Instrumental record
Annual area mean for each box, degrees Celsius versus 1971–2000. Click an ocean to isolate it. Dashed lines are the Atlantic pair and the southern band.
Instrumental record · 1880–2025
Ocean temperature
A graph on each ocean: the full sea-surface record, and how fast that rise is now. Switch to heat content for the joules stored below the surface. What is plotted here was measured. The paleoclimate past is a bias model, taken up at the bottom of the page.
ERSSTv5 anomaly versus 1971–2000. Copper is the northern basins, bronze the tropical Pacific and Atlantic, teal the Indian and Southern oceans.
Swipe the map to reach each ocean.
Acceleration
A trend takes a long stretch of the record. This chart is only a 20-year slope, slid forward one year at a time, and those slopes have already reversed. From 1950 to 1980 the North Pacific ran -0.10 and the North Atlantic -0.15, while the Indian Ocean ran +0.16 and the southern band +0.14. The rate that has had time to settle is the whole instrumental series: +0.06 °C/decade for the global ocean, 1880–2025. The steep end of this chart is the current short window. It is trivial — a few tenths of a degree, inside swings the same oceans have already made and given back.
Rate since 2000
Linear trend, 2000–2025. The North Pacific surface is the fastest. The southern band is the slowest. Same window as the AMS State of the Climate basin rates.
Hemisphere oceans
A different series: NOAAGlobalTemp ocean-only, anomaly versus 1901–2000, not the boxes above. Northern ocean +1.04 °C from the 1880s to the last decade; southern +0.75 °C. Since 2000 the north is running +0.28 °C/decade and the south +0.15. From 1950 to 1980 the south was +0.12 while the north was -0.03.
Full record
Trailing 20-year rate
°C per decade. It takes many decades to establish a trend; a 20-year fit is shorter than one climate normal, and this one has run negative as well as positive. The settled rate is the full record, +0.06 °C/decade. The current end of the line is a short slope. It is trivial beside that long rate and beside the 3–4 °C the northern ocean already stands above the southern one.
Rate by era
Linear trend in each window, °C per decade, on the left axis. Global ocean runs +0.20 °C/decade since 2000, against +0.06 from 1950 to 1980. The blue bar is not temperature. It is absorbed shortwave sunlight, from Nikolov and Zeller’s reading of NASA CERES: 0.80 W/m² per decade, right axis. There is no blue bar before 2000 because CERES was not flying.
Nikolov and Zeller (Geomatics, 2024) use CERES EBAF 4.2. All-sky albedo fell about 0.79% after 2000, and the planet absorbed about 2.7 W/m² more shortwave. Absorbed sunlight is (1 − albedo) × TSI / 4. TSI itself barely trends over those years. It only wobbles by ±0.48 W/m². The step-up in the temperature rate is the albedo term, and it stays in tune with TSI because that is the beam being absorbed. In their regression the albedo drop plus those TSI variations account for 100% of the global surface warming trend since 2000 and 83% of the year-to-year variation (R² = 0.78 against surface temperature, 0.8 against the top 100 m of the ocean). Cloud albedo does most of the work. TSI modulates it. Their surface trend is +0.23 K/decade, about 0.288 K for each extra W/m² absorbed. The ocean series on this page is a different average and a different baseline, so the bars will not match that number exactly.
Where the hemispheres cancel
North minus south, same record. The difference swings from about −0.27 °C to +0.37 °C and comes back. Its trend from 1880 to 2025 is only +0.008 °C/decade, against a global ocean trend of +0.061. The two hemispheres do not cancel every year — year to year they usually move together — but the difference between them does not accumulate. What remains in the global mean is the part they share. The clear offset inside a single window is 1950–1980: the south ran +0.12 °C/decade while the north ran -0.03. The global rate in that window, +0.06, is what is left after those moves offset.
Only the instrumental record
A thermometer, a bucket, a buoy, a satellite radiometer, a flask of air. That is the record we actually have: sea surface here from 1880, heat content in the top 700 m from 1955, the top 2000 m from 2005, CERES sunlight from 2000, and direct CO2 from the Keeling flasks since 1958. Before those instruments, both temperature and CO2 are reconstructions. A reconstruction is a measurement plus a bias model. Change the bias and the past moves.
The other labs, temperature
The paleoclimate labs are not a longer version of this atlas. Tree rings, ice, and sediment need a calibration, and the calibration is a bias you choose. Mann, Bradley and Hughes centered their tree-ring principal components on the 20th century. McIntyre and McKitrick showed that this one choice loads hockey-stick shapes, especially strip-bark bristlecone, and that an ordinary centering lifts the early 15th century. The National Research Council in 2006 still called the last decades of the 20th century likely the warmest of the previous 400 years, and only “plausible” across the previous thousand. It also said strip-bark samples should be avoided as a temperature proxy. After 1960 many ring series fall while the thermometers rise. Some graphs simply stop the proxy there. Leave that segment in, or take the bristlecones out, and the recent blade changes.
The other labs, CO2
The flask record since 1958 is an instrument, in the same sense as the thermometers above. Ice-core CO2 before that is not. Air trapped in Antarctic ice is smoothed by diffusion in the firn, and the gas age sits on a densification model that can be centuries or thousands of years off the age of the ice. Shift that model and the CO2 curve slides against temperature. Low-accumulation cores cannot see a decade. Law Dome can see the industrial rise because the snow piles up fast. Stomatal-index series, Wagner and coauthors in 1999 and Kouwenberg and coauthors in 2005, put tens of ppm of century-scale CO2 change into the Holocene that the smoothed ice cores do not show. Pick the stomatal calibration or pick the diffusion smoothing, and the past CO2 level changes.
Biases inside the instruments
Even this page is not free of a bias choice. Ships read about 0.12 °C warmer than buoys. Karl and coauthors in 2015 put that offset into the sea-surface record and the recent “hiatus” shrank. Canvas buckets in the early 20th century were biased cold; the Folland and Parker corrections move those years by tenths of a degree. The shape of the instrumental rate depends on which of those adjustments you accept. That is a smaller freedom than a proxy reconstruction, because a sensor is still under the adjustment.
What that leaves
Both numbers people argue from — paleo temperature and paleo CO2 — can be raised or lowered by changing which bias is removed. The curves above are the part that does not require that step. On that instrumental window, the faster rate since 2000 lines up with more absorbed sunlight, and the hemispheric difference averages out instead of explaining the rise.