How we know the climate is changing

Start with the part that most arguments skip. The physics came first.

Nobody noticed the planet warming and then went hunting for a culprit. The mechanism was worked out in laboratories in the 1850s, the effect of adding carbon dioxide to the atmosphere was calculated in 1896, and the measurements that confirmed it were made decades later by people testing a prediction that already existed on paper.

That order matters. A theory that predicts something before it's observed is doing a very different job from one assembled afterwards to explain a result.

The nineteenth-century laboratory work

Joseph Fourier noticed the problem in the 1820s. Work out how much energy the Earth receives from the Sun, work out how much a body at that distance should radiate away, and the planet comes out far colder than it actually is — below freezing, everywhere, permanently. Something in the atmosphere was holding heat in.

Eunice Foote ran an experiment in 1856 with glass cylinders of different gases left in sunlight, and reported that the one filled with carbon dioxide got hottest and stayed hot longest. Her paper was read out at a scientific meeting in the United States and then largely forgotten for a century and a half.

John Tyndall did the careful version in 1859 in London, and his apparatus is the thing that settles the question. He pushed infrared radiation down a tube filled with one gas at a time and measured how much came out the far end. Nitrogen and oxygen — 99% of the atmosphere — were effectively transparent. Water vapour, carbon dioxide, methane and ozone absorbed strongly. He'd found that a tiny minority of the air does essentially all the heat-trapping.

Svante Arrhenius took the next step in 1896 and did the arithmetic by hand: what happens to surface temperature if you double the carbon dioxide. He got somewhere around 5 to 6°C, which is high by modern estimates but the right sign and roughly the right size, from a man with a pencil. Guy Callendar came back to it in 1938 with compiled temperature records and argued the warming had already begun. He was largely ignored.

None of this was controversial physics. It was gas absorption spectra, and it's the same physics that makes infrared cameras and CO2 lasers work.

Why carbon dioxide and not oxygen

The reason isn't arbitrary, and the precision matters here, because "trace gas, only 0.04%, can't possibly matter" is the objection you'll meet most often.

A molecule absorbs infrared radiation when the vibration it triggers changes the molecule's electrical asymmetry — its dipole moment. Nitrogen and oxygen are two identical atoms in a straight line. However they stretch, they stay symmetric, so there's nothing for infrared light to grip. They're transparent to it. Carbon dioxide has three atoms, and its bending and asymmetric stretching modes do change the charge distribution, so it absorbs hard in exactly the band the Earth's surface radiates in. Methane and water vapour do the same for their own reasons.

Concentration isn't the whole story, absorption is. A drop of ink in a bathtub is a trace substance too.

The heat-trapping itself works by altitude, not by a blanket. Sunlight arrives as short wavelengths and passes through mostly unimpeded. The warmed surface radiates back out in the infrared, where the atmosphere is not transparent. That energy is absorbed and re-emitted repeatedly on the way up, and it only escapes to space from a height where the air is thin enough for the radiation to get out. Add more carbon dioxide and that escape altitude rises. Higher air is colder, colder bodies radiate less, so for a while the planet is emitting less than it receives — and the surface has to warm until the balance is restored. That's the mechanism, and it's why the top of the atmosphere is the place to look.

What the thermometers say

Systematic instrumental records start in the mid-nineteenth century, sparse at first, and reach reasonable global coverage by around 1900. Several teams compile them independently: NASA's Goddard Institute, NOAA, the UK Met Office with the University of East Anglia, the Japan Meteorological Agency, and Berkeley Earth. They share many of the same raw stations, but they make different choices about corrections, gap-filling and ocean data.

They produce nearly the same curve. Warming of roughly 1.1°C from the second half of the nineteenth century to the 2011–2020 decade, most of it since about 1975.

The Berkeley Earth project is worth knowing about specifically. It was set up by Richard Muller, a physicist who had publicly criticised the existing temperature records and their handling of station siting and urban heat, and it took funding from sources sceptical of the mainstream result. The team rebuilt the analysis from scratch with different statistical methods and more stations. It reproduced the existing curve, and Muller said so in print.

The corrections people worry about are real and they're documented. Weather stations get moved, instruments get replaced, observation times change, cities grow up around thermometers that used to be in fields. Homogenisation adjusts for those breaks, and you can check whether it's doing something dishonest by looking at data it can't touch. Sea surface temperatures, measured from ships and buoys with no cities anywhere near them, warm. Rural-only station subsets warm. Weather balloons and satellites warm. The trend doesn't depend on the adjustments.

The things that don't need a thermometer at all

If you distrusted every temperature record on Earth, the physical world keeps its own accounts.

• Sea level. Tide gauges going back over a century, and satellite altimetry since 1993 measuring the whole ocean surface directly. The satellite era shows roughly 3.3 mm a year, against a twentieth-century average nearer 1.4. Two causes, both thermal: seawater expands as it warms, and land ice is melting into it.
• Ocean heat content. More than 90% of the extra energy has gone into the ocean, not the air, which makes this the single cleanest indicator of the planet's energy balance. Since the 2000s it's been measured by the Argo array — around 3,900 autonomous floats that sink to 2,000 metres, drift, and surface to report a temperature and salinity profile. Ocean heat content has risen steadily and with far less year-to-year noise than surface air.
• Glaciers. The World Glacier Monitoring Service has mass balance records from mountain glaciers on every continent. The great majority are losing mass, and many have retreated past positions marked by dated photographs and moraines.
• Ice sheets. The GRACE satellites, launched in 2002 and replaced in 2018, measure ice loss by weighing it — two spacecraft in formation whose separation changes minutely as they pass over regions of different gravity. Greenland and Antarctica are both losing hundreds of gigatonnes a year.
• Arctic sea ice. Continuous satellite coverage since 1979. The September minimum extent has fallen by something like a tenth per decade, and the ice that's left is younger and thinner.
• Living things. Records of first leaf, first flowering, egg-laying and migration dates, some kept by families and monasteries for centuries, have shifted earlier. Species ranges have moved poleward and uphill.

These are independent instruments, run by different institutions, measuring different quantities. They agree.

How we know the extra carbon is fossil

This is the part that turns a correlation into an identification, and it's mostly chemistry rather than climate science.

Carbon dioxide has gone from about 280 parts per million before industrialisation to past 420 today. Charles Keeling started measuring it continuously at Mauna Loa in 1958, when it read about 315, and that record — the famous sawtooth, rising every year with a seasonal wobble as northern hemisphere plants leaf out and die back — has never been interrupted.

Four separate lines of evidence say where the extra came from.

Carbon-13. Carbon comes in isotopes, and photosynthesis is slightly fussy: plants preferentially take up the lighter carbon-12, so plant tissue is depleted in carbon-13 relative to the atmosphere. Fossil fuels are ancient plant material, and they carry that same depleted signature. As atmospheric CO2 has risen, the ratio of carbon-13 to carbon-12 in the air has fallen, exactly as it should if the new carbon is coming out of old biology. Volcanic carbon has a different isotopic signature and doesn't do this.

Carbon-14. This isotope is radioactive with a half-life of 5,730 years, and it's constantly created in the upper atmosphere by cosmic rays. Anything millions of years old has none left at all. So burning fossil carbon dilutes the atmosphere's carbon-14. Hans Suess measured precisely that decline in tree rings in 1955, before atmospheric nuclear testing swamped the signal, and it's named the Suess effect after him.

Oxygen. Combustion consumes oxygen in a fixed ratio to the carbon burned. Ralph Keeling — son of the Mauna Loa Keeling — developed measurements precise enough to track atmospheric oxygen from 1989, and it's been declining in almost exactly the proportion the combustion arithmetic predicts. Carbon dioxide released by volcanoes or vented from a warming ocean wouldn't consume any oxygen.

The ocean's pH. If the ocean were the source of the extra carbon, it would be losing carbon and getting less acidic. It's doing the opposite. Surface ocean pH has fallen by about 0.1 units since pre-industrial times, which is roughly a 30% increase in hydrogen ion concentration. The ocean is absorbing carbon, not emitting it — it's a sink, and it's taken up something like a quarter of everything we've released.

The volcano question can be closed with arithmetic. The US Geological Survey puts global volcanic CO2 emissions at a fraction of a gigatonne a year. Fossil fuel combustion and cement production run above 35. Volcanoes are around 1% of the human figure, and a large eruption cools the planet briefly anyway, by throwing sulphate aerosols into the stratosphere — as Pinatubo did in 1991, to a degree that had been forecast in advance.

Air from 800,000 years ago

Ice cores are the strangest and most direct archive available. Snow falling on Antarctica or Greenland doesn't melt. It compacts under later snow, and as it turns to ice it seals off bubbles of the air that was there at the time. Those bubbles are literal samples of ancient atmosphere. Nothing is being reconstructed or inferred — the gas goes into a mass spectrometer.

The EPICA Dome C core in Antarctica reaches back about 800,000 years, covering eight full glacial cycles. Across all of it, carbon dioxide oscillates between roughly 170 and 300 parts per million, and never once goes above that band. We're now well past 420, and the rate of increase is faster than anything in the record by a large margin.

The same ice gives you a temperature record alongside the gas. The ratio of heavy to light isotopes of oxygen and hydrogen in the ice itself depends on how cold it was when the snow formed, because heavier water molecules condense out of air masses preferentially on the journey from the ocean. Layers can be counted like tree rings near the top, and volcanic ash layers from known eruptions provide fixed dating points that let cores from different continents be lined up.

Now the honest complication, because it gets used as a gotcha and deserves a straight answer. In the ice core record, at the end of ice ages, temperature starts rising several hundred years before CO2 does. That's true and it's well understood. The trigger for a deglaciation isn't carbon dioxide — it's a shift in the Earth's orbit and axial tilt, the Milankovitch cycles, which redistribute sunlight by latitude and season on timescales of tens of thousands of years. That initial nudge warms the oceans, which release CO2, which amplifies the warming, which releases more. The gas is a feedback in that story, and the amplification is required to explain how such a small orbital change produces a full glacial cycle. A thing that can be an effect can also be a cause. And the orbital cycles are calculable: they aren't pointed towards rapid warming now, and they don't act over 150 years.

Fingerprints that rule out the Sun

Greenhouse warming and a brighter Sun would both raise surface temperatures. They leave different marks, and the marks have been checked.

The stratosphere is the decisive one. If the Sun were pumping out more energy, the whole atmosphere would warm together. Greenhouse gases do something different — they trap heat lower down and reduce what reaches the upper atmosphere, so the troposphere warms while the stratosphere cools. That's what balloons and satellites have measured for decades. Arrhenius-era physics predicted it; observation found it.

Nights have warmed faster than days, and winters faster than summers, which is what you'd expect from reduced heat loss rather than increased heat input. The Arctic has warmed several times faster than the global average, largely because melting bright ice exposes dark water that absorbs more sunlight.

Two measurements close the loop directly. Harries and colleagues published a comparison in 2001 of infrared spectra of the Earth taken from orbit in 1970 and again in 1997, and found less radiation escaping to space at precisely the wavelengths carbon dioxide and methane absorb. Feldman and colleagues did the mirror experiment at ground level, published in 2015, measuring the increase in downward infrared radiation at CO2's wavelengths at two American sites over a decade, and got a number consistent with the calculated forcing.

Meanwhile, solar output has been measured continuously from satellites since 1978. It cycles over about eleven years and has been flat to slightly declining across the period when surface temperatures rose fastest.

What the scientific community actually says

Plainly, without softening: the warming is happening, and human activity is the cause. The IPCC's Sixth Assessment Report in 2021 opens its physical science summary with the word unequivocal — human influence has warmed the atmosphere, ocean and land. Every national academy of science in every major country agrees, as do the professional bodies of physics, chemistry, geology and meteorology.

Surveys of the published literature and of publishing climate scientists have run repeatedly since 2004, using different methods and different samples. They keep landing between 97% and above 99% agreement among researchers actively working in the field.

What's genuinely uncertain is worth stating just as plainly, because pretending otherwise helps nobody. The exact climate sensitivity — how much warming a doubling of CO2 eventually produces — is still a range rather than a number, narrowed in the most recent assessment to a likely 2.5 to 4°C with a best estimate near 3. Cloud behaviour is the largest remaining contributor to that spread. Ice sheet dynamics, and therefore the speed of sea level rise this century, are poorly constrained at the fast end. Regional projections are far shakier than global ones.

Those are arguments about magnitude and timing. None of them is an argument about direction, and none of them touches the isotopes.

Checking it yourself

Nearly all of it is public. The Mauna Loa carbon dioxide series is published monthly by NOAA and Scripps. The major temperature datasets are downloadable, and Berkeley Earth releases its code. Tide gauge records from individual harbours go back a century or more. Sea ice extent is updated daily from satellite passes.

You can even see the seasonal breathing in the Keeling curve — CO2 dipping every northern summer as forests leaf out, rising again every winter as they shed and decay. There's more land, and more vegetation, north of the equator, so the whole planet's carbon dioxide follows the northern hemisphere's growing season.

That wobble sits on top of a line that hasn't stopped climbing since 1958.