Heat Stress, Labor Productivity, and the Hidden Economic Cost of Warming

Climate damages are most visible as events: a flood that destroys homes, a storm that grounds a supply chain, a drought that wipes out a crop. Disaster databases and insurance losses count these. Heat's effect on work is different. It operates every hot afternoon, in every field, construction site and unventilated workshop, and it is self-protective: people slow down because continuing at full pace would make them ill. The International Labour Organization estimated in 2019 that heat stress would cut global working hours by 2.2% by 2030, equivalent to 80 million full-time jobs and US$2.4 trillion, even if warming stayed at about 1.5 °C, with agriculture accounting for roughly 60% of the loss and South Asia and West Africa losing about 5% of hours.

This loss is easy to miss in economic statistics. Much of it falls on informal and self-employed workers, especially in farming, whose reduced effort shows up only as a smaller harvest or lower earnings. It is spread thinly over long periods rather than concentrated in a dated event. And it depends on local heat and humidity during working hours — something an annual average temperature does not capture. A day can be mild on average and still have four hours in the afternoon in which heavy work becomes unsafe.

That is why aggregate climate-economy models struggle to represent it. The DICE model, for example, expresses all climate damages as a single quadratic function of global mean temperature, calibrated on studies of aggregate damages; the 2016 version implies about 2.1% of world output at 3 °C. Such a function has no explicit way to respond to humidity, to the hours in which people work or to the sectoral mix of an economy, and so cannot show where heat's labour costs fall. Empirical work points the same way: Burke, Hsiang and Miguel found that economic productivity peaks at an annual average temperature of about 13 °C and declines sharply above it, which places most poor countries on the steep side of the curve.

Occupational health measures heat stress with the wet-bulb globe temperature, which combines air temperature, humidity, radiant heat and wind into a single index used in international standards for safe work. Kjellstrom and colleagues fitted curves linking WBGT to the share of normal work capacity that can be sustained, based on field and laboratory studies of gold miners, rice harvesters and people working in climate chambers. The curves, used by the Lancet Countdown and in economic studies such as Orlov and colleagues', differ by the intensity of the work: heavy work of about 400 watts, such as farming and construction; moderate work of about 300 watts, such as manufacturing; and light work of about 200 watts, typical of services.

For heavy work, capacity begins to fall noticeably above a WBGT of about 25 °C, is roughly halved at 31 °C and reaches the curves' floor — 10%, the share of time assumed always possible in short bursts — in the high 30s. Moderate and light work tolerate a few more degrees. The steepness is the point: a few degrees of extra heat at the top of the range remove far more work than the same few degrees lower down, which is why losses grow faster than temperature.

The curves are approximations. The fit to the underlying data is not tight, the evidence comes from a small number of settings, and real workers adapt by changing pace, timing and clothing. They also take no account of acclimatisation, which helps, or of the health damage from working through heat, which hurts. The slider shows how capacity changes across the range of WBGT recorded in the sampled cities in 2024; the highest hourly value, in Dubai, reached 34.6 °C even in the shade.

For each city, the shade WBGT is estimated for every hour of 2024 as 0.7 times the wet-bulb temperature plus 0.3 times the air temperature, and the capacity loss is averaged over a standard working day of 08:00 to 17:00 on every day of the year. Dubai lost 14% of heavy-work capacity across the year, Jeddah 12.6%, Kolkata 12.2% and Muscat 12.0%; Bangkok, Chennai, Phnom Penh, Ho Chi Minh City, Dhaka and Yangon all lost between 9.6% and 11.2%. Moderate work lost between a third and a half as much, and light work a tenth or less, because their thresholds are higher.

The grid for each city shows where the loss comes from: every hour of the day across the fortnights of 2024, shaded by heat stress, with the standard working day outlined. In Kolkata the danger runs from March to October, peaks in June — when heavy work loses a third of its capacity — and is greatest around late morning and early afternoon; nights are hot but cooler than the working day. In the humid tropics, hours of high heat stress stretch across most of the year, while in the Gulf they are concentrated in a long, intense summer.

Humidity matters as much as temperature. Dry-heat cities such as Cairo, Delhi in May or Niamey reach higher air temperatures but lower shade WBGT than humid coastal cities, because sweat evaporates easily; the shade estimate therefore understates their risk for people working in the sun, where radiant heat dominates. Altitude matters too: highland capitals such as Mexico City, Bogotá and Addis Ababa show almost no loss, although their countries' lowlands do — a reminder that the city sample stands in for national conditions imperfectly.

Weighting each city's losses by the country's mix of jobs and output — agriculture as heavy work, industry as moderate, services as light — gives national estimates for 123 countries. In 2024 about 1.65% of their labour force's working hours were lost to heat, the equivalent of 61 million full-time workers, and about $492 billion of output, 0.43% of their GDP, was at risk. Farm workers bore 74% of the lost hours; because farm output per worker is low in the hottest economies, agriculture accounts for a smaller 35% of the money lost, and industry, including construction, for half.

The burden is strongly regressive. Lower-middle-income countries lost 3.0% of working hours and 1.8% of GDP; low-income countries, whose sampled cities are mostly dry Sahelian capitals, 1.2% of hours; upper-middle-income countries 1.3%; and high-income countries 0.19%. As a share of GDP the heaviest losses fall in the United Arab Emirates (4.3%), Myanmar (3.9%), Cambodia, Niger and Oman (about 3.6%), Saudi Arabia (3.2%), Pakistan and Vietnam (2.9%), Bangladesh (2.7%), Thailand (2.5%) and India (2.0%). In money, China ($157 billion) and India ($81 billion) dominate, followed by Saudi Arabia, Indonesia and the UAE.

These figures are close in pattern to independent estimates. The Lancet Countdown's 2025 report puts India's 2024 loss at 247 billion potential labour hours and about $194 billion in potential income, two-thirds of it in agriculture — more than double our India estimate, because it counts work in direct sun and the whole rural workforce. That gap is a measure of how much a shade-based, city-based estimate understates the total, and of how much of the cost lies outside cities.

Integrated assessment models translate warming into economic damage through a damage function. In DICE, the best known, damages are a single quadratic function of global mean temperature: a coefficient of 0.00236 in the 2016 version and 0.003467 in the 2023 version. At 1.5 °C of warming these imply total climate damages of 0.53% and 0.78% of world output; at 3 °C, 2.1% and 3.1%. Those totals are meant to include every channel — agriculture, sea-level rise, health, energy and the rest.

Set beside them, the cost of heat to work alone is large. Our conservative 2024 estimate is 0.43% of the GDP of the countries covered; the Lancet Countdown's broader estimate of lost labour income is almost 1% of world GDP. These are not like-for-like quantities — the labour figures measure the total cost of heat, including the heat that would exist without climate change, while the damage function measures only the cost of warming — but the comparison shows the scale of a channel that a global-mean damage function does not explicitly represent. Section 6 shows that the warming-related part is itself large and grows faster than temperature.

The deeper problem is structural. Labour losses depend on humidity, on hours of the day, on the share of heavy outdoor work and on income levels that determine whether people can afford to stop. A function of global mean temperature cannot reflect any of these, and so tends to understate damages where they are most concentrated — in hot, humid, labour-intensive economies whose people have contributed least to warming. Studies that build heat–labour effects into economic models, such as Orlov and colleagues', find substantial costs that rise steeply under high-emission pathways.

To see how the loss responds to warming, every hour's WBGT in every city is shifted up or down by the same amount and the calculation repeated. This is a simplification — warming is not uniform, and humidity changes too — but it shows the shape of the response. Removing 1 °C from every hour would cut 2024's output at risk from $492 billion to $274 billion, by 44%; removing 1.5 °C, to $202 billion. Adding 1 °C raises it to $851 billion, 2 °C to $1.41 trillion and 3 °C to $2.26 trillion, almost 2% of the covered GDP.

The increments grow: the first added degree costs about $360 billion a year, the second about $560 billion and the third about $840 billion. That convexity follows directly from the shape of the capacity curves — more and more working hours cross into the steep part of the curve — and it is why the labour channel matters more, not less, as warming proceeds. Lost hours follow the same path, from 61 million full-time equivalents today to 100 million with an extra degree and 157 million with two.

The shift is applied to WBGT, which in humid conditions rises roughly as fast as air temperature. Land and cities warm faster than the global mean, so a degree of WBGT corresponds to less than a degree of global warming in many hot regions. The numbers are therefore best read as sensitivities: they show that the cost of the heat already present is large, that the part removed by taking a degree away is also large, and that further warming would raise it steeply.

Because heat stress peaks in the early afternoon, when the work is done matters as much as how hot the day is. Starting the working day at 06:00 instead of 08:00, with the same nine hours, cuts the 2024 loss from $492 billion to $392 billion, by 20%. A split shift — 06:00 to 11:00 and 15:00 to 19:00, avoiding the hottest four hours — cuts it to $343 billion, by 30%, and reduces lost hours from 61 million to 42 million full-time equivalents. The gains are similar across income groups: 16–22% for an early start and 30–40% for a split shift.

Time is not the only lever. Shade, ventilation and cooling at indoor workplaces protect moderate and light work; mechanisation reduces the share of heavy outdoor work; and work–rest schedules, water and protective clothing let people keep working more safely. The ILO's 2019 assessment stressed that these measures need to be combined with social protection, because the workers most exposed — informal, agricultural and self-employed — are the least able to stop without losing income.

Many of these measures already have legal form. A growing number of countries and states set maximum working temperatures, mandatory midday breaks during summer months, or heat-illness prevention rules for outdoor work, and heat-health warning systems can trigger them. Their limit is coverage: rules rarely reach informal work, and schedules can shift only so far when nights are hot too. That is why adaptation reduces, but does not remove, the case for limiting warming: the curves in section 6 show that the losses rise fastest exactly where adaptation is hardest.

Hourly 2-metre air temperature and wet-bulb temperature for 2024 come from the ERA5 reanalysis via the Open-Meteo archive API, in local time, for 149 large cities chosen as in this series' heat and air-pollution report (the largest city in every country of 5 million or more, plus up to seven others in the most populous countries). Shade WBGT is approximated as 0.7 × wet-bulb + 0.3 × air temperature, treating the natural wet-bulb as the psychrometric wet-bulb and the globe temperature as air temperature; this omits solar radiation and therefore understates WBGT in the sun.

Work-capacity loss uses the Hothaps exposure–response functions (Kjellstrom et al. 2018; Orlov et al. 2020) for heavy (≈ 400 W), moderate (≈ 300 W) and light (≈ 200 W) work, averaged over the working hours of every day of 2024. Country losses are city-population-weighted averages. Agriculture is treated as heavy work, industry as moderate and services as light. Hours lost = Σ employment share × capacity loss; output at risk = Σ value-added share × capacity loss, applied to latest GDP in current US$ (World Bank). Full-time equivalents apply the hours share to the labour force.

The estimate leaves out work in direct sun, rural workers outside sampled cities, heat illness and death, school and cognitive effects, and the ways employers and workers adjust; it assumes output falls in proportion to capacity. Warming scenarios shift every hour's WBGT uniformly. DICE values are computed from published damage coefficients. Comparisons with the Lancet Countdown and ILO are of scale, since methods differ. Related reports: the geography of survivability under climate stress and the compounding burden of heat and air pollution.