Earth's Temperature Through Time

4.5 billion years of climate data — from Hadean magma oceans to the present interglacial — assembled from ice cores, ocean sediments, fossil pollen, geochemical proxies, and instrumental records. Each section explains how we know what temperatures were.

Global Mean Surface Temperature — Deep Time

4,500 Ma → present · anomaly relative to pre-industrial baseline (14 °C) · data from geochemical proxies (δ¹⁸O, δ¹³C, GEOCARB, stomatal indices)

Temperature values before the Cambrian (541 Ma) carry uncertainty of ±5–10 °C. The "faint young Sun" paradox — the Sun was ~30% dimmer 4 Ga ago — makes high Archean temperatures attributable to a dense CO₂ / methane greenhouse atmosphere. Snowball Earth episodes at ~720 Ma and ~635 Ma are the most extreme cold events in the record.

Key Warm & Cold Events

Major excursions in deep time

  • 4.4 Ga Hadean magma ocean. Surface solidifies; first oceans form from volcanic outgassing. +100 °C est.
  • 3.5 Ga Archean greenhouse — dense CO₂/CH₄ atmosphere compensates for dim Sun. Oceans liquid. +15 °C
  • 720 Ma Sturtian Snowball Earth. Glaciers reach the equator; global mean near −50 °C. −60 °C
  • 635 Ma Marinoan Snowball Earth. Second full glaciation, ended by volcanic CO₂ buildup. −55 °C
  • 500 Ma Ordovician greenhouse. No land plants; CO₂ ~4,000 ppm. Tropics above 40 °C. +12 °C
  • 444 Ma Late Ordovician glaciation — rapid CO₂ drawdown by weathering of Appalachians. −5 °C
  • 55 Ma PETM — Paleocene–Eocene Thermal Maximum. CO₂ spike, 5–8 °C warming in ~20,000 yr. +8 °C

Atmospheric CO₂ — Deep Time

Estimated from GEOCARB III, stomatal indices, boron isotopes

Cenozoic Climate — Benthic Foram δ¹⁸O Stack

65 Ma → present · Zachos et al. 2001/2008, Westerhold et al. 2020 · proxy: benthic foraminifera oxygen isotopes (δ¹⁸O)

How it works: The ratio of ¹⁸O to ¹⁶O in carbonate shells of bottom-dwelling foraminifera records both sea-surface temperature and global ice volume. Heavier ¹⁶O preferentially evaporates into glaciers, leaving ocean water enriched in ¹⁸O. Combined with Mg/Ca ratios (pure temperature signal), we can separate the two effects. Resolution: ~10,000 years. Uncertainty: ±1–2 °C.

Cenozoic Timeline of Key Events

  • 65 Ma K-Pg impact. Brief impact winter followed by rebound greenhouse from wildfire CO₂.
  • 55 Ma PETM — CO₂/CH₄ pulse (+5–8 °C), crocodiles in Arctic. Recovered over ~200 kyr. +7 °C
  • 50 Ma Early Eocene Climatic Optimum. Warmest sustained period of the Cenozoic. +13 °C
  • 34 Ma Eocene–Oligocene transition. Antarctic ice sheet forms as Drake Passage opens. −4 °C
  • 17 Ma Mid-Miocene Climatic Optimum. CO₂ ~500 ppm; Greenland briefly deglaciated. +5 °C
  • 14 Ma East Antarctic ice sheet expands permanently. CO₂ falls below ~350 ppm.
  • 2.6 Ma Pleistocene begins. Northern Hemisphere glaciation intensifies.

Ice Age Cycles — Antarctic & Greenland Ice Cores

800,000 yr → present · EPICA Dome C, Vostok, NEEM · proxy: δD and δ¹⁸O of ice, trapped air bubbles

How it works: Ice cores preserve annual layers of compressed snow going back 800,000 years. The ratio δD (deuterium/hydrogen) in the ice records local temperature when the snow fell. Trapped air bubbles preserve the exact atmospheric composition — CO₂, CH₄, N₂O — directly measured with mass spectrometers. The EPICA Dome C core covers 8 complete glacial–interglacial cycles. Orbital forcing (Milanković cycles) drives the ~100 kyr periodicity.

CO₂ vs Temperature — Ice Core

800 ka. Note: CO₂ and T track closely; CO₂ lags T by ~800 yr (orbital forcing → temperature → ocean outgassing → CO₂ amplification)

Sea Level During Ice Ages

Coral terraces, sediment records. Last glacial maximum (21 ka): sea level ~120 m lower than today.

Last 50,000 Years — Greenland Ice Core (NGRIP / GISP2)

50,000 yr BP → present · δ¹⁸O-derived temperature · Dansgaard-Oeschger events, Last Glacial Maximum, deglaciation

How it works: Greenland ice cores (NGRIP, GISP2, GRIP) provide near-annual resolution for this period. The δ¹⁸O signal in ice directly tracks local temperature over Greenland, with global teleconnections. The rapid Dansgaard-Oeschger (D-O) warming events — 25 of them between 115 ka and 11.7 ka — saw Greenland temperatures rise 8–16 °C within decades, linked to reorganisations of Atlantic Ocean circulation (AMOC). Antarctic ice cores (EPICA) show the same events inverted: when Greenland warmed quickly, Antarctica warmed slowly and vice versa — the bipolar seesaw driven by AMOC heat transport.

Dansgaard-Oeschger Events

25 abrupt warm interstadials between 115 ka and 11.7 ka — each a sudden jump of 8–16 °C over Greenland within decades, then a gradual cooling

  • DO-25 (~116 ka) First major interstadial of the last glacial. Greenland warms ~12 °C in <100 yr, linked to AMOC resumption. +12 °C
  • DO-14 (~55 ka) Strong Greenland warming event. Antarctic Isotope Maximum 4 — Antarctic cooling as AMOC carries heat north. +10 °C
  • H6–H1 (~60–15 ka) Six Heinrich Events: collapse of Laurentide ice sheet lobes, massive iceberg armadas flood N. Atlantic, shut down AMOC, trigger global cooling. −3 °C global
  • LGM (~21 ka) Last Glacial Maximum. Laurentide ice sheet 3–4 km thick over Canada. Sea level 120 m lower. Global mean ~6 °C colder than today. CO₂ = 185 ppm. −6 °C
  • DO-1 / Bølling-Allerød (~14.7 ka) Most recent D-O event. Greenland warms +12 °C in ~40 years. Deglaciation begins. Sea level starts rising rapidly. +12 °C
  • Younger Dryas (~12.9 ka) Meltwater pulse 1A floods North Atlantic → AMOC collapses → 1,200-year cold reversal. Greenland drops ~10 °C in decades. −10 °C GIS
  • Holocene onset (~11.7 ka) Transition from last ice age to interglacial. Greenland ice core shows +10 °C in a single decade — the sharpest natural climate transition in the record. +10 °C

Atmospheric CO₂ — Last 50 ka

From EPICA Dome C and Vostok ice cores. LGM minimum: 185 ppm. Pre-industrial: 280 ppm.

Sea Level — Last 50 ka

Coral terraces, sediment cores (Lambeck et al. 2014). LGM: −120 m. Current rate: +3.7 mm/yr.

Holocene Temperature — Multi-Proxy Reconstruction

12,000 yr BP → 1850 CE · Marcott et al. 2013, PAGES 2k, pollen, lake sediments, speleothems, tree rings

How it works: Multiple independent proxies are combined statistically. Pollen assemblages in lake sediments reveal past vegetation and thus temperature. Speleothems (cave stalagmites) record δ¹⁸O in annual growth bands. Tree rings (dendrochronology) provide annual resolution for the past ~11,000 years. Lake varves (annual sediment layers) count individual years. The Holocene Thermal Optimum (~8–4 ka) was 0.5–1 °C warmer than pre-industrial due to stronger northern-summer insolation (orbital precession). The Medieval Warm Period and Little Ice Age were regional (North Atlantic) not global phenomena, superimposed on a gentle long-term cooling trend.

Key Holocene Events

  • 12,900 BP Younger Dryas — sudden cold snap (−3 °C, ~1,000 yr), caused by AMOC shutdown from meltwater pulse. −3 °C
  • 11,700 BP Holocene onset. Temperature rises ~10 °C in a few decades in Greenland ice core record.
  • 8,000 BP Holocene Thermal Optimum — peak warmth. Sahara was green ("African Humid Period"). +0.8 °C
  • 8,200 BP 8.2 ka cold event — brief (~200 yr) cooling from final draining of glacial Lake Agassiz. −1.5 °C
  • 500–150 BP Little Ice Age — regional Northern Hemisphere cooling, glacier advances, Thames freezing. −0.5 °C
  • 1850–now Industrial warming. +1.3 °C above pre-industrial, rate 10–100× faster than any natural Holocene change. +1.3 °C

Instrumental Record — 1850 to 2025

HadCRUT5, GISTEMP, Berkeley Earth, ERA5 reanalysis · anomaly vs 1850–1900 baseline

How it works: Since 1850, temperature is measured directly by land weather stations (GHCN network: 25,000+ stations), ocean buoys and ship intake measurements, and since 1979 by satellite (MSU/AMSU microwave sounders). Multiple independent groups (NASA, NOAA, UK Met Office, Berkeley Earth, Copernicus/ECMWF) process raw data and reach essentially the same result despite different methodologies — strong evidence that the warming signal is real, not an artefact. The 2023–2024 anomaly exceeded +1.5 °C for the first time.

CO₂ Concentration — Mauna Loa

Direct measurement, Keeling Curve 1958–2025. Pre-industrial: 280 ppm. 2024: 426 ppm.

Rate of Warming by Decade

Each bar = mean °C/decade for that period. Acceleration clearly visible post-1970.

IPCC AR6 Projections — 2025 to 2100

SSP1-1.9 (best), SSP2-4.5 (middle), SSP3-7.0, SSP5-8.5 (worst) · anomaly vs 1850–1900

SSP1-1.9 — Net-zero by 2050, aggressive mitigation (~1.5 °C) SSP2-4.5 — Moderate action, policies close to current pledges (~2.5 °C) SSP3-7.0 — Fragmented policies, regional rivalry (~3.5 °C) SSP5-8.5 — Fossil-fuel intensive, no new policies (~4.4 °C)
Projections use Earth System Models (ESMs) from CMIP6, ensembled across 40+ models. Shaded bands show the 5–95% likely range. Unlike past reconstructions, projections carry genuine uncertainty from: (1) emissions trajectory, (2) climate sensitivity (ECS: 2.5–4 °C per CO₂ doubling, likely range), (3) carbon-cycle feedbacks (permafrost, Amazon dieback, ocean solubility), and (4) social tipping points. The current trajectory (as of 2024 policies) tracks SSP2-4.5 to SSP3-7.0.

Where Do We Stand in the Deep-Time Context?

Current trajectory on the 65 Ma scale. Even SSP2-4.5 would produce temperatures not seen since the Miocene Thermal Maximum.

Paleoclimate Proxy Methods

Seven independent lines of evidence — how scientists reconstruct past temperatures without a thermometer

🧊
Ice Cores
800,000 yr — present

Drilled from Antarctica (EPICA Dome C, Vostok) and Greenland (NEEM, GISP2), ice cores preserve annual snow layers. The isotope ratio δD (deuterium vs hydrogen) in the ice water records the temperature when the snow precipitated. Crucially, trapped air bubbles preserve the exact atmospheric gas composition, giving direct CO₂, CH₄, and N₂O measurements — no inference required.

Precision: ±0.5–1 °C · Resolution: annual (recent), decades (deep)

🐚
Foraminifera δ¹⁸O
65 million yr — present

Microscopic marine organisms (forams) build calcite shells whose oxygen isotope ratio (¹⁸O/¹⁶O) depends on seawater temperature and global ice volume. During ice ages, light ¹⁶O preferentially evaporates and gets locked in glaciers, enriching the ocean in ¹⁸O. Benthic (bottom-dwelling) foram stacks like Zachos et al. 2001 and Westerhold et al. 2020 cover the entire Cenozoic at ~10 kyr resolution.

Precision: ±1–2 °C · Resolution: ~10,000 yr

🌊
Mg/Ca Ratios (Forams)
65 million yr — present

The Mg/Ca ratio in foraminiferal calcite responds to temperature alone (unlike δ¹⁸O which conflates temperature and ice volume). Used alongside δ¹⁸O, scientists can separate the two signals: how much of the δ¹⁸O shift is from cooling vs ice growth. This allows independent sea-surface temperature reconstruction.

Precision: ±1–1.5 °C · Resolution: ~10,000 yr

🌳
Tree Rings (Dendrochronology)
11,000 yr — present

Trees produce one ring per year. Ring width and density record temperature (and moisture) during the growing season. The PAGES 2k consortium has compiled millennial-length tree-ring networks from 7 continental regions. Bristlecone pines in California and Finnish sub-arctic pines extend reliable records back 7,000+ years. Oldest living individual: 5,000 yr; fossil sequences extend further.

Precision: ±0.3–0.8 °C · Resolution: annual

🗿
Speleothems (Cave Deposits)
600,000 yr — present

Stalagmites and stalactites grow in annual-to-decadal layers. The δ¹⁸O of the calcite records the isotopic signature of rainwater (which varies with temperature and monsoon intensity). U-Th radiometric dating constrains ages to ±50–500 yr precision. Especially useful for tropical and Asian monsoon reconstruction where ice-core coverage is absent.

Precision: ±1–3 °C · Resolution: decadal

🌿
Pollen (Palynology)
50 million yr — present

Pollen grains are chemically resistant and preserve in lake and bog sediments for millions of years. Pollen assemblages (which plant species were present, in what proportions) are matched to modern calibration datasets to infer mean annual temperature and precipitation. The Modern Analog Technique finds the modern climate whose pollen assemblage most closely matches the fossil one.

Precision: ±1–3 °C · Resolution: centuries

⚗️
Boron Isotopes & GEOCARB
500 million yr — present

The boron isotope ratio (δ¹¹B) in marine carbonates tracks ocean pH, which in turn constrains atmospheric CO₂. GEOCARB is a long-range geochemical model that tracks the carbon cycle via weathering rates, volcanic outgassing, and organic carbon burial. Together they provide CO₂ estimates back 600 million years, allowing temperature inference via CO₂ radiative forcing and climate sensitivity.

Precision: ±3–8 °C · Resolution: millions of years

🌡️
Instrumental Record
1850 — present

Direct thermometer measurements from 25,000+ land stations and ocean buoys, cross-checked against satellite microwave sounders (1979–) and weather balloon radiosondes. Independent analyses (NASA GISS, NOAA, HadCRUT, Berkeley Earth, ERA5) use different station networks and homogenisation methods yet agree within ±0.05 °C — confirming that the ~1.3 °C rise since 1850 is robust.

Precision: ±0.05 °C (recent) · Resolution: monthly

Why multiple proxies matter: Each proxy has its own biases, spatial coverage gaps, and calibration uncertainties. When δ¹⁸O forams, Mg/Ca, tree rings, pollen, and speleothems all agree on the same temperature history — despite being chemically, biologically, and geographically independent — the convergence is strong evidence that the reconstruction is correct. This is the same logic as triangulating a position from multiple independent instruments.