Data & AI
Forecasts & uncertainty
70 percent rain, and then it stays dry. Was the forecast wrong? Not necessarily. Here is how to read forecasts like a pro.
The weather app shows 70 percent rain, you carry an umbrella around all day, and not a single drop falls. Typical, they just cannot do it? Do not judge so quickly. Forecasts are not promises but probabilities, and if you read them correctly you make better decisions: for weather, elections, sports and everywhere else people look into the future.
What 70 percent rain really means
A 70 percent chance of rain means: on 10 days with exactly this weather situation, it rains on about 7 and stays dry on about 3. Your dry day was not a forecast error, it was one of the planned 3 out of 10. The forecast would only be wrong if, over time, such days brought rain much more or much less often than announced. A single observation can never refute a probability, only many observations together show whether the weather service is well calibrated. Well calibrated means: when it announces 70 percent, it does rain on about 70 percent of those days in the long run.
Why forecasts scatter
Where do the percentages come from? Weather services do not run their computer models just a single time but many times, each run with tiny changes to the starting conditions. The atmosphere is chaotic: small differences at the start can lead to completely different weather a few days later. If 35 out of 50 model runs show rain, the app reports 70 percent. The scatter of the runs is not a weakness, it is the most honest account of what the data can tell.
But correlation is not causation: the ice cream does not warm the weather, summer drives both.
The further away, the more uncertain
Tomorrow's forecast is fairly reliable, the day after a little less, and beyond about a week things get properly shaky. Small model errors amplify with every day. That is why good forecasts draw a funnel: a narrow band for the near future that widens further out. This goes far beyond weather, uncertainty in economic and climate forecasts also grows with the time horizon. Anyone promising you an exact single number for ten years from now is hiding the funnel.
Deciding well despite uncertainty
Uncertainty is not a reason to ignore forecasts, it is one more piece of information. With 70 percent rain the umbrella is worth it, even if it comes along for nothing on 3 out of 10 days: getting soaked costs more than carrying it. Ask two things about every forecast. First: what does it cost me if the less likely event happens? Second: how far into the future does the forecast reach? Then you are using the probability instead of being surprised by it.
Exercises
0 of 6 solvedTime to try it yourself. You can't break anything, every attempt counts.
The app reports a 70 percent chance of rain, but it stays dry. Was the forecast wrong?
On each of 10 days the chance of rain is 70 percent. On how many of these days should you expect rain on average?
On each of 20 days the chance of rain is 20 percent. How many rainy days should you expect among them on average?
A 70 percent chance of rain means: on 10 such days it rains on about ….
Which weather forecast is usually more uncertain?
Order the weather forecasts from the most certain to the least certain.
- 1The forecast for one week from now
- 2The forecast for the day after tomorrow
- 3The forecast for tomorrow
- 4The forecast for two weeks from now