by providing fine-grained predictions of future heat-affected populations at the administrative level. These localized projections supply critical inputs for spatial allocation models, such as ICLUS (USEPA, 2017), enabling more targeted, effective, and context-sensitive planning strategies.

5.3. Policy implications

Our results provide crucial scientific support for governments in shaping early policymaking for climate change adaptation. Firstly, it is essential to conduct local-level vulnerability assessments to identify areas most susceptible to heat-related impacts. These assessments should consider local climate patterns, urban heat island effects in neighborhoods, population density, and socioeconomic factors. By understanding the specific vulnerabilities within each jurisdiction and local government area, planning authorities can prioritize resources and interventions where they are most needed. This will enable the implementation of highly targeted adaptation strategies to address the diverse needs of heat-affected populations. To raise awareness about the health risks of heat exposure and provide information on preventive measures, targeted outreach campaigns can be designed. This can involve distributing educational materials in multiple languages to cater to the needs of the Black, Hispanic, Asian, and Native American communities. Partnering with community organizations to conduct workshops and leveraging trusted community leaders to disseminate information effectively can also be valuable strategies. Furthermore, collaboration with community-based organizations and social service agencies can help establish and strengthen social support networks in areas with high concentrations of heat-affected populations. This is particularly important for the poverty group, who may lack the financial capacity to use air conditioning. Specific assistance measures can be provided to enhance their access to public resources and services, such as cooling assistance programs, financial support for utility bills, and transportation to cooling centers.

Our findings also highlight the importance of scenario-specific planning. Policymakers should consider the unique characteristics and vulnerabilities of each city to develop context-specific approaches. For instance, in Orlando, which exhibits the highest rate of increase under the SSP2 scenario, representing a middle-of-the-road development in terms of mitigation and adaptation challenges, policymakers should concentrate on implementing a range of strategies to address the projected heat impacts. Given the expected rise in heat-affected populations, it is crucial to prioritize measures that enhance resilience and protect vulnerable communities. Likewise, for St. Louis under the SSP5 scenario, which represents a fossil-fueled development pathway with rapid economic growth but significant mitigation challenges, policymakers should prioritize strategies that simultaneously address heat impacts and reduce greenhouse gas emissions. This may involve implementing energy-efficient building codes, promoting renewable energy sources, and incentivizing sustainable transportation options. These measures aim to mitigate both the health risks associated with heat exposure, such as stroke and heat-induced mortality (Alahmad et al., 2024; Yang et al., 2024), and the environmental impacts of increased energy consumption.

5.4. Limitations

This study has several limitations that can be addressed in future studies to expand upon our findings. First, the definition of heat-affected populations is based on relative temperature (i.e., the threshold of the 80th percentile) under the assumption that people have some ability to adapt to high temperatures. However, it would be valuable to conduct sensitivity analyses based on both relative and absolute temperature (e. g., a specific temperature threshold) to define heat hotspots. This would provide a more comprehensive understanding of the impact of temperature on population sensitivity to heat.

Second, our forecast of sociodemographic trends for the years 2030, 2040, and 2050 relies on historical data from 1980, 1990, 2000, 2010, and 2019 by using a second-order polynomial regression model - a specification widely recognized as the minimal extension beyond linear trend that can capture curvilinear demographic trajectories when only four to five decennial observations are available (Smith et al., 2005). Quadratic trend surfaces have a long history in spatial analysis and, when combined with local borrowing of strength, perform comparably to more complex spatiotemporal regressions in sparse series settings (Chi and Voss, 2011). It assumes the continuation of past patterns without accounting for potential inflection points or disruptive events. Nonetheless, this choice entails several caveats: residual long-range spatial dependence may remain unmodelled; abrupt structural breaks associated with migration shocks, policy interventions, or extreme climate events are smoothed over; the short temporal horizon constrains the complexity that can be estimated without over-fitting; and empirical evidence for the model’s transferability beyond the U.S. context is still limited. Future extensions that embed explicit spatial-econometric terms, integrate breakpoint-detection or policy-switching mechanisms, and exploit richer annual data streams will be valuable for further reducing uncertainty and enhancing generalizability.

Third, our approach incorporates a neighborhood-aware forecasting method that integrates adjacent areas into the predictive process. This is based on the premise that proximate locations are likely to exhibit similar trends, following the principle of the first law of geography. However, further examination of this assumption is warranted, as geopolitical, economic, and sociocultural factors can significantly impact and differentiate the development trajectories of even closely located regions. Therefore, our model may not fully capture the complexity of future sociodemographic trends. Addressing these limitations in future research would provide a more nuanced understanding of heat-affected populations and their sociodemographic dynamics.

Finally, these SSP-RCP scenarios are fundamental tools for projecting future climate change and informing adaptation and mitigation decisions, although it is subject to large uncertainty and arbitrary. First, even within a single SSP, there can be multiple emissions scenarios that result in different levels of radiative forcing and, consequently, varying global-mean temperature increases. Second, differences in global average temperature changes associated with RCPs and SSPs can also arise from advancements in climate modeling between different CMIP phases (e.g., CMIP5 and CMIP6). Third, global climate models typically operate at coarser spatial resolutions (e.g., 0.70–3.75°). To provide more localized climate information for regional adaptation strategies, these projections often undergo downscaling to higher resolutions (e.g., 0.1° or 1 km). While downscaled projections generally show overall agreement in the magnitude and sign of projected temperature changes with the larger ensemble, the downscaling process itself can introduce complexities and uncertainties, especially for very localized effects and patterns. In addition, the SSP-RCP scenarios primarily focus on CO2 emissions, while other factors like non-CO2 greenhouse gases, aerosols, and land-use changes also influence climate, and their treatment can be a source of uncertainty. Despite these uncertainties, the use of multiple SSP-RCP scenarios in our study enhances the robustness of the study’s findings. By considering a range of plausible future pathways, our study provides a comprehensive spectrum of possible outcomes, which can be further improved to be comprehensive with the above uncertainties considered in future studies.

In addition, it is acknowledged that we did not perform an independent evaluation of the climate models’ simulation capabilities for historical periods, as this has already been rigorously conducted by the CMIP6 data providers. The CMIP6 ensemble underwent comprehensive validation across a wide range of climate variables, spatial domains, and historical timelines by the modelling centers themselves. Given the extensive scope of those assessments and their established credibility, repeating such evaluations falls beyond the scope of our study. Instead, our focus remains on the projected disparities in future heat exposure, for which we adopt the CMIP6 projections as scientifically validated inputs.

6. Conclusion

In summary, this study predicts the impact of future global warming on heat-affected populations in the United States by 2050, highlighting how it is likely to exacerbate disparities in heat exposure among social groups. Using four combined socioeconomic and climate scenarios, the analysis forecasts that populations living in poverty, as well as Black and Asian communities, will be the most affected. Residents of the 30 most populous metropolitan areas are projected to experience particularly high exposure. By providing fine-grained predictions of heat-affected populations and identifying areas requiring targeted heat interventions and resource allocation, this research supports efforts to address the needs of ethnically diverse populations with varying vulnerabilities. Moreover, the approach can be extended to assess the impact of other environmental stressors on populations and their subgroups, evaluating social disparities and promoting environmental justice across different modeling scenarios. These findings underscore the importance of targeted heat mitigation and the prioritization of social equity, in alignment with the United Nations’ Sustainable Development Goals. They further call for cross-disciplinary efforts to enhance climate resilience and sustainable urban development in the United States.

HIGHLIGHTS.

• There are an increasing number of heat-exposed populations in the continental US by 2050.

• The change of heat-exposed populations is predicted by ethnicity and income level.

• Global warming will exacerbate social inequality by 2050.

• People living in large metropolises are predicted to experience more social inequality.

• The poverty group will experience the greatest impact of heat, followed by Black and Asian.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.jclepro.2025.146767.

Data availability

Provided in the Data Statement.