There are roughly 6,000 weather stations feeding the global temperature record, and most of them sit near cities. That single fact has driven more arguments between climate scientists than any other piece of the machinery—because the thing everyone quotes, "the planet is 1.3 °C warmer," is not a measurement at all. It's an inference. A consensus built from scattered thermometers, satellite radiometers, ocean buoys and weather balloons, all stitched together by humans who argue about the stitching.
Here's what surprised me when I dug into how the numbers actually get made: no single instrument on Earth measures the global average temperature. Not one. Every published figure you've ever seen—NOAA, NASA, the Met Office—is a reconstruction. And once you understand the reconstruction, the "is it accurate?" question starts sounding very different.
Key takeaways
- No single thermometer reads global temperature; every published figure is an aggregation of thousands of local readings.
- Stations measure land air temperature two metres above the ground; oceans are measured by ships, buoys and satellites.
- Raw readings are corrected for station moves, instrument swaps, time-of-day changes and urban warming before they count.
- Results are expressed as anomalies against a reference period, not as absolute temperatures—because anomalies survive station changes far better than raw values.
- Independent groups using different methods land within a few hundredths of a degree of each other.
- The record goes back to around 1850 for land and sea surface data, with satellites only from the late 1970s.
How scientists measure global temperature records, from thermometer to headline
The chain starts with a single thermometer in a white louvered box—a Stevenson screen—standing on grass, roughly 1.5 to 2 metres above the ground, away from buildings and pavement. That placement is not arbitrary. It's designed to sample the air that people and plants actually live in, while shielding the sensor from direct sun and ground radiation.
Those stations are not evenly spread. Europe, North America and East Asia are dense with them. Parts of Africa, the Amazon basin and the interior of Antarctica are effectively blank. Scientists call this the coverage problem, and it's the reason global averages are computed on a grid: the planet is divided into boxes, stations within each box are averaged, and boxes with no data are filled by interpolation from neighbouring regions.
The four measurement streams that feed the record
Land thermometers are only one channel. A complete global figure pulls from four:
- Surface stations — the ~6,000 network, plus thousands more historical records
- Sea surface temperature — ship engine-intake readings, drifting buoys, moored buoys; historically the messiest data of all, because ships used to haul buckets of water on deck
- Satellites — microwave radiometers measuring lower-troposphere temperature since 1979
- Radiosondes — instrument packages on weather balloons, launched twice daily from hundreds of sites
Which stream dominates the final number? Land. Because land warms faster than ocean and is where humans live, surface station data drives most of the headline anomaly figure you read in news articles. The ocean, which covers 71% of the planet, moderates everything—and it's the hardest part to measure consistently.
Why the answer is an anomaly, not a temperature
Ask what Earth's average temperature is and you'll get a number around 15 °C. Ask a climate scientist and they'll give you an anomaly: "0.85 °C above the 1951–1980 mean," or "1.3 °C above the pre-industrial baseline."
The reason is subtle and, honestly, the most elegant part of the whole method. Absolute temperature is fragile. Move a station 200 metres up a hill and its absolute reading shifts. Switch from a mercury thermometer to a digital sensor and the absolute value drifts slightly. But if you compute the difference between that station and its own long-term average, the shift largely cancels out—because you're comparing like with like. Anomalies travel; absolute values don't.
How scientists correct the raw data before it counts
Raw station readings are not used as-is. They go through a process called homogenisation, and this is where most public confusion lives.
Picture a station that moved from a town centre to a nearby airfield in 1975. The old site was warmer—concrete, traffic, buildings radiating heat. The new site is cooler. A naive reading would show a sudden temperature drop in 1975 that has nothing to do with climate. Homogenisation algorithms detect that break point statistically, compare the station against its neighbours, and adjust the series so the break disappears.
The urban heat island question, honestly
The most common objection I hear is that stations sit in cities, cities are hot, so the record is inflated. It sounds reasonable. The evidence points the other way: when researchers compare urban and rural stations separately, the long-term trends are nearly identical. The urban warming effect exists locally, but it's small compared to the century-scale signal, and the correction steps catch most of it.
What genuinely did matter, historically, were three things: ships switching from wooden buckets to engine-intake measurements, a shift in observation times from morning to afternoon at many stations, and the transition from liquid-in-glass thermometers to electronic sensors. Each of these produced a real bias in the raw data. Each is corrected in the published record. I don't think that's a conspiracy—I think it's bookkeeping.
How the major datasets compare
NASA's GISTEMP, NOAA's global temperature product, the Met Office HadCRUT series, Berkeley Earth and Copernicus all publish annual figures. They use different coverage, different ocean corrections and different interpolation techniques. They do not agree perfectly.
| Dataset | Coverage approach | Ocean treatment | Typical divergence |
|---|---|---|---|
| NASA GISTEMP | Gridded, extrapolates into gaps | HadISST sea surface | Within a few hundredths of a degree |
| NOAA | Gridded with spatial smoothing | ERSST | Within a few hundredths |
| HadCRUT | Leaves gaps unfilled | HadSST | Slightly cooler in some years |
| Berkeley Earth | Full global reconstruction | Blended | Close to GISTEMP |
| Copernicus (ERA5) | Reanalysis, model-based | Modelled | Can differ more at monthly scale |
Zoom in on any single month and the spread gets wider—monthly values can differ by a tenth of a degree or more because of how each product handles ocean gaps. Zoom out to a decade and the lines converge so tightly that the disagreement is smaller than the year-to-year noise from El Niño. That convergence across five independent teams, using five different methods, is the real evidence the record holds.
What the record shows, and the four questions people keep asking
How much hotter is the Earth than 100 years ago?
Roughly 1.1 to 1.3 °C warmer compared with the late nineteenth century, depending on which baseline and which dataset you use. Against the 1951–1980 reference period that NASA uses, recent years sit well above one degree. In plain terms: Earth's surface is over a degree Celsius warmer than it was when your great-grandparents were young—and most of that increase arrived after 1970.
Do 97% of scientists believe in global warming?
The figure comes from surveys of published climate research, and it's roughly right as a measure of consensus—the overwhelming majority of climate scientists who publish on the topic attribute recent warming to human activity. I'd caveat the number itself: "97%" is a headline compression, and the underlying studies measured slightly different things. But directionally, the scientific consensus on anthropogenic warming is about as strong as consensus ever gets in any field.
What places will be unlivable by 2050?
This is where the temperature record stops being abstract. Projections suggest parts of the Persian Gulf and South Asia could see wet-bulb temperatures—heat combined with humidity—approaching the threshold where the human body can no longer cool itself through sweating. That's a physics limit, not a policy question. In practice "unlivable" usually means seasonally dangerous and economically unviable rather than permanently abandoned, and the specific places depend heavily on emissions between now and then.
Is Mars hotter or cooler than Earth?
Cooler. Much cooler. Mars averages around minus 60 °C, with a thin CO₂ atmosphere that provides almost no heat retention. The comparison is worth keeping in mind because it shows what actually drives surface temperature: not distance from the Sun so much as what the atmosphere does with the heat that arrives.
When the records begin, and why that matters
Systematic instrumental records start around 1850, when a coordinated network of land stations and ship logs became organised enough to be usable. Before that, you're relying on proxies—tree rings, ice cores, corals—which are far less precise year to year but show the broad shape of the past.
Global satellite coverage only begins in the late 1970s. That's a remarkably short window for a planet whose climate cycles run on decades. The trend lines across those four decades are consistent, the decade-by-decade ranking is stable, and every independent reconstruction arrives at the same shape. Which brings me back to the thing I keep thinking about: the record isn't a thermometer reading. It's an argument that has been audited, corrected, cross-examined by rival teams—and has held. That's a stronger claim than a single number could ever be.
