Step one: measure the whole atmosphere
Every forecast begins with a snapshot of the sky as it is right now. That snapshot is assembled from tens of thousands of surface weather stations, twice-daily weather balloons launched from hundreds of sites, commercial aircraft reporting conditions along their routes, ocean buoys, radar sweeps, and a fleet of satellites watching clouds, moisture, and temperature from orbit. No single country could afford all of it; the data is shared worldwide through agreements coordinated by the World Meteorological Organization — one of the quieter triumphs of international cooperation.
Step two: let physics run forward
The atmosphere obeys physical laws — fluid dynamics and thermodynamics chief among them. A weather model divides the atmosphere into a three-dimensional grid of millions of cells and repeatedly applies those laws to advance the whole system a few minutes at a time, hours into days. Doing this fast enough to be useful takes some of the largest supercomputers on Earth. The best-known global models are Europe's ECMWF and America's GFS; regional models add finer detail for shorter ranges. Forecast sites and apps — including the weather page here, which uses the open Open-Meteo service — present a blend of these model outputs for your exact coordinates.
Why forecasts go fuzzy after a few days
In the 1960s, meteorologist Edward Lorenz discovered that tiny differences in the starting snapshot — differences far smaller than any instrument can measure — grow into completely different weather within about two weeks. This is the famous "butterfly effect," and it sets a hard ceiling on forecasting: no amount of computing power will ever deliver reliable day-by-day forecasts a month out. Modern centers embrace the chaos by running "ensembles": dozens of forecasts from slightly varied starting points. When the ensemble members agree, confidence is high; when they scatter, the forecasters know — and honest forecasts say — that the future is genuinely uncertain.
What "70% chance of rain" really means
A 70% probability of precipitation does not mean it will rain 70% of the day, or over 70% of the area, in most presentations. It means that in the model's judgment, given this setup, measurable rain reaches your location in about 7 out of 10 such situations. Practical reading: below ~20%, leave the umbrella; 30–60%, have a plan B; above 70%, assume rain and be pleasantly surprised otherwise.
A fair rule of thumb for trust
- Days 1–3: temperature and rain forecasts are strong — plan confidently.
- Days 4–7: the broad pattern (warm/cold, settled/stormy) is usually right; the exact timing and amounts wobble. Re-check as the day approaches.
- Beyond day 7: treat it as a trend, not a promise. Even the 7-day view on our forecast page is best read as "the shape of the week," firming up as days draw closer.
It's worth remembering how far this science has come: today's 5-day forecast is about as accurate as a 1-day forecast was in 1980. The forecast on your phone is quietly one of the most sophisticated products of modern science — refreshed several times a day, for free.