How far ahead can you actually trust a weather forecast?

Open almost any weather app and it will show you a neat row of days stretching two weeks into the future, each with a tidy icon and a temperature. It looks equally confident on day 1 and day 14. It should not.

Forecast quality drops off sharply the further ahead you look, and knowing roughly where it drops is the difference between planning well and being caught out. Here is the honest version.

Forecast skill is a measurable thing

Meteorologists do not just guess at how good forecasts are; they score them. The usual measure is called forecast skill, and centres like ECMWF publish it openly. In simple terms, skill compares the forecast against what actually happened, and against a naive baseline such as 'assume tomorrow is like the seasonal average'. A skilful forecast beats that baseline. Once a forecast stops beating it, the forecast is no longer telling you anything you could not have guessed from the calendar alone.

The rough timeline

Skill varies with the weather situation and the variable, but as a working guide for everyday planning:

• Days 1 to 3: reliable. This is the sweet spot. Temperature, and usually rain, are dependable enough to plan around with real confidence.
• Days 4 to 7: good for the trend. The broad picture, warmer or colder, wetter or drier, is often right. The exact timing of a shower or the precise high temperature is less certain.
• Days 8 to 14: low skill. You can sometimes see a general pattern, but specific daily details are close to guesswork. Treat any single day's numbers with heavy scepticism.
• Beyond 14 days: mostly climatology. At this range the 'forecast' is usually little better than telling you what a typical day in that season looks like. NOAA's own explainers put the practical limit of a useful specific forecast at around a week to ten days.

Ensembles: turning uncertainty into a number

How do forecasters know skill fades? They run the model many times with slightly different starting conditions, producing an ensemble of possible futures. Early in the forecast the members stay close together. Further out they fan apart. That fanning is a direct picture of growing uncertainty, and it is why a good service reports a range rather than a single false-precise number. When the ensemble members and different models still agree at day 7, that is a genuinely trustworthy day-7 forecast. When they scatter at day 4, even day 4 is shaky. This is the same why forecasts disagree model-spread idea, viewed through the lens of time.

Forecasts really have got better

The good news is that these ranges keep improving. A landmark review in the journal Nature, 'The quiet revolution of numerical weather prediction', documented that forecast skill has advanced by roughly one day per decade. In plain terms, a modern 6-day forecast is about as accurate as a 5-day forecast was ten years ago, and a 5-day forecast today rivals the 3-day forecast of the 1980s.

A today's 6-day forecast is about as good as a 5-day forecast was ten years ago. Steady, unglamorous progress of roughly a day of skill per decade.

That gain has come from better satellite data, faster supercomputers, smarter data assimilation and improved physics. It is real, but it does not repeal chaos. The atmosphere still has a hard predictability limit, and no amount of computing makes a specific day-20 forecast trustworthy.

Why the 14-day number is not gospel

So why does your app show 14 days at all? Because the model produces those numbers, and showing them is easy. The problem is presentation: a day-14 temperature displayed in the same confident style as tomorrow's implies a certainty that simply is not there. The number is a single line plucked from an ensemble that, by day 14, is spread all over the place. It is a plausible scenario, not a promise.

Practical guidance

• Plan firmly inside 3 days. For anything that must go ahead, an outdoor event, a long drive, rely on the near-term forecast and you will rarely be surprised.
• Use days 4 to 7 for direction, not detail. Great for deciding whether next weekend is looking wet or fine. Do not bank on the exact hour it rains.
• Treat 8 to 14 days as a heads-up. Note the general signal, then recheck as the day approaches. Expect it to change.
• Look for convergence over time. If successive updates keep pointing the same way, confidence is building. If they lurch around, the situation is inherently uncertain.
• Prefer a forecast that shows its uncertainty. A range or a confidence flag is more honest, and more useful, than a single crisp number that pretends day 12 is as sure as tomorrow.

The two-week forecast is not useless, but it is a sketch, not a photograph. The skill is knowing which days to plan around and which to simply keep an eye on. See how confidence changes day by day for your area on compare US forecasts or compare UK forecasts.