How it's measured
How thousands of thermometers become one number.
Pick your depth. Same facts at every level — only the detail changes.
Scientists have been writing down the temperature all over the world for 175 years. Three different teams add the numbers up in their own ways — and they all find the same thing: Earth has been getting warmer, and fastest in the years you were alive.
Each number here is an anomaly — how much warmer or cooler a year was than the 1951–1980 average. Three independent teams (NASA, NOAA, Berkeley Earth) build their estimates from thousands of weather stations plus ocean readings, using different methods. Their answers almost always land within about 0.04°C of each other — a difference far too small to feel. The agreement itself is the strongest piece of evidence on this site.
Global mean surface temperature can't be measured with one thermometer — it's estimated by combining land station records (GHCN and others) with sea-surface temperature datasets, gridding the globe, and averaging with area weighting. Sparse regions (early oceans, the poles) are the hard part, and each team handles them differently — which is exactly why three independent answers agreeing within ~0.04°C (median, across 145 overlap years) matters. Anomalies are used instead of absolute temperatures because anomalies correlate over much larger distances, making sparse networks far more reliable.
Series shown: GISTEMP v4 L-OTI (GHCNv4 + ERSSTv5, 1200 km interpolation), NOAAGlobalTemp v6.0 (GHCN + ERSST, 90°S–90°N), and Berkeley Earth's land+ocean reconstruction (kriging over 50,000+ station series; air temperature above sea ice), the latter with published 95% uncertainty. Each series is re-baselined here by subtracting its own 1951–1980 mean — the only transformation applied; curve shapes are exactly as published. The raw files, with retrieval dates, are linked below; the fetch pipeline is open and fails rather than fabricates when a source is unavailable.
The pipeline
1 · Fetch
An open script downloads each team's published file directly from its source. If a source is down, the script fails — it never invents, extrapolates, or smooths a value.
2 · Line them up
Each team publishes its numbers against its own reference period. For comparison, every series shown has had its own 1951–1980 average subtracted, so all three share one zero line. The shape of each curve is exactly as published.
3 · Snapshot
The result is stored with provenance beside every number — source URL, retrieval date (2026-08-21), reference period, method notes — and exported unedited to the CSV you can download from the sidebar.
Where every number comes from
NASA GISTEMP v4
NASA Goddard Institute for Space Studies
Land–Ocean Temperature Index (L-OTI); thousands of meteorological stations plus sea-surface temperatures.
Coverage 1880–2025 · published against 1951–1980 · margin of error not included
NOAAGlobalTemp v6
NOAA National Centers for Environmental Information
Merged land (GHCN) and ocean (ERSST) surface temperature analysis, 90°S–90°N annual means.
Coverage 1850–2025 · published against per NOAAGlobalTemp v6 climatology · margin of error not included
Berkeley Earth
Berkeley Earth
Independent statistical analysis of 50,000+ station series plus ocean observations; publishes per-year 95% uncertainty.
Coverage 1850–2024 · published against 1951–1980 (air temperature above sea ice) · margin of error published yearly