Methods and provenance

This tool places an observed or forecast heat extreme in the context of a warming climate. It is contextualization, not attribution: it does not make a formal causal statement about this specific event, but shows how rare an event of this magnitude is now and how rare it becomes at higher global warming levels.

Method in brief

• Present-day rarity is observation-anchored. A non-stationary GEV is fit to observed annual block maxima of the event metric, with the observed global-mean temperature anomaly (relative to 1850-1900) as a covariate, so the present-day distribution accounts for warming to date. The current-climate return period comes from observations, not from model tails. The present-day GMST level is the end point of a linear trend over the most recent 30 complete years, which estimates warming now rather than a flat decadal mean (centred several years in the past), following the Indicators of Global Climate Change.
• Future panels come from CMIP6 change factors. The model ensemble is used only for the response: the shift in GEV location and scale per degree of additional global warming. Those per-degree change factors are applied to the observation-anchored GEV as deltas from the present.
• A global-warming-level axis, not scenario-by-year: +1.5, +2, +3 °C relative to 1850-1900.

Two representations: France-wide average vs local peak

The headline figure and the first two maps use France-wide / coarse-gridbox averages of daily maximum temperature on the 1.5° 6-hourly reference footing. This is honest about scale: it reads several °C below the station values in the news, because it averages cooler regions, coasts, and high ground.

Alongside it the tool shows a local peak (local_txx): the per-cell annual maximum daily-max on the 0.25° hourly ERA5T product, over metropolitan France. This resolves the afternoon peak and inland cities, so it reads much closer to station-scale heat. The France-peak headline is the single hottest metropolitan-France 0.25° cell for the event.

• Source / footing. local_txx is fit directly on ERA5T 0.25° hourly, so the event value sits on that same native footing — no reference-product offset (the 1.5° pipeline needs one; this one does not). The present climatology uses the 1991–2020 period (30 annual maxima).
• Future shift — a documented approximation. CMIP6 change factors are fit at the models' native (coarse, ~1–2.5°) resolution and then interpolated to the 0.25° grid before being applied as deltas to the fine present fit. Applying a coarse change factor to a fine present distribution is an approximation: it carries the large-scale warming response, not fine-scale changes in the response itself.
• ERA5 vs stations. Even the 0.25° local peak is a gridbox mean, so it runs roughly 1–2 °C below the hottest individual station (a known ERA5 cool bias). The displayed local-peak value is honest about being a gridbox figure, not a station record.
• Metropolitan France only. Cells are masked to the Natural Earth France polygon, excluding Spain, Italy, and sea inside the bounding box.

Data sources

CMIP6 models used

Sensitivity cross-check

The observed GMST sensitivity is kept as a check against the CMIP per-degree scaling; a large disagreement is flagged rather than averaged over.

Documented offsets

Statistical convention

Licensing and citation

ERA5 is produced by the Copernicus Climate Change Service (C3S); use is governed by the Copernicus licence and ERA5 should be cited per the C3S terms. CMIP6 data are made available under the terms of the participating modelling groups; the ssp585 and historical experiments and the listed source models should be cited per CMIP6 / WCRP terms. ECMWF open data is published under CC-BY-4.0. The NOAAGlobalTemp series stands in for HadCRUT5 in this build. The present-day warming level follows the Indicators of Global Climate Change (Forster et al., 2024, Earth Syst. Sci. Data 16, 2625–2658, doi:10.5194/essd-16-2625-2024).