Our analysis indicates that rising urban temperatures will have profound implications for the future livability of cities and confirms previous studies on this topic3,7. Under SSP5-8.5, up to 217 of today’s largest cities (>300,000 inhabitants) could exceed the 29 °C MAT threshold by 2071-2100, affecting (sim)320 million people. Cities sort into three recurring risk profiles: (1) already-hot megacities where exceedance intersects with very large exposed populations (e.g., Mumbai, Chennai); (2) temperate but dense legacy cities facing steep background warming despite slower growth; and (3) hot-and-growing cities with constrained adaptive capacity, where rapid urbanization and informality amplify exposure.
Across threshold-exceeding cities, intra-urban morphology systematically modulates exposure. From 217 cities, 61.3% have bare natural surfaces above their continent-specific median, 57.6% have green-area fractions below the median, and 58.1% have low-plant shares below the median. LCZ profiles of exceeders show higher shares of bare natural surfaces and lower vegetated fractions relative to non-exceeders (Fig. 2), consistent with greater heat absorption and reduced evapotranspirative cooling. In Africa, exceeders tend to have markedly lower green-area shares; in Asia, the reduction is concentrated in low-plant LCZs. Conversely, cities with larger pervious and vegetated fractions show damped exposure—pointing directly to morphology-leveraged adaptation opportunities.
In dense, cooler-climate cities (e.g. in Europe or North America), the median projected MAT increase exceeds 4.1 °C by 2100 under SSP5-8.5 (Fig. 2)—a large shift for infrastructures designed for milder conditions14. In many hot-and-growing cities, baseline MATs are already near or above threshold; rapid population increases magnify exposure—e.g., up to 600% growth in some cases (and 1500% under SSP3-7.0). Even where populations stabilise or decline (e.g., some North American and South American cities), high densities mean that a (sim) 4 °C rise in MAT will still exacerbate urban heat risks.
Delhi’s large informal-settlement share15 heightens vulnerability as background heat rises; Kabul’s potential expansion (up to 400%, scenario-dependent) combines rapid demographic change with increasing thermal stress. Madrid and Belgrade face steep warming atop dense, legacy-built fabrics. Toronto’s exposure grows with or without population increase due to high population numbers. Santiago’s dense informal areas face amplification under notable warming16. These examples reflect the three profiles mentioned above.
Linking LCZ diagnostics to our adaptation results (Fig. 6A), four measures consistently rank highest11: increase pervious-surface fractions; deploy reflective building materials in compact/dense LCZs; expand greenspaces/vegetation; and add or enlarge urban water bodies for evaporative cooling and co-benefits. Local feasibility varies: some Asian cities combine high MAT projections with limited GDP (Fig. 6B,C), constraining large-scale greening or retrofits. In wealthier metropolises (assuming political and societal will), reflective roofs and permeable pavements can scale rapidly. The contrast between Cherthala (high vegetative baseline, low pervious-surface score) and Kuwait City (desert-locked, high pervious-surface potential) illustrates how climate context, LCZ composition, and water resources shape feasible portfolios.
Thematically, the three profiles suggest different policy mixes: (1) for already-hot megacities, near-term heat action plans, reflective-surface mandates, and targeted greening in the hottest LCZs; (2) for temperate-but-warming dense cities, code updates for cool materials, retrofitting programs, and climate-fit species selection in parks; (3) for hot-and-growing, capacity-constrained cities, low-cost LCZ-targeted measures (cool coatings, shade trees, permeable lanes) and investments in basic services that also cool (water and sanitation). Further studies should also analyse which financing mechanisms best suit the different types of adaptation measures and risk profiles.
Together, these findings indicate that who is exposed (population trajectories), where exposure concentrates (LCZ patterns), and what is feasible (adaptive capacity) co-determine risk. Framing results by risk profiles avoids regional repetition and clarifies how morphology can either amplify or damp projected warming, while pointing to LCZ-informed, finance-aware pathways for implementation.
Limitations
It is important to acknowledge that the concept of the human climate niche is inherently complex and contested17, even if it remains useful for showing global trends. While the (hbox {MAT} =) 29 °C threshold provides a benchmark5, urban habitability is influenced by multiple interacting factors—including humidity, access to resources, and socioeconomic conditions—so results should be interpreted as indicative rather than deterministic.
Climate data Our study relies on downscaled climate projections (CHELSA, 1 km resolution), which are not designed to capture urban heat island (UHI) effects. As such, they may underestimate microclimatic variability where impervious cover and sparse vegetation amplify thermal stress. Local UHI spikes of 1–5 °C above surrounding areas3, when combined with humidity, can push MAT and wet-bulb thresholds beyond critical levels in low-greenness neighborhoods18. Although MAT >29 °C provides a broad niche indicator4, future urban analyses should incorporate wet-bulb metrics, humidity patterns, and high-resolution urban climate models or observations. Another limitation is that only a subset of CMIP6-SSP combinations is available at CHELSA’s native resolution; notably, SSP2-4.5 is missing. Inclusion of additional pathways would help test sensitivity to intermediate emissions scenarios.
Demographic projections Population estimates are drawn from downscaled SSP products for consistency with climate inputs, but demographic projections vary in both totals and spatial allocation. Alternatives such as WorldPop use different urbanization algorithms that could shift local exposure estimates19. Moreover, even current counts—particularly for residents of informal settlements—are incomplete15. Future work could improve coverage by integrating high-resolution slum maps, household survey data, and sensitivity analyses across multiple population datasets.
LCZ classification and land-use change Adaptation recommendations are tied to Local Climate Zone (LCZ) profiles, which capture the distinct morphological characteristics of urban neighborhoods. However, uncertainties in LCZ mapping stem from remote-sensing resolution and seasonal land-cover dynamics. More critically, LCZs are assumed static, although infill development, densification, or greening initiatives will change their distribution over time. Periodic updates and scenario testing are therefore needed to ensure that adaptation strategies remain aligned with evolving urban morphologies.
Adaptation scoring The assignment of measures to LCZ classes follows a binary scheme, which oversimplifies suitability. A more nuanced, weighted approach would better reflect gradations in effectiveness across settings. Additionally, we did not account for existing city heat action plans, which may already address some of the recommended measures.
Economic capacity Finally, we approximate adaptation capacity using total GDP per city. While useful as a first proxy, GDP does not capture institutional strength, governance quality, or socio-economic vulnerability. Incorporating per-capita indicators, governance metrics, and vulnerability indices could provide a more complete assessment of adaptive capacity.