Did you know that a forecast saying ‘30% chance of rain’ can be reasonable even when it rains—and also when it stays dry? A probability forecast is not a confident instruction for just one day. It describes uncertainty about a defined event.
A 30% probability of rain means a forecast assigns probability 0.30 to a particular rain event at a specified location and during a specified period. Under the US National Weather Service definition of probability of precipitation, the event concerns measurable precipitation at a point during the forecast interval; check the NWS glossary for that provider-specific definition. Do not assume every weather app, country, threshold or time window uses exactly the same convention. Testing a probability needs clearly matched observations and many comparable forecasts, not one dramatic afternoon.
Diagnose the popular misunderstanding
Picture two fictional students. One says, ‘The app predicted 30% rain, and it rained, so the forecast was wrong.’ The other says, ‘It stayed dry, so the forecast was correct.’ Neither judgement follows from a single observation. A 30% forecast leaves 70% probability for the opposite outcome, but neither outcome is impossible.
Ask what the ‘30%’ refers to. Does it mean rain for 30% of the afternoon? Rain over 30% of the city? Thirty per cent of clouds will rain? These are common guesses. For the NWS point-and-period probability just defined, they are not the correct meanings. The event concerns the chance of qualifying precipitation at a specified place within the chosen interval.
The first skill is therefore not a calculation. It is reading the forecast’s event, place, threshold and time window before trying to verify it.
Read the complete forecast, not only the large number
Imagine a fictional card that says ‘Location: the school garden; Time: 12:00–18:00; Event: measurable rain at that point; Probability: 30%.’ That is precise enough for a class exercise. It does not say ‘there will be exactly 1.8 hours of rain’, because 30% of six hours is not the interpretation of the probability.
The NWS forecast terminology guide states that its probability of precipitation refers to measurable precipitation at a given point during a specified period. A forecast that simply says ‘rain 30%’ without its underlying terms should be treated cautiously until the provider’s definition is checked.
Also match forecast issuance times. A new 30% prediction issued on Friday is not the same forecast as a 30% prediction issued on Wednesday for Friday. For a fair study, decide which version will be tested and save its issue time before the weather happens. Do not choose whichever version looks best afterward.
A fully worked invented forecast-verification log
Here is a small thought experiment, not actual weather observations or a forecast evaluation of Singapore. A teacher constructs ten independent-looking fictional days, each with the same 30% probability for a comparable defined rain event. The observed outcomes for those ten cases are:
- Days 1–5: dry, rain, dry, dry, rain.
- Days 6–10: rain, dry, dry, rain, dry.
- Count: 4 rainy cases and 6 dry cases out of 10.
The observed fraction with rain is 4 ÷ 10 = 0.40, or 40%. The forecast probabilities sum to ten times 0.30 = 3 expected rainy cases on average across similarly defined opportunities. The realised count happened to be four. That difference is not proof of failure: ten cases are a very small sample, and chance outcomes fluctuate even when probability forecasts are well calibrated.
To check the counting, list the rainy days explicitly: 2, 5, 6 and 9. Count four, not five. The comparison is therefore 40% observed in a tiny illustrative group versus 30% assigned to each event. Neither figure is the percentage of hours during which rain fell.
What reliable long-run evaluation would ask
A calibrated forecast system is one for which events assigned probabilities near 30% occur on roughly 30% of comparable occasions over a sufficiently large, well-defined set. This is a long-run description, not a statement that exactly three of every ten consecutive days must be wet.
Suppose, as a second invented illustration, there were 100 comparable 30% forecasts and measurable rain occurred on 28 of those occasions. The observed frequency would be 28 ÷ 100 = 28%, two percentage points below 30%. That small difference alone is not a comprehensive reliability assessment. The sample size, selection, dependence, measurement quality and expected statistical variability all matter.
To do better, group forecasts by predicted probability, retain their original values, define the verifying observation consistently and collect enough examples. A school exercise can show the principle using fictional datasets without pretending to conduct professional meteorological verification.
A gentle extension: a probability error score
Older students can use the Brier score for binary outcomes. Let p be the predicted chance of the event as a number between 0 and 1, and let o be 1 when the event happens or 0 when it does not. For one case, the squared error is (p − o)². Average those squared errors over all cases. Lower is better for the evaluated set, but the score must be interpreted against suitable comparison forecasts and data.
For the ten fictitious days above, every p is 0.30. The four rainy days each contribute (0.30 − 1)² = 0.49. The six dry days each contribute (0.30 − 0)² = 0.09. The mean is (4 × 0.49 + 6 × 0.09) ÷ 10 = (1.96 + 0.54) ÷ 10 = 0.25. All arithmetic is shown so the pupil can check it.
That 0.25 is a descriptive score for this small invented set, not a professional verdict about a weather service. Students who are new to probability should master the event definition and long-run interpretation before learning the optional scoring extension.
Where school projects can go wrong
- Mismatch of place: a rain shower at home does not automatically verify the school’s point forecast.
- Mismatch of time: rain at 19:00 is outside a forecast defined for 12:00–18:00.
- Mismatch of threshold: a few drops or trace moisture may not meet the forecast provider’s measurable-precipitation definition.
- Changing the forecast version: checking a later update instead of the originally saved forecast biases the comparison.
- Missing observations: a rain gauge overflow, absent observer or faulty sensor must be flagged, not replaced with a guessed Yes or No.
- Treating chance as certainty: both outcomes remain possible when the probability is between 0% and 100%.
- Confusing calibration with guaranteed outcomes: long-run frequencies can support an evaluation, but small groups fluctuate.
Use the Core Aim guide to an overflowing rain gauge when discussing missing measurements. A rainfall observation that cannot establish whether the chosen threshold was crossed should be labelled uncertain rather than forced into the rainy or dry category.
Guided and independent practice
First step: rewrite the claim
Give the learner this fictional statement: ‘There is a 30% chance of measurable rain at the school during the afternoon.’ Ask for a one-sentence explanation without using the phrases ‘30% of the time’ or ‘30% of the city’. A strong answer says, ‘The provider assigns a 30-in-100 probability to the specified rain event at that place during that period.’ Add that this wording depends on the provider’s declared convention.
Second step: test the time window
The forecast period is 12:00–18:00. It rains only at 18:40. Does this count as a verifying event? No, not for that interval, even if the rain was real. Now imagine qualifying rain at 17:45: it is within the period and can count, assuming the location and threshold match.
Independent challenge
Create a fictional set of 20 matching 25% forecasts. Let qualifying rain occur six times. Compute the observed frequency: 6 ÷ 20 = 30%. The numerical difference from the forecast probability is five percentage points. Does that prove the service was miscalibrated? No. Twenty cases alone do not justify that conclusion. Ask what extra data, definition and checks would strengthen the evaluation.
Let advanced students explain why two consecutive 30% forecasts are not necessarily independent. A weather system can influence several days or neighbouring locations. That is one reason not to infer certainty from a tiny cluster of cases.
What pupils and adults can do with this skill
A learner can begin a forecast notebook with columns for provider, forecast issue time, predicted event, place, period, probability, verifying observation and uncertainty. Do not ask children to stand outside in hazardous weather or to collect personal location histories. Use supervised, publicly available records or invented examples.
A parent can help by asking ‘What exactly was predicted?’ instead of saying the weather app ‘got it wrong’ whenever an unlikely event happens. A teacher can use two or three fabricated forecast cards to diagnose misunderstandings, then a longer simulated set to practise frequencies. The observable improvement is a learner who can state the event, calculate a fraction correctly and resist a sweeping judgement from one case.
Weather decisions involve consequences as well as probabilities. Even a relatively low chance of dangerous conditions may warrant a precaution, depending on the severity and applicable official guidance. This article is about interpreting probabilistic information, not an instruction to ignore warnings or choose outdoor activity against school advice.
Useful links for the learning pathway
The weather forecast reliability specialist guide explores deeper verification and decisions. The Core Aim outlier lesson offers another way to think about surprising evidence. For a broader approach, see the Education Hub and the Clementi Mathematics teaching reference.
Frequently asked questions
Does 30% chance mean it rains on three of the next ten days?
No. Not necessarily. There is an expected long-run frequency under appropriate conditions, not a fixed quota for consecutive days.
If it rains despite a 10% forecast, was the forecast wrong?
An event assigned 10% probability is less likely but not impossible. To evaluate forecast quality, examine definitions and many comparable predictions, not only the surprising result.
Can I test a forecast against a photograph of wet pavement?
Sometimes the image can support an observation, but it does not automatically establish when rain fell, where the picture was taken or whether the provider’s measurable threshold was met. Preserve those limitations.
What this lesson is really teaching
Probability is a disciplined way of speaking honestly about uncertainty. Pupils who learn to read a forecast as a defined event, compare observations fairly and separate a single outcome from long-run reliability have acquired a skill they can carry into medical claims, risk estimates, surveys and everyday decisions—without pretending to know the future.