Where every number comes from

Electricity generation mix

Country shares are normalised so that domestic generation sums to 100%. This drops net imports and statistical differences, which is why a country's low-carbon total in the stacked chart can sit one to two percentage points away from the published low-carbon share shown in the ranked bar chart. The UK is the clearest case: it imports 10–13% of its supply, mostly low-carbon French nuclear, which a generation-only view cannot see.

Global renewables and coal shares, 2000–2025

Still compiled — this one is not yet a direct pull. Stitched from successive Ember annual reviews rather than a single vintage. Ember restates earlier years as country data improves, so intermediate years here may move by up to about 0.3 percentage points against the current Ember series. The 2024 and 2025 endpoints are as published.

The build script already implements the Ember read, including the country generation mix; it needs a free API key to run (EMBER_API_KEY). Until that key is set these series stay flagged as compiled. Ember publishes under CC-BY 4.0, which is what makes the API route redistributable — the published-summary route is not.

Temperature anomaly series, 1850–2024

Direct pull. Read from the HadCRUT.5.0.2.0 annual global summary series by tools/fetch-data.mjs. One dataset end to end: the chart, the stat tiles and the overview hero all come from this single file, and the year rankings are computed from the series rather than written by hand.

Complete years only. The annual file publishes a row for the current year as soon as it has any months at all, with nothing marking it as partial — the 2025 row in this release is January and February averaged, and the only visible tell is a confidence band three times as wide as its neighbours’. The build reads the monthly file alongside the annual one, counts the months, and excludes any year with fewer than twelve from the record and from every ranking. An earlier version of this page did not, and consequently reported a two-month year as the warmest on record.

The one transformation applied. The Met Office publishes this series against a 1961–1990 baseline. Every figure on this site is against 1850–1900, so the build subtracts the mean of the 1850–1900 window of this same series — a shift of 0.356 °C. That keeps one dataset throughout instead of splicing a modern reanalysis onto a historical record. The uncertainty bands are the published 95% range, shifted by the same constant.

On 1.5 °C. Exactly one complete year in this dataset has exceeded 1.5 °C: 2024, at +1.54 °C. The Paris threshold is defined on a multi-decade average, not a calendar year, and the 2022–2024 mean is +1.38 °C. Higher figures quoted elsewhere for recent years generally come from Copernicus ERA5, which runs warmer than HadCRUT5 for the same year. Both are defensible; blending them is not. Earlier versions of this page quoted ERA5 headline figures above a HadCRUT5 chart, which produced a ranking the chart itself contradicted. The site now quotes HadCRUT5 throughout and carries the 1991–2020 figure alongside the 1850–1900 one so the two baselines can be reconciled without mixing datasets.

Plug-in vehicle sales by region, 2019–2024

2025 regional totals are as published. Earlier years are compiled so that each year's regional split sums to that year's published global total. Treat the split as indicative and the totals as firm.

Heat pump sales, 2019–2021

Compiled onto the current EHPA 21-country basis. EHPA's country coverage has changed over time, so a figure quoted from an older report will not match. 2022 onward are as published.

2025 insured losses by peril

Severe convective storm, wildfire and flood are published figures. The fourth row, tropical cyclone and other perils, is the residual to the $107bn total — derived by subtraction, not published, and flagged accordingly.

Fleet shares

Plug-in share of cars in use is compiled from registration stock estimates and is the least firm dataset on the site. It is included because the gap between fleet and sales share is the single most misread thing in EV coverage, and an indicative figure serves that point better than no figure at all.

• Insured losses under-count the poorest regions. Loss data measures insured value. Most economic damage in Asia, Africa and Latin America is uninsured and therefore invisible to the series on the extreme events page. The protection gap is not a rounding error — it is most of the loss.
• Definitions of "EV" vary by source. This site counts battery-electric plus plug-in hybrid and excludes conventional hybrids. Chinese NEV figures include extended-range EVs. Japanese "electrified" targets include conventional hybrids. Comparing a headline across sources without checking the definition will produce a wrong answer.
• Heat pump comparisons are not like-for-like. Nordic figures include cheap air-to-air units in homes that already heat with electricity. UK and German figures are dominated by air-to-water systems replacing wet gas central heating, which is a far larger job. The per-1,000-households chart normalises market size but not this difference.
• Target years are not commitments of equal weight. A statutory duty, a strategy document and a summit announcement are three different objects. The grid page separates them; most coverage does not.
• NOAA's billion-dollar disaster series is discontinued. It ends at 2024 and there is no successor with the same method. This site shows it stopping rather than splicing in a different series.
• Incentive tables decay fastest. Checked July 2026. Schemes change mid-year, exhaust budgets, and reopen with different rules. Verify before relying on any specific figure.

Figures compiled by this site are published under CC-BY 4.0 — use them, with attribution. Underlying sources keep their own licences, listed in the table above. Ember and NOAA data are openly licensed. IEA, Rho Motion, EHPA, Swiss Re and Munich Re figures are cited from published summaries under fair use for commentary and analysis; redistributing those datasets in bulk needs a licence from the publisher, not from this site.

If you cite something here, cite the primary source with the vintage shown, and cite this site only for the figures it derives itself.