How much hotter India has become, and what comes next.

Read across for years, down for months. Blue cells were cooler than the 1991-2020 normal for that month, ochre cells warmer, white cells close to it. The baseline is itself recent: a white cell is a month as warm as the 1990s and 2000s, already warmer than the 1940s.

ERA5 is a reanalysis: a weather model run backwards over every observation available (stations, balloons, satellites), producing a complete, consistent picture of each day. It is not a thermometer reading, and its early decades rest on far fewer observations than its recent ones.

Each panel counts days in a year above a line that matters for the body. The count is for an average spot: the days above the line in every quarter-degree cell of India's land, averaged. A count of 12 can mean 12 days everywhere or 60 days across a fifth of the country, mostly the Gangetic plain and the coasts.

The panels share no common scale: each is drawn to its own range, so read the change within a panel, not the height across panels.

Each line is one year's average temperature over India's land, minus that record's average for 1951-1980, so both start from the same zero. Year-to-year swings of half a degree are ordinary: El Niño years run hot, strong monsoons cool the summer. The signal is the drift underneath: flat for half a century, then climbing.

The warming rate is a straight line fitted through the years: the average change per decade over the period named, not a prediction.

For each month, the hollow dot is how much warmer its days have become, the blue dot how much warmer its nights. The monsoon months, June and July, warmed least: cloud and rain cap the heat. The dry season, from November to March, warmed most. A warmer winter shortens the cool spell wheat needs to fill its grain, which is why a hot February or March worries farmers more than a hot May.

Where the two dots sit close, the two instruments agree. Where they part, read the gap as uncertainty: the two records do not even agree on which regions warmed most. The station record, built from sparse gauges in hilly and remote states, leans on interpolation there; the reanalysis leans on its model. They also part in the irrigated north-west: canals and wet fields cool the afternoon there, aerosols from crop fires and traffic dim the sun, and the two records treat those effects differently. In Delhi the reanalysis finds days slightly cooler than in the 1950s and nights clearly warmer; the station record finds both warmer.

Two bars a state. The blue bar is how many more muggy days a year the state now has; the grey bar how many more (or fewer) scorching dry days. In the plains of the north, dry 40°C days have become fewer while wet-bulb days rose sharply: the heat did not leave, it took on water. Irrigation, which puts moisture into the air, is part of the reason.

Each bar is one year's rain. About four-fifths of it falls from June to September. A bar far below the line is a drought year, and in an economy where much of the farmland still depends on rain, it shows up within months in farm incomes, rural wages and food prices.

The bars are the recent average for each month, the dots the old one. A shift in timing matters to a farmer even when the total does not move: a late June delays sowing, a wet September can flatten a ripe crop. Monthly averages over the whole country cannot show the onset date; the India Meteorological Department dates onset and withdrawal, and those dates are not in these sources. We do not know from this data whether the monsoon arrives later.

An extreme downpour falls on a district, not on a country. Averaged over all of India, a cloudburst in one valley is diluted by dry days everywhere else, so national averages of extremes move slowly even when local floods grow more frequent. This national average can neither confirm nor rule out more frequent local downpours.

These are territorial emissions: the carbon dioxide released inside each country's borders, mostly by burning coal, oil and gas, and by making cement. They are estimated from fuel and production statistics, not measured in the air, and are revised each year as those statistics are revised.

Divide a country's emissions by its people and the ranking turns over. India has as many people as China and four times as many as the United States, so its total is large while its share per person is small. A per-person average says nothing about how emissions split within India: a household with two air conditioners and a car emits many times what a household with one bulb does.

Carbon dioxide stays in the air for centuries, so warming follows the total ever emitted, not this year's flow. Countries that industrialised early carry a larger share of the stock than of today's emissions; India, a late emitter, the reverse. The warming share also counts methane and nitrous oxide and emissions from clearing land, estimated by Jones and colleagues.

Most of India's coal is burned in power stations, the rest in steel, cement and brick kilns. Oil goes mostly into transport and cooking gas; natural gas is a small fuel in India, unlike in the United States. Cement releases carbon dioxide twice: from the fuel that heats the kiln and from the limestone it bakes.

Methane and nitrous oxide warm the air far more per tonne than carbon dioxide; they are converted to "CO₂-equivalent" tonnes so the gases can be added up. India's methane comes mostly from flooded rice paddies, cattle digesting fodder, and landfills; its nitrous oxide mostly from fertiliser on fields. These are farm emissions, tied to food and livelihoods, and much harder to cut than a coal plant's.

Primary energy is everything a country uses before conversion: coal into a power station, crude into a refinery. Electricity from wind, sun and water is counted as the fossil fuel it would have taken to generate it (the "substitution method"), so its share is comparable with coal's. Traditional biomass, still the cooking fuel of many rural homes, is left out of these figures.

This is electricity generated, not capacity installed. A solar plant produces at full power only around midday on clear days, so its share of generation is much smaller than its share of capacity. For capacity month by month, see the renewable energy guide.

Each group is a 20-year window; within it, one hollow dot per emissions pathway. The pathways barely differ until the 2040s, because the warming of the next two decades is largely set by gases already emitted. After that the choice of pathway decides most of the outcome. A child born in 2026 is 14 to 33 years old in the 2040-2059 window and 54 to 73 in 2080-2099.

The baseline here is 1995-2014, the models' reference period, not 1951-1980. The models' change from 1995-2014 cannot be added to the observed change since 1951 to make one number: they are different instruments.

The grey line is what the models reproduce for 1995-2014; the dots are what they project. Compare dots with the grey line, not with the observed counts in section 02: a model's past is its own ruler. The hollow, dashed marks are a reminder that none of this has happened.

Land far from the sea warms more than land beside it, and high ground more than low. The ordering is more robust than the numbers: models agree better on where it warms most than on by how much.

Cooling degree days measure the heat a building has to get rid of to stay at a comfortable temperature, and so track the work fans, coolers and air conditioners must do. They are the standard input to estimates of electricity demand for cooling.

The figure is the change in deaths from heat and cold combined: warmer winters save some lives, hotter summers cost more. It is a projection from statistical links between temperature and deaths, observed across many countries, applied to India's projected climate and population. India's death registration does not record heat as a cause reliably enough to count these deaths directly.

A satellite sees a scorch mark, not its cause. In India, most burnt cropland is stubble set alight after the harvest, in Punjab and Haryana after rice in October and November, and across the plains before the summer crop. Its smoke adds to north India's winter haze. In the weekly satellite record, burning peaks in March and April and again in October and November.