Most panels fit their visible values and shaded bands. Orbit keeps a fixed vertical range. Check the units: similar shapes can represent very different amounts of change.

Hover labels briefly name the evidence or method. A proxy is an indirect clue to past conditions. Tags summarize each source; its mix of evidence can vary over time. “+ model” indicates an additional modelling step, not a direct measurement of the plotted quantity.

Dashed vertical markers identify source joins; lines and bands stop there. A jump can reflect different reconstruction methods. Lines connect available samples within a source without adding measurements.

Shading shows reported uncertainty. Percentages and ±1σ or ±2σ labels describe different ranges; σ means one standard deviation. The bands exclude uncertainty from aligning sources. Records without bands also have uncertainty.

BP means years before 1950. CE and BCE are calendar years. Year markers fall on January 1; close views show calendar dates. Sample dates, stage intervals and averaging periods appear on hover. Zooming cannot reveal changes finer than the source data.

Timeline flags give familiar landmarks for the scale. Hover or focus a label for its name and date; click or tap for an explanation and a Wikipedia link. Close with ×, Escape or a click elsewhere. Labels can extend beyond short bands; the bands show the duration. Older dates and broad periods are estimates, explained in each card.

Visual correlation helps you ask questions but does not establish causation. Some records share models or inputs: Osman's temperature reconstruction includes greenhouse-gas forcing in its model priors, and Miller's sea-level calculations use temperature-related inputs.

Values show the difference from the 1961–1990 average. From oldest to newest, the curve uses:

• About 485–66 million years BP: PhanDA, by geological stage.
• 66–2 million years BP: Hansen's global estimate.
• 2 million–24,000 BP: Snyder, a smoothed reconstruction.
• 24,000 BP to before 1 CE: Osman's 200-year means.
• 1–1879 CE: PAGES2k's annual reconstruction.
• 1880 CE onward: GISS monthly observations.

GISS combines thermometer readings on land and at sea. PAGES2k uses tree rings and other records, including corals, ice cores and sediments. Older estimates use chemistry preserved in seafloor sediments: Hansen and Snyder convert those signals to global temperature; Osman and PhanDA combine proxies with climate simulations.

GISS uses its 1961–1990 mean. Hansen uses its published 14 °C reference; PhanDA adopts the same value as an application assumption. Osman is aligned to PAGES2k over 150–1750 CE; Snyder uses Osman's aligned 0–5,000 BP mean. These constant shifts preserve variability while the methods and temperature definitions differ.

Shading shows PhanDA's 90% ensemble range, PAGES2k's 95% ensemble range, Snyder's 95% reconstruction interval, and Osman's ±1σ ensemble spread. Horizontal dating uncertainty is not shown. PhanDA's stages span millions of years; Snyder's dense grid is smoothed by dating uncertainty. These records cannot precisely rank the coldest period or establish rates and timing lags across joins.

CO₂ is measured in parts per million (ppm). Daily Scripps and NOAA air measurements follow the Antarctic composite, measured from air trapped in ice cores. Ice-core shading shows ±1σ concentration measurement uncertainty; gas-age uncertainty is excluded.

For about 66 million years to the ice-core record, CenCO₂PIP shows median 500,000-year averages with a 95% credible interval. The band describes uncertainty in each average, not all short-term changes.

Before 66 million years BP, Foster et al. reaches about 420 million years BP with a smoothed, most likely estimate and a 68% confidence range. Its 500,000-year grid is finer than the changes the reconstruction can resolve.

CenCO₂PIP's “mixed proxies” includes marine and land evidence. The older Foster segment draws on fossil plants and ancient soils. Both combine indirect methods whose mix changes through time.

Foster and CenCO₂PIP disagree strongly near their join. Their published ppm values are preserved; the separated ends do not measure a sudden atmospheric change. Compare broad trends, allowing for different methods and averaging periods.

Zeebe's ZB18a solution describes the shape of Earth's orbit over the last 58 million years, with samples 1,600 years apart. Older orbital phase is unconstrained and is omitted. From 2000 CE, La2004 extends the curve to the present using a line between its 1,000-year samples. The model change is marked; their values are not adjusted to match.

This is a calculation of planetary motion under gravity, rather than a climate measurement or sediment reconstruction.

Total solar irradiance is the power of sunlight per unit area, measured in W/m² at a fixed distance of 1 AU. The curve combines SATIRE / CMIP6 reconstructions, the NNL model from 1874, and TSIS satellite observations from 2018.

The early SATIRE segment uses tree-ring radiocarbon to infer solar activity; later reconstructions use sunspots and other activity records. These feed solar models. TSIS measures sunlight directly with a sensor on the International Space Station.

Historical records receive constant shifts based on shared dates, using the median of monthly differences to reduce the influence of exceptional months. TSIS keeps its measured level. Models can differ in trends and cycle amplitudes.

Views wider than 200 years show annual means. Closer views use daily dates with 50-day means from 1850 onward. Hover shows averaging periods, partial-year coverage and provisional readings. Averages stay within each source. Solar-cycle changes are typically about 0.1% of the roughly 1361 W/m² total, so automatic scaling makes small changes look large.

HolVol / eVolv2k / CMIP6 supplies annual global stratospheric aerosol optical depth from 9500 BCE to 2014 CE. Optical depth describes how strongly aerosols block light; it does not count eruptions. Component changes are marked.

HolVol and eVolv2k infer volcanic sulfur from deposits in Greenland and Antarctic ice cores, then model its aerosol effect. CMIP6 combines historical reconstructions, modelling and instrumental observations, including satellites.

From oldest to newest, the curve uses Marcilly (520–70 million years BP), Miller (about 67 million–798,000 BP), Spratt & Lisiecki (798,000–24,000 BP), Lambeck (about 24,000–3,000 BP), Kopp (about 3,000 BP–1800 CE), tide gauges from 1807, and satellites.

Modern values are relative to the 1950 mean; satellites are aligned using matching months with tide gauges. Older records use approximate modern references. Spratt keeps its published calibration at 5,000 BP and the Last Glacial Maximum.

Shading shows Kopp's ±1σ reconstruction spread, Lambeck's ±2σ accuracy, and Spratt's 95% bootstrap interval. Spratt uses seven reconstructions through 430,000 BP and five thereafter, with this change marked. Its 1,000-year grid does not resolve every fluctuation.

Lambeck and Spratt mainly describe sea-level changes associated with ice volume. Modern observations also include thermal expansion. Marcilly uses flooding maps and a modern land-height relationship, sampled every 10 million years in corrected Table 1. It has no supplied uncertainty band. Plate tectonics and ocean-basin changes also matter over these long periods.

Miller and Spratt infer sea level from marine sediment records and models. Lambeck uses coastal evidence, including corals, corrected for land movement; Kopp combines coastal proxy records and tide gauges. Modern gauges measure water level at coasts, and satellite radar measures sea-surface height from orbit.

Our World in Data provides historical estimates through 2023. Values from 2024 are projections in this snapshot, including years that have passed, and appear as dashed lines. Lines connect annual samples and reach the present using the next year's projection. Interpolated endpoints are identified in the hover text.

Historical estimates combine records and models, with archaeology informing early periods and censuses informing recent ones. Projections use demographic models of future population change.

Scroll or hold +/D or −/A to zoom; 0–9 select presets. C or Space toggles the cursor. Tab focuses dataset buttons; Enter or Space selects one. On a focused button, Left/Right selects an adjacent dataset and Home/End selects the first/last. Close this guide with × or Escape.

GitHub checks current GISS, NOAA, TSIS and Colorado observations on the 1st and 16th of each month. Reload to load the latest deployed snapshots. Historical reconstructions and population stay on their documented versions.

Install using your browser's Add to Home Screen or Install app command. Once saved, the charts work offline with cached data. Reconnect and reload for current observations. Source websites need an internet connection.