A 30% Chance of Rain—Who Exactly Has a Three-in-Ten Chance of Getting Wet?

Geoffrey Chen

When Numbers Start Making Decisions · Season Five, “Things That Have Not Happened Yet” · Article 1

1. The same 30% can support opposite decisions

Imagine an outdoor event scheduled for Saturday morning. The forecast says there is a 30% chance of rain, with a possible total of zero to one millimetre. The organiser says that a dry outcome is more likely and keeps the plan unchanged. The technician wants a shelter because even a brief shower could damage the sound equipment. Both have read the same number and reached opposite decisions.

That does not necessarily mean either person misunderstands probability. Thirty per cent describes a conditional uncertainty. It does not contain the value of the event, the size of a possible loss, the cost of an alternative venue or the decision-maker’s tolerance for risk. A forecast compresses a possible future into a number, but it cannot eliminate the judgement between evidence and action.

The first question is what the 30% refers to. It is not a prediction that rain will fall for 30% of the day. It does not say that exactly 30% of the forecast area will become wet, or that three people in ten will encounter rain. In the Bureau of Meteorology’s precipitation products, the probability is tied to a location or grid, a forecast period and a measurable-rain threshold. The relevant event is at least 0.2 millimetres of precipitation. Bureau of Meteorology: ADFD precipitation forecast grids

The same statement can be expressed in reverse: there is a 70% chance of receiving less than 0.2 millimetres. Location, time, threshold and forecast version are all part of the proposition. Remove any of them and the percentage can silently become an answer to a different question.

2. Chance of rain and amount of rain are different questions

Whether measurable rain occurs and how much falls if it does occur cannot be answered by one percentage. A few drops may be visible on a window while the recorded amount remains below the measurement threshold. Conversely, a low overall probability can coexist with a high total in one of the less likely scenarios.

The Bureau therefore publishes rainfall ranges as well as probabilities. Its current guidance explains that the lower displayed amount has a 75% chance of being reached, while the higher amount has a 25% chance. An expanded display may also show a 50% value. These are exceedance probabilities for particular totals, not promises that the observed amount will remain between two bounds. Bureau of Meteorology: How to interpret the daily rainfall forecast

Different users need different parts of this information. A walker may care about encountering any rain. A reservoir operator cares about volume across a catchment. Drainage and emergency managers care about the high-impact tail. A forecast icon has very little space in which to communicate probability, amount, period and type of weather. It is useful as an alert, but it should not carry a high-consequence decision alone.

3. Where does 30% come from?

Modern forecasting is not a guess made by one observer looking at the sky. Measurements from stations, radar, satellites, oceans and the atmosphere are assimilated into numerical weather models. Different models and initial conditions produce a family of possible futures. Automated systems and forecaster judgement turn those results into products for particular places and periods.

No individual knows the complete state of the atmosphere, and no instrument directly measures Saturday’s rain in advance. The probability arises from relationships among observations, physical equations, model limitations, historical performance and product rules. A local shower may be smaller than the model grid. Initial measurements contain error. Possible paths generally spread as lead time increases. New observations can change the forecast.

A useful probability forecast should also be calibrated. Across a large set of comparable 30% forecasts, the event should occur roughly three times in ten. Calibration evaluates a class of forecasts; it does not guarantee one case. A dry Saturday does not prove the forecast wrong. If comparable 30% forecasts almost never produce measurable rain, that would reveal a systematic problem.

This distinction matters because people often evaluate a probabilistic forecast as if it were a categorical prediction. “It did not rain” settles what happened, but not whether 30% was a reasonable estimate given the information available at the time.

4. When does the number begin to decide?

The public forecast does not order anyone to cancel an event. Thirty per cent acquires decision-making power only when another actor attaches an action rule.

A person may take an umbrella above 40%. A film contract may connect specified weather forecasts to insurance. A construction site may combine rain, lightning, wind and task type in a stop-work procedure. An emergency service may rely on specialised products rather than the public daily icon. Because each rule enters a different chain of consequences, different thresholds can be reasonable.

There are therefore two layers of numerical authority. The Bureau defines and publishes a meteorological probability. The user decides what probability triggers what action. If an organiser makes 30% an automatic cancellation threshold, that threshold is not drawn by the Bureau. The organiser remains responsible for safety, cost and alternatives.

Saying “the forecast made us do it” hides the normative choice. Evidence does not decide whether the cost of an unnecessary cancellation is more serious than the cost of continuing and causing damage. A human or institution must make and explain that comparison.

5. Error is more than being right or wrong

A decision rule can produce four broad outcomes: it triggers and rain occurs; it triggers and rain does not occur; it does not trigger and rain occurs; or it does not trigger and the day remains dry. These include hits, false alarms, misses and correct negatives.

Reducing false alarms sounds desirable until the associated increase in misses is considered. For a wedding, a false alarm may mean the cost of a marquee. For flash flooding, thunderstorms or aviation, a miss can involve lives. Overall accuracy cannot determine the acceptable distribution of error.

Mismatch can also occur even when the meteorological number is sound. The user may check the nearest town while the activity is in a valley, on a coast or in a different grid. A daily probability may be applied to a two-hour event. The 0.2-millimetre threshold may not match the amount that damages equipment. The number can answer its own question accurately while being unsuitable for the user’s question.

High-frequency users should therefore record forecast version, location, period and outcome, then review whether their own action threshold is too sensitive or not sensitive enough. Product changes should also explain definitions and display changes so users can separate a changing forecast from a changing interface.

6. Probability is an interface for action

From the perspective developed in this series, 30% is not a fixed property attached to Saturday’s weather. It is formed by the atmospheric state, observing network, model, time window, grid and definition of measurable rain. The structure is stable enough to support action without turning the future into a known fact.

Society delegates part of its knowledge of the future to models and meteorological institutions. That delegation is necessary. It does not transfer the remaining commitment. A model can estimate a probability, but it cannot explain why an event organiser cancelled, accept an employer’s duty to workers or select a government’s acceptable risk of missing danger.

The more uncertain the future, the more valuable reversible action becomes. At 30%, moving equipment, arranging a removable shelter and setting a review time may be more proportionate than immediately cancelling. Measures can escalate as forecasts update. That is not indecision; it aligns the strength and reversibility of action with the strength of evidence.

Conclusion: let probability prompt action without pretending to command it

Rain probabilities should remain visible as percentages because they preserve uncertainty better than a bare phrase such as “rain possible”. But the display should also make the event definition, forecast period and possible amount easy to see. High-consequence users should rely on products suited to their exact place, period and hazard.

Thirty per cent can justify preparation. It should not automatically cause the same response in every setting. Whoever connects it to cancellation, work stoppage or evacuation must explain the expected errors, review time and exceptions. The greater the consequence, the less defensible it is to preserve only a screenshot of one percentage.

The value of probability is not that it announces the future in advance. It helps us arrange a present that remains corrigible while the future is unsettled. If the feared outcome never arrives, preparation was not necessarily wasted. The right question is whether the response was proportionate to what could reasonably be known and what was at stake.

Primary sources

Discover more from Geoffrey Chen

Subscribe to get the latest posts sent to your email.