Pohoda: know your weather ahead← back to the map

Data and method

Last updated: 4 October 2026

Overview

Pohoda is built on the normal climate: typical weather at each place for each fortnight of the year. The optional seasonal forecast adjusts those normal values for the next three months, only where the forecast models have demonstrated skill.

From data to what you see Past climate and the seasonal forecast models feed a skill check, which decides where the forecast adjusts the climate, giving the map, popups and charts. Past climateERA5, 1996 to 2025 Seasonal forecast5 models, Copernicus C3S Skill checktested on 1993 to 2016 Normal climateper place, per fortnight Climate adjustedonly where the forecast is reliable map, popups, charts
The path from data to what you see on the map. The forecast only changes the normal climate where it passes a skill check.

1. The normal climate

The base layer is ERA5, the Copernicus and ECMWF reanalysis, for 1996 to 2025. Each place is described by its seasonal cycle plus a gentle long-term trend. Values are averaged into 26 fortnights per year at about 28 km resolution. Variables: temperature (highest, lowest, average), rain, wind, cloud, sunshine, wet, frost and snow days, and sea temperature.

The map colours, the match score and the best-time tables use these normal values.

2. The seasonal forecast

A seasonal forecast gives the lean of the coming months: warmer or cooler, wetter or drier than normal. It does not forecast individual days. Pohoda uses the Copernicus Climate Change Service multi-model forecast from five centres (ECMWF, Météo-France, DWD, ECCC, CMCC). It covers up to three months ahead and is refreshed monthly, around the 5th.

Thirds, not degrees

The forecast is expressed as odds for three equal groups. For one month at one place, the last 24 years are sorted into the coolest (or driest) third, the middle third and the warmest (or wettest) third. In a typical year each group has a one in three chance. The forecast states how those odds shift, for example a 55% chance of the warm third.

Three equally likely groups A bell curve split into below normal, normal and above normal thirds, and two bars: a typical year with one third each, and a forecast leaning above normal. cooler / drier third middle third warmer / wetter third How the past 24 years of one month at one place split into three equal groups A typical year: each third equally likelybelow 33%normal 34%above 33% An example forecast: leaning to the warmer / wetter thirdbelow 15%normal 30%above 55%
Odds for the three groups. The example forecast is illustrative.

Is the forecast any good here?

Seasonal forecasts are skilful in some places and seasons and not in others. Each model is tested against 1993 to 2016: for each place, do its forecasts beat guessing the average? This gives a skill score per place, tested so that chance results do not count. Where no skill is shown, the forecast is not used.

How much of the forecast is used, by skill A line that stays at zero until the skill reaches 0.02, rises to full weight at 0.10, and stays at full weight above. no skill moderate high all none half 0 0.02 0.10 0.15+ forecast skill at this place (how much better than "just use the average") share of forecast applied
Below a skill of 0.02 the forecast is ignored. It is phased in above that and used in full from 0.10.

Skill falls with lead time and is higher for temperature than for rain. Rain has skill mainly in the tropics, where patterns such as El Niño steer rainfall.

Where the forecast passes the skill test, by month ahead Share of the world where the skill test passes: temperature 92, 66 and 53 percent, rain 71, 34 and 32 percent, for the first, second and third month ahead. 0%25%50%75%100%92%71%1st month66%34%2nd month53%32%3rd month temperature rain share of the world (area weighted) where the skill test passes, September 2026 release
Share of the globe where the forecast is used, September 2026 release. This varies by release and season. Over Europe, rain skill is largely absent beyond the first month.

How the numbers change

Where the forecast has skill, the normal value is shifted toward the forecast, scaled by confidence.

Temperature and sea temperaturenormal + shift
21.0 °C plus 0.9 °C shows 21.9 °C.
Rainnormal × factor
40 mm times 1.12 shows about 45 mm. The factor is the odds-weighted average of how much wetter or drier each third has been at that place historically.

Clicking a place shows a note in the popup stating what the forecast changed there. Where there is no reliable signal, the note says so and values are unchanged. Adjusted fortnights are shaded on the slider and in the timeseries; darker means higher confidence.

What this cannot do

The raw forecast maps

The unadjusted forecast layers (odds and anomalies, with a confidence view) are at pohoda.app/?outlook=global&raw=1.

Sources

Contains modified Copernicus Climate Change Service information (2026). Neither the European Commission nor ECMWF is responsible for any use that may be made of the Copernicus information or data it contains. Pohoda is a planning aid and not a forecast service: check official forecasts before you travel.

Pohoda is a free, map-first planner for finding when and where the weather suits your trip. Open the map

legal & privacy · best time to visit