Risk drivers

Scenario projection of the key drivers of climate-health risks

Annual average of daily mean temperatures and number of hot days

Annual average PM2.5 concentrations

Total city population, by age group

Health impacts

Projected health impacts from heat and air pollution, alongside the climate and demographic drivers behind them

Improve resilience

Impact of interventions on health outcomes with benefits on avoided healthcare costs and emissions, alongside total deployment costs

Improve underlying health

Reducing high BMI prevalence will reduce population vulnerability to climate change and reduce pressure on the healthcare system, making it more resilient

Technical methodology

This tool quantifies climate-related health risks in cities worldwide, and the benefits of acting on them. It covers two hazards: extreme heat and ambient air pollution (fine particulate matter, PM2.5). It runs at neighbourhood resolution, because exposure, vulnerability and health systems vary within a single city.

This methodology was developed and the analysis carried out by Mode Economics in partnership with the Yale School of Public Health.

The model works in three stages. First, it sets the risk drivers; who is exposed, to which hazard, and how that changes to 2050. Second, it converts exposure into health impacts, namely mortality and hospitalisations, using the baseline disease burden and hazard–health response functions. Third, it tests interventions that improve resilience, under a range of deployment scenarios. It then estimates the healthcare costs and healthcare-related emissions avoided by the hospitalisations prevented. Alongside these, it reports the health burden associated with high body-mass index (BMI), as a measure of underlying population vulnerability.

Heat and air pollution are modelled and reported separately.

1. Risk drivers

Risk drivers are the inputs that determine how much hazard each population faces. We calculate exposure to them for every neighbourhood area and project them to 2050.

Spatial resolution

Results are modelled and displayed for every neighbourhood, but the inputs behind them are resolved at different scales. Some vary neighbourhood by neighbourhood; others carry a single city-wide value, or a national one.

Population

Population comes from WorldPop constrained 100-metre estimates for 2015–2030 (Bondarenko et al., 2025). WorldPop distributes each country's latest census counts across its built-up area, and constrains every year's national total to the UN World Population Prospects 2024 (WorldPop, 2025). We sum the grid cells falling within each neighbourhood, from a base year of 2023. Beyond 2030 we estimate population growth along the SSP2 trajectory (Wang, Meng and Long, 2022), anchored to the 2030 WorldPop baseline and adjusted for city ageing using country age structures from the UN World Population Prospects (2024). Population is split by sex into three age groups: under 5, 5 to 64, and 65 and over.

Because national totals follow the UN projection, a city's trajectory can differ from its recent census trend or from official local projections. We use global data so that one method covers every city.

Neighbourhood boundaries are the units each city plans with, taken from national statistics offices, municipal open data or UN OCHA administrative datasets; the city boundary is their aggregate. The unit differs by city, for example alcaldías in Mexico City, bairros in Rio de Janeiro and boroughs in London. The source for any city is available on request.

Heat risk

Daily temperatures come from NASA's NEX-GDDP-CMIP6 downscaled projections for 2020–2050, under the SSP2-4.5 scenario (Thrasher et al., 2022). We use a five-model ensemble to average out volatility in individual climate models. Daily mean near-surface air temperature is aggregated across each city boundary and applied to the neighbourhoods within it.

Urban heat island: Dense, built-up neighbourhoods within the city run hotter than the surrounding areas, often called the urban heat island effect. We add a satellite-derived urban heat island adjustment for each neighbourhood, taken from a global canopy-level dataset of urban–rural air temperature differences (Yang et al., 2024). The adjustment is a monthly climatology, so it varies through the year as well as between neighbourhoods.

Air pollution hazard

Ambient PM2.5 comes from the ACAG V6.GL.02.04 satellite-derived data set (van Donkelaar et al., 2021; Shen et al., 2024), population-weighted within each neighbourhood. Because global air-quality projections are highly uncertain given the role of anthropogenic change, we project PM2.5 to represent a continuation of current trends. This implies that current policy momentum continues, i.e. in cities where PM2.5 has been decreasing due to policy and market forces, we assume that this trend continues in line with increasing electrification, cleaner energy, modal shifts etc. We project each city forward with its own linear trend, fitted to its observed PM2.5 over 2000–2022. Outliers and the COVID-affected years 2020 and 2021 are excluded, because they are not representative of the underlying trend. We implement a floor for the projection at 5 µg/m³, the WHO annual guideline (World Health Organization, 2021), as few urban areas in the world currently have pollution levels below this due to a concentration of human activity. Reliable projections to 2050 require bottom-up data for each city, and for this you could use resources such as those developed by C40.

2. Health impacts

We convert heat and air pollution exposure into mortality and hospitalisations by linking the projected future level of mortality and hospitalisations to the modelled change in exposure to each hazard.

Projected hospitalisations and mortality

We use country-level current mortality rates based on the Global Burden of Disease (GBD) 2023 study (GBD 2023 Causes of Death Collaborators, 2025). Hospitalisation rates are proxied based on inpatient utilisation estimates (Moses et al., 2019). Hospitalisation rates are then adjusted from the national level to an urban one, using ILO estimates of the share of the population not covered because of health staffing shortfalls and because of financial barriers, weighted equally; mortality rates are used at the national level. Where granular data is publicly available for hospitalisations and mortality, we use that instead e.g., DATASUS for Brazilian municipalities, where recorded admissions are scaled up to all-sector estimates from the National Health Survey (Stopa et al., 2023), and CDC WONDER and HCUPnet for cities in the United States.

We then combine the baseline rates of mortality and hospitalisation obtained above with the demographic projections to produce city-level mortality and hospitalisation rates.

Finally, we project mortality rates to 2050 with the IHME GBD Foresight mortality forecast (GBD 2021 Forecasting Collaborators, 2024). This already embeds expected changes in income, health systems and the epidemiological transition, so we do not re-model them. As no equivalent dataset exists for hospitalisations, the rate for each demographic group (i.e. age and sex) is held flat at current levels.

Temperature–health relative risks

Heat risk is modelled with published temperature exposure–response curves. These curves show the excess risk of mortality, relative to an ambient day, at each daily temperature, and are specific to each age group. The risk varies by city, due to existing physiological and sociological adaptation to heat, and by age, as older adults are more vulnerable to the health impacts of heat. Each curve has a minimum-mortality temperature, below which no heat risk is attributed.

Where a city matches one of the 854 European cities in Masselot et al. (2023), we use that city's curve, because a city-specific curve captures local acclimatisation. Masselot publishes five adult age groups; we combine these into the tool's 5 to 64 and 65 and over groups, weighted by population. Three cities outside Europe — Sydney, Melbourne and Christchurch — are assigned the European city whose curve shape best matches their own climate. Every curve is then re-anchored onto the city's own modelled temperature distribution, so it is the shape of the response that transfers rather than the absolute temperatures. For the remaining cities and for the under-5 group, we use climate-zone specific exposure–response curves (Burkart et al., 2021), with climate zones assigned from the Köppen-Geiger classification recorded for each urban centre in the JRC Global Human Settlement Layer Urban Centre Database (European Commission JRC, 2024). Beyond the range each curve is fitted over, the heat tail is extended log-linearly rather than held flat, following established practice in the temperature–mortality literature (Gasparrini et al., 2017; Vicedo-Cabrera et al., 2019).

We use 17 heat-sensitive causes identified by Burkart et al. (2021): ischaemic heart disease, stroke, hypertensive heart disease, cardiomyopathy and myocarditis, chronic obstructive pulmonary disease, lower respiratory infections, chronic kidney disease, diabetes mellitus, and nine injury causes (drowning, road injuries, other transport injuries, animal contact, exposure to mechanical forces, other unintentional injuries, self-harm, interpersonal violence, and exposure to forces of nature). Under 5, every injury cause except exposure to forces of nature is excluded: these act through activity patterns rather than heat stress, and no modelled intervention prevents them. We combine them into a single all-cause attributable fraction for each population group, weighting each cause by its share of mortality. Hospitalisations are approximated by applying the same attributable fraction to baseline hospitalisations.

Air pollution–health relative risks

Air pollution risk is modelled with PM2.5 concentration–response curves, which provide a relative risk that varies with the annual-mean concentration. These curves show the excess risk of mortality relative to a theoretical minimum-risk exposure level (TMREL) of 4.15 µg/m³, which defines the concentration of PM2.5 below which there are no health effects (GBD 2019 Risk Factors Collaborators, 2020). These curves do not differ by city, as people and health systems cannot adapt to the health impacts of air pollution the way that they can for heat. We use the IHME GBD 2023 curves for the six causes with established PM2.5 links: ischaemic heart disease, stroke, COPD, lung cancer, type 2 diabetes and lower respiratory infections. We combine them into a single all-cause attributable fraction for each population group, weighting each cause by its share of mortality. This lets the strength of the relationship vary by age and sex for each city. Hospitalisations are approximated by applying the same attributable fraction to baseline hospitalisations.

3. Improve resilience

We model interventions with evidenced reductions in heat or pollution risk. Each intervention is targeted where the need and the potential for impact are greatest. We model a range of deployment assumptions, calibrated against existing comparable programmes. For each intervention we set out how its deployment is constrained and adapted to local context, how effective it is in reducing the risk or health impacts, and how it is deployed and costed.

Heat

Street trees

Street trees reduce ambient air temperatures, reducing the temperatures that people are exposed to and the resulting health impacts. Street trees also have co-benefits related to health for air quality, flooding and wellbeing, although these are not modelled here.

  • Effectiveness. Added canopy lowers air temperature by 0.08°C for each percentage point of additional canopy cover, applied uniformly across cities (Zaerpour et al., 2025). Tree growth and survival are modelled over time, so cooling builds up gradually rather than arriving at once. Mature canopy size per tree varies by climate zone; polar climates are excluded, as street trees are not a viable heat intervention there.

  • Feasibility constraints and contextual amendments. We assess existing canopy, mapped at 10-metre resolution from ESA WorldCover. We estimate each neighbourhood's plantable area as its paved, non-building surfaces, such as roads and car parks. This is found by taking the built-up area and subtracting building footprints. We then scale it down by a realism factor to reflect how much of that space can practically be planted (ESA WorldCover v200 (Zanaga et al., 2022); GHSL Built Surface, GHS_BUILT_S (Pesaresi and Politis, 2023)). New canopy is allocated first to the neighbourhoods that need it most, ranked on three equally weighted factors: population density, mean temperature including the urban heat island effect, and how little canopy they already have. Planting in any neighbourhood is capped at 25% canopy cover.

  • Deployment and costs. Planting begins in 2027 and is phased over seven years, with canopy continuing to grow afterwards and the full canopy target reached around 19 years after the programme starts. Deployment scale is calibrated against the size of real street-tree programmes (C40 Cities, 2019; Jones et al., 2024). In arid cities, currently Ciudad Juárez and Monterrey, deployment is halved: the plantable-area layer counts paved ground that cannot sustain a street tree, and irrigation water is scarce. Costs use a single street-tree unit cost, taken from a UK street-tree cost-benefit analysis: $2,916 of capital cost per tree planted and $45 a year of maintenance per surviving tree, both at 2018 prices and restated to 2024 dollars (about $3,643 and $56) before being adjusted to each city's country price level using PPP (GreenBlue Urban and Treeconomics, 2018; Panduro et al., 2024).

Cool roofs

Cool roofs lower indoor and ambient temperature, reducing the temperatures that people are exposed to and the resulting health impacts (Arup, n.d.; Greater London Authority, 2023).

  • Effectiveness. Cool roofs reduce the temperature occupants experience by around 4-5% in tropical, arid and temperate climates (Synnefa et al., 2007), which feeds the same temperature response curves used for the baseline. The intervention is modelled only in those zones, and no city in the tool lies outside them.

  • Feasibility constraints and contextual amendments. Eligibility is limited to low-rise residential buildings below 10m high, identified from the GHSL building-volume and surface layers. We exclude the share of households with air conditioning based on Falchetta et al., 2024. Cool roofs are prioritised towards neighbourhoods that are lower-income (Wang and Sun, 2022), hotter (including the urban heat island effect), and have more residents aged 65+, weighted equally. Deployment is capped at half of the eligible population to reflect feasibility and access constraints, and further to the share of that stock whose roofs can physically take a coating, which follows each country's typical roof type (from flat concrete slab, where most roofs qualify, to pitched tile, slate and asphalt shingle, where few do). The ambition levels themselves are expressed as a share of the whole city population — from 0.8% to 2.5% of residents by 2050 — and the slider reports coverage on that basis. In Barcelona, Berkeley, Broward County, London and Medellín the coatable pool is exhausted before the highest levels are reached, so those levels add no further coverage. In Barcelona it binds at every level, holding coverage near 0.5%.

  • Deployment and costs. Deployment begins in 2027 and is phased over five years. A single reflective coating specification is applied in every city, costed at $5/m² of materials and $7/m² of labour and spread over a six-year recoating cycle. Labour is adjusted to local price levels using PPP; materials are commodity-priced and are not adjusted. Unit costs are drawn from US consumer and trade pricing sources. Costs are charged only for the roofs actually treated.

Community health education

Community health workers reduce heat-related illness and mortality by educating residents, helping them recognise heat symptoms early, and supporting vulnerable households to adopt protective behaviours.

  • Effectiveness. We model a 41% reduction in the excess relative risk of heat-related mortality for people aged 65 and over, taken from a social-support programme for older adults in Rome (Liotta et al., 2018). In lower income cities an odds ratio of 0.62 is applied to heat-related hospitalisation, based on a randomised controlled trial in Pakistan (Razzak et al., 2022). In higher income cities, the hospitalisation effect is assumed to be the same as the mortality effect for over-65s.

  • Feasibility constraints and contextual amendments. The mortality effect applies to residents aged 65 and over in all cities, based on the Long Live the Elderly programme in Rome (Liotta et al., 2018). In lower income cities an additional hospitalisation effect applies across all ages based on a randomised controlled trial in Pakistan (Razzak et al., 2022); and in higher income cities the hospitalisation effect is only applied to the over-65s.

  • Deployment and costs. The programme is phased in over 2026–2030, after which it runs each year at full effectiveness. Costs are set per person reached: assumed to be a lower rate in lower income cities ($6 per capita) and a higher rate in higher income cities ($80 per capita), adjusted for PPP.

Early warning systems

Heat warnings cut mortality and hospitalisations by prompting households and institutions to act before extreme-heat events.

  • Effectiveness. The reduction is applied on days above the local minimum-mortality temperature, in four steps ranging from about 13% to 22%, with the largest reductions at the most extreme temperatures, following evaluations of the Ahmedabad Heat Action Plan (Hess et al., 2018). The same reduction applies to hospitalisations as to deaths, across all age groups.

  • Feasibility constraints and contextual amendments. We account for cities that already have a warning system. Where a city has an EWS from before 2023, we do not apply the intervention, as the effect is already included in the baseline. This applies to Greater Manchester, London, Melbourne, Surat, and Sydney. Where a city has an EWS in place set up after 2023, we model the benefits and OPEX costs, but do not include CAPEX costs for setting up the programme, which applies to Rio de Janeiro.

  • Deployment and costs. The deployment year is determined by the level of ambition selected by the user. Capital costs are assumed to be a fixed value of $500,000, based on a triangulation of existing warning systems and expert assessment driven by, first, the indicative cost of a technical assistance programme and, second, that cities with lower PPP could face higher set-up costs. Operating costs are modelled at $0.05 per capita, based on a range of sources including the UNDRR/WMO Early Warnings for All programme and Uejio et al. (2011).

Rest, shade, hydrate

Outdoor workers cannot easily avoid heat. Rest, shade and hydration measures lower their risk of heat-driven health impacts.

  • Approach and effectiveness. The policy reduces the excess relative risk of heat-related mortality and hospitalisations by between 33% in the most informal economies and 51% in the most formal, based on literature assessing impacts of these programmes (Dean and McCallum, 2025). The reduction varies with labour-market informality. More-informal economies are assumed to have lower adherence to the policy as it is harder to enforce labour protections in informal settings through fines or penalties, so a smaller reduction is applied.

  • Feasibility constraints and contextual amendments. We estimate the target population of outdoor workers in each city based on the share of the working-age population in outdoor-exposed sectors who work outside. This is calculated using ILO data for the working-age population and share of urban employment per sector, and the US Bureau of Labor Statistics Occupational Requirements Survey, which reports outdoor exposure by occupational group (Bureau of Labor Statistics, 2024). Three sector groups are included: construction, where 91% of workers are treated as working outdoors, the survey's figure for construction and extraction occupations; mining, quarrying and utilities, at 55%; and trade, transport, accommodation, food and business services, at 22%. The latter two are blended from the occupational groups that make up each sector. Agriculture is not included. The sector mix and the outdoor shares are held constant over time, so the worker count moves only with the population forecast.

  • Deployment and costs. Coverage is phased over 2026–2030. Cost is charged per treated worker and adjusted by PPP to each city: $13 a year of operating cost, plus $20.50 of capital cost in the first deployment year and every five years thereafter as the programme is refreshed. Productivity costs are excluded, as we do not assess productivity benefits (Washington State Department of Labor & Industries, 2023; Occupational Safety and Health Administration, 2024).

Air pollution

Reducing road-transport emissions

The road transport intervention models the impact of a suite of policies to reduce PM2.5. In high-income cities this could typically include ULEZ, modal shifts to increase active and public transport, and electrification. In middle- and lower-income cities this could typically include vehicle and fuel standards, modal shifts to increase public transport use, and bus electrification.

  • Approach and effectiveness. The intervention is modelled as a reduction in the share of PM2.5 driven by road transport, between 15% and 35% for higher income cities, and 20% to 45% for lower income cities. These ranges are based on a triangulation of existing studies of the effectiveness of these interventions (Fraser et al., 2025; Greater London Authority, 2025; Simeonova et al., 2021; Chamberlain et al., 2023; World Bank, 2025; Pan et al., 2023). The achievable reduction differs by city context, because lower income cities typically have older, higher-emitting vehicle fleets, leaving more exhaust emissions to abate, whereas higher income cities have often already implemented many of these policies.

  • Feasibility constraints and contextual amendments. The road-transport share of ambient PM2.5 is taken from WHO source-apportionment data for the city's world region (Karagulian et al., 2015). The model never takes a city below the WHO annual guideline of 5 µg/m³.

  • Deployment. The reduction is assumed to be driven by a combination of policy and market forces, and so is phased in from 2026, reaching its full effect in 2045. Implementation cost is out of scope for this intervention, due to the variety of policy instruments that could be used to achieve this reduction, and their context-specific costs.

Self-care during pollution events

On the most polluted days, people can cut their own exposure using well-fitted masks, sheltering indoors with windows closed, reducing strenuous outdoor activity, and by using portable air purifiers. We model this as a reduction in exposure on the most polluted days for each city, adjusted for the uptake of these protective behaviours in response to public health advice.

  • Approach and effectiveness. As the concentration response function is based on annual concentrations, we model the annual distribution of pollution, centred on the annual mean from the linear projections from the GHAP satellite daily PM2.5 data for 2017–2019 (Wei et al., 2023). We assume a constant distribution to 2050 and use this to identify the number of days considered high risk. We see in the literature that people who are more at risk are more likely to adopt protective behaviours and so our modelled uptake is higher among those aged 65 and over at 45% and lower for everyone else at 22% (D'Antoni et al., 2017). People who adopt protective behaviours reduce the part of their exposure that sits above the alert threshold by between 20% and 40%, based on a triangulation of studies measuring the impact of these types of interventions (Carlsten et al., 2020; Allen et al., 2011; Xu et al., 2020; Cherrie et al., 2018).

  • Feasibility constraints and contextual amendments. People are modelled to adopt the behaviours only on the worst air pollution days. This is due to 'alert-fatigue' and city-specific perceptions of risk from air pollution. The alert threshold is the higher of the 95th percentile of the city's daily concentrations and 25 µg/m³, the WHO 24-hour interim target (World Health Organization, 2021). In cleaner cities the 25 µg/m³ floor binds, so fewer than 5% of days qualify, and where no day exceeds it the intervention has no effect.

  • Deployment and costs. The intervention applies from 2027, with no ramp, assuming that the city starts issuing alerts that result in people adopting these behaviours. Implementation cost is out of scope for this intervention.

Healthcare avoided costs and emissions

We translate the avoided hospitalisations into avoided healthcare spend and avoided healthcare emissions. Avoided deaths are reported but are not costed.

Hospitalisation costs

Each avoided hospital episode is costed as a sum of the inpatient stay (based on the WHO-CHOICE cost per bed-day times an average length of stay), the related outpatient and primary-care visits, and a proportionate allowance for drugs, diagnostics and imaging (Stenberg et al., 2018). Length of stay is the country's all-cause average where OECD reports one, which covers about half the cities in the tool and ranges from roughly 4 to 8 days; the remaining cities use a 7-day default, the OECD median (OECD, 2024). WHO-CHOICE unit costs are price-adjusted to 2024 US dollars.

Emissions

Avoided emissions are calculated as the emissions intensity of the health care system, in kilograms of CO2-equivalent per dollar of health spend, multiplied by the avoided hospitalisation costs. Intensities come from Lenzen et al. (2020) where available. Where a country has no health-sector value we use its whole-economy emissions intensity, calibrated by the median ratio of health-sector to whole-economy intensity across the countries Lenzen et al. do cover; where neither is available we fall back to the income-group values of Rasheed et al. (2021).

4. Improve underlying health

Alongside heat and air pollution, the tool reports the health burden associated with high body-mass index (BMI). High BMI is not a climate hazard, but it shapes how exposed a population is to one. It raises the underlying rate of the cardiovascular, metabolic and respiratory conditions that heat and air pollution act on, and it adds pressure to the health system that has to absorb those shocks. Lowering it is a way for a city to reduce climate-health risk indirectly, and over a longer horizon than the heat and air pollution interventions work on.

Population exposure to high BMI

We estimate the whole distribution of BMI for each country, by sex and five-year age band, for every year to 2050. Health risk rises steeply at the upper end of the range, so the shape of the distribution matters as much as its centre. The shape is fitted to the banded BMI tabulations published by the NCD Risk Factor Collaboration (NCD-RisC). We use obesity prevalence estimates from the GBD studies (GBD 2021 Adult BMI Collaborators, 2025; GBD 2021 Adolescent BMI Collaborators, 2025), which run to 2050, as an input to anchor the share of adults with a BMI of 30 or above.

City exposure is scaled from the national distribution using the ratio of urban to rural mean BMI reported by NCD-RisC (2019), so cities in countries where urban populations are heavier than rural ones carry a correspondingly higher distribution. Children and adolescents aged 5 to 19 are handled separately, using BMI-for-age z-scores against the WHO 2007 growth reference, with overweight and obesity defined at one and two standard deviations above the reference median.

Attributable burden

Exposure is converted into attributable mortality and hospitalisations using published relative-risk curves for high BMI, integrated across the fitted distribution, with risk measured against a minimum-risk BMI of 22.5 (Afshin et al., 2017). The resulting attributable fraction is applied to the city's projected all-cause mortality and hospitalisation rates, on the same basis as the heat and air pollution pathways, and is reported across the same three age groups used throughout the tool: under 5, 5 to 64, and 65 and over. No high-BMI pathway is modelled below age 5.

Climate-sensitive conditions

Each condition contributing to the high-BMI burden is flagged by whether it is also sensitive to climate, so the share that overlaps with heat and air pollution risk can be read on its own.

Following are treated as climate-sensitive: ischaemic heart disease, stroke (ischaemic stroke, intracerebral haemorrhage and subarachnoid haemorrhage), hypertensive heart disease, type 2 diabetes, and chronic kidney disease (from type 2 diabetes, from hypertension, from glomerulonephritis and from other or unspecified causes). These are the cardiovascular, metabolic and renal conditions Burkart et al. (2021) also links to non-optimal temperature.

The remainder are treated as not climate-sensitive: atrial fibrillation and flutter, gallbladder and biliary disease, and the obesity-related cancers — oesophageal, colon and rectum, liver (hepatitis B, hepatitis C and alcohol-related), gallbladder and biliary tract, pancreatic, breast, uterine, ovarian, kidney and thyroid cancer, multiple myeloma, and leukaemia (acute and chronic lymphoid, acute and chronic myeloid, and other).

References