About the data

The short version

Everything historical is measured or reconstructed from measurements — no climate models are involved in the past. The past comes from ERA5, a scientific weather reconstruction of the whole planet since 1950. Comparisons are against the classic 1961–1990 reference period. The future (to 2050) comes from three high-resolution CMIP6 climate models and is a scenario, not a prediction. We show what the data says — including when it contradicts expectations, and including its blind spots.

Data sources

All global datasets are bundled with the site (no third-party requests from your browser) and refreshed about yearly; local data is fetched live and cached (see “Freshness” below).

What a reanalysis is — and how far to trust it

ERA5, by the EU’s Copernicus programme, blends millions of real observations (weather stations, ships, balloons, satellites) with a physics model to reconstruct the weather everywhere on Earth, hour by hour, back to 1950. It is the standard dataset of climate science for exactly our purpose: consistent, comparable local climate history for any point on the planet, including places that never had a weather station.

Its limits, honestly:

· Values describe a grid cell of roughly 10–25 km — a neighborhood average, not your street’s thermometer. We never round or move your coordinates (elevation matters: a few km can change the answer in hilly terrain), but the value is still an area average.
· The decades before the satellite era (~1979), and especially the 1950s, rest on fewer observations — trends there carry more uncertainty.
· In some data-sparse regions (high mountains, parts of the tropics) precipitation is weakly constrained, and changes in what ERA5 assimilates can create artificial steps in local series. Where a chart shows an abrupt, persistent jump, treat it with caution.
· The most recent ~3 months are preliminary (“ERA5T”) and can be slightly revised.

Baselines, anomalies, and the stripe colors

1961–1990 is the reference everywhere on this site — the classic WMO climate-normal period, and roughly the climate today’s adults grew up in, which makes it a fair answer to “was it like this before?”. An anomaly is simply a year’s value minus the 1961–1990 average. Trend lines are 10-year rolling means: single years are weather, the decade line is climate. One global exception: the “1.5 °C / 2 °C” thresholds on The big picture use the pre-industrial baseline (1850–1900), which was about 0.4 °C cooler than 1961–1990 — the site says so wherever both appear.

Stripe colors are globally fixed: the same color means the same °C anomaly in every chart and every place, saturating at ±2.5 °C (the ColorBrewer palette used by showyourstripes.info). That is why the planet’s stripes look paler than your city’s — and why any two places on this site can be compared at a glance. Drought stripes use the BrBG palette on the SPEI scale (below).

The metrics, one by one

Warming (last 10 yrs): mean temperature anomaly of the ten most recent complete years. Warmest year: highest annual mean on record. “X of the 10 hottest years in the last decade”: of the ten hottest years since 1950, how many fell in the last ten years — with 76 years of data, pure chance would put ~1.3 there; values of 7–10 are the signature of a strong trend (WMO uses the same framing).

Hot days: days per year with a maximum of 30 °C or more. Tropical nights: nights that never fall below 20 °C — a health-relevant threshold, and in temperate places a phenomenon that simply did not exist in the baseline. Longest dry spell: the year’s longest run of days below 1 mm (the standard “consecutive dry days” index).

Heavy rain: the year’s wettest single day (Rx1day), days of 20 mm or more, rain per rainy day, and the share of the year’s rain falling in its five wettest days. Warmer air holds ~7 % more moisture per degree, so many places see rain concentrate into bursts even as totals fall. Honest limit: an actual cloudburst is a sub-hourly event a few kilometres wide, which daily ~25 km data smooths over — these metrics show the trend that makes such events more likely, not individual disasters.

Drought (SPEI-12): for each month we take the previous 12 months of precipitation minus reference evapotranspiration (FAO-56 Penman-Monteith — the atmosphere’s “water demand”), then standardize that balance against the same calendar month in 1961–1990. The result is in standard deviations: below −1 is drought, below −2 extreme drought, positive values are wetter than normal. Because warmer air is thirstier, SPEI can deepen even where rainfall is unchanged — that is the point of using it. Values are comparable with published drought research.

The last 12 months: a fresh daily view compared against day-of-year normals — for each calendar day, the 1961–1990 distribution of that day ±5 days (temperature) or ±7 days (rain), so the seasonal cycle is built in and “unusually warm for a March day” means exactly that.

Scenarios and the 2050 outlook — in full

Why scenarios, not forecasts. The physics of the climate is predictable; human emissions are a choice still being made. Climate science therefore works with SSPs (Shared Socioeconomic Pathways) — internally consistent “what if” storylines ranging from strong climate action (SSP1-2.6) through roughly current policies (SSP2-4.5) to fossil-fueled growth (SSP5-8.5). A scenario says: if the world emits this much, physics does that. The spread between them — about 1.8 °C vs 4.4 °C by 2100 — is the part humanity still controls.

Two experiments, two jobs. Both charts on this site come from CMIP6, the current generation of the global climate-model intercomparison — but from different sub-experiments:

Our local outlook uses three HighResMIP models (CMCC-CM2-VHR4, MRI-AGCM3-2-S, EC-Earth3P-HR), bias-corrected by Open-Meteo against the same ERA5 data as our historical charts. We show each year as the three-model mean with a band spanning the models — the band is model disagreement, not a confidence interval.

Why the high-emission pathway — and is that alarmist? HighResMIP ran only this one pathway (city-scale runs are computationally enormous). As a story about 2100, SSP5-8.5 is now widely considered unlikely — it assumes a return to massive coal growth. But that criticism concerns the second half of the century, which we deliberately do not show: up to 2050 the pathways differ by only ~0.2 °C, less than the model spread, because near-term warming is largely already committed. So the 2050 outlook is a fair approximation of any future short of rapid decarbonization — not a worst case. For the era where the pathway choice dominates (after 2050), see the full scenario fan on Global.

What the models cannot do: ~25 km cells smooth over neighborhoods; local extremes (heatwaves, cloudbursts) are systematically underestimated — your city’s worst future days will likely be worse than the smooth model average suggests.

Freshness, caching, and reproducibility

Local history runs to about one week ago and is recomputed at most every 30 days (recent months only — history is immutable, and everything is recomputed from scratch at least yearly). Projections are static science and cached long-term. The place cards on the homepage use small monthly-refreshed summaries. Global datasets are bundled with the site and refreshed about yearly. Every metric on this site is computed from the public datasets above with the documented definitions — anyone can reproduce our numbers.

Honest by design

We show what the data says: places that got wetter show wetter; where a story describes floods but the data shows drying, both can be true — and we say so rather than smoothing it over. Weather is not climate: single years prove nothing, decades do. Projections are labeled as scenarios, never predictions. And where the data has blind spots — sub-daily extremes, data-sparse regions, preliminary months — the relevant chart says so in plain language. Found something that looks wrong? Tell us: openforests.com.