PSLE Science Reality Lab Vol No.041 | “The Forecast Says 30°C” — Is That One Certain Future?

PSLE-SCI-REALITY-0041

Wait, What? The forecast can say 30°C even though the future has not been measured yet.

Your weather app says: Tuesday, 3 pm — 30°C.

The number looks exactly like a thermometer reading. It has a unit. It sits beside a clock time. It may even appear on a smooth hourly graph.

But there is a crucial scientific difference. A thermometer reading describes an observation that has already happened. A forecast describes a future state that a model predicts from current observations, physical rules and calculations.

Modern forecasting systems often run not just one future, but many related model runs. These ensembles explore how forecasts change when plausible starting conditions or model details vary. The result is not twelve different futures waiting to happen. It is a scientific way to investigate uncertainty.

That makes a forecast a perfect Reality Lab object. The learner’s job is not to decide whether forecasts are “right” or “wrong” in general. It is to ask what the displayed number represents, how certain the forecast appears to be, what could change it, and how far the prediction can reasonably travel.

Quick Answer

A forecast value such as 30°C is a model-based prediction, not a future observation. Check the forecast time, lead time, update time and—when available—the spread of plausible model outcomes. A narrow spread can indicate that the model runs are relatively consistent with one another; a wide spread shows that the modeled future is less tightly constrained. Neither guarantees what will actually happen. Forecast uncertainty is not proof that forecasting is useless. It is part of the scientific evidence.

Reality Lab rule: A single forecast number is a summary of a prediction, not a measurement of the future.

The learner job this page owns

This Reality Lab owns one applied evidence-transfer job: how a Primary 5/6 learner should evaluate a real-world scientific forecast that is communicated as one line or one number. It does not replace weather science, computer modelling, probability theory or the existing eduKate pages on predictions, model limits and insufficient evidence. Instead, it applies those ideas to a familiar communication object: a forecast that looks more certain than the underlying evidence may be.

Original case: the school field-trip forecast

A fictional school plans a science field trip for Tuesday afternoon. The weather app displays 30°C at 3 pm.

Behind the simple display, imagine that twelve related model runs give these possible 3 pm temperatures:

Model run	Predicted temperature
1	27°C
2	28°C
3	28°C
4	29°C
5	29°C
6	30°C
7	30°C
8	30°C
9	31°C
10	31°C
11	32°C
12	34°C

A public app might choose one central summary value and display 30°C. That can be useful. But it hides the fact that the model system produced a range of plausible outcomes from 27°C to 34°C.

A student writes: “It will be exactly 30°C at 3 pm.”

That sentence is stronger than the evidence. A more scientific statement would be:

“The forecast summary is 30°C for 3 pm, while the model runs show a range of plausible temperatures. The actual temperature will be observed only when 3 pm arrives.”

Observation, model and forecast: three different jobs

Scientific object	What it does
Observation	Measures or records the world at a place and time.
Model	Uses rules and inputs to represent how a system may behave.
Forecast	Uses observations and models to predict a future state.

The same unit can appear in all three. A thermometer may record 30°C now. A model may calculate 30°C for tomorrow. The identical unit does not make the two statements the same kind of evidence.

Why run an ensemble?

Weather systems are sensitive to starting conditions. Scientists never know the complete state of the atmosphere perfectly at every location and altitude. Measurements have limits. Observations are unevenly distributed. Models simplify an extraordinarily complex world.

An ensemble explores this problem by running a set of plausible forecasts. Members may begin from slightly different estimates of the atmosphere or use other controlled variations. If the members stay close together, the model system is showing less spread for that quantity and time. If they separate widely, the modeled future is more uncertain.

NOAA training materials describe ensemble spread as useful information about forecast uncertainty and predictability. The crucial word is information. Spread does not magically reveal the one future that will occur. It tells you something about how strongly the model system constrains the future under its tested assumptions.

Twelve model members are not twelve independent observations

This is an important Reality Lab boundary. The twelve values in our table are not twelve thermometers observing Tuesday afternoon. They are related model outputs produced before Tuesday afternoon happens.

If all the models share an important missing process or biased input, many ensemble members can be wrong together. Agreement inside an ensemble can strengthen confidence about model consistency without becoming twelve independent pieces of real-world evidence.

This connects to Reality Lab Vol No.008: repetition does not automatically equal independence.

Spread is not the same as accuracy

Imagine every ensemble member predicts between 29°C and 31°C, but the observed temperature later turns out to be 35°C. The ensemble was tightly grouped but inaccurate for that event.

Now imagine another forecast whose members range from 27°C to 34°C and the actual temperature is 31°C. The ensemble had wider spread but included the observed outcome.

This is why forecast systems are checked against later observations. Scientists examine how often forecasts succeed, how errors behave and whether the stated uncertainty is well calibrated. A narrow range should not be treated as a guarantee.

Lead time changes the evidence

A forecast for one hour from now begins with more immediate information than a forecast for seven days from now. As predictions travel farther into the future, small uncertainties can grow and different model outcomes can separate.

That does not mean “longer forecast = useless”. It means the communication should preserve lead time. A learner should always ask:

When was this forecast produced?
What future time is being predicted?
How far ahead is that?
Has a newer forecast incorporated new observations?

The update-time trap

At 7 am, an app predicts 30°C at 3 pm. At noon, after new observations enter the forecasting system, it predicts 32°C.

A student says, “The first forecast was false because the app changed its mind.”

That is too simple. A forecast is conditional on the information available when it is made. Updating a prediction after receiving new evidence is part of good scientific practice. The important question is whether the forecast system improves decisions and whether its uncertainty is communicated honestly—not whether a future prediction remains frozen while new evidence arrives.

A seven-question forecast audit

What quantity is being forecast? Temperature, rainfall, wind, water level or something else?
For what place and time? Forecasts are bounded by location and timing.
When was the forecast issued? Older guidance may have been replaced by a newer run.
How far ahead is the forecast? Lead time affects uncertainty.
Is one number hiding a range? Look for ensemble spread, probability ranges or uncertainty bands when available.
Do the model runs share dependencies? Agreement among related runs is not the same as many independent observations.
What happened when reality returned? Later observations let us check the forecast.

Worked reasoning case 1: narrow spread

Ten model members predict between 29°C and 31°C. The public forecast says 30°C.

A careful conclusion is: the ensemble is tightly grouped around 30°C. A careless conclusion is: 30°C is guaranteed. Tight agreement can increase confidence relative to a widely spread ensemble, but the future still has not been observed.

Worked reasoning case 2: wide spread

Ten members range from 24°C to 35°C, yet the app still displays 30°C as a central summary.

The single number now hides much more disagreement. A decision sensitive to temperature—such as whether a field experiment may overheat—should consider the range rather than behave as if 30°C were certain.

Worked reasoning case 3: one outlying member

Nine members predict 29–31°C. One predicts 38°C.

Do not automatically delete the 38°C run because it looks inconvenient. Ask why it differs. Is it a plausible alternative produced by a changed starting state? A numerical failure? A different physical pathway? The outlier may be unimportant, or it may reveal a possible scenario worth watching. Evidence evaluation comes before deletion.

Worked reasoning case 4: all members share one weakness

Every member predicts 30–31°C, but the model poorly represents a local sea-breeze effect at the school site. The actual temperature later reaches only 26°C.

The lesson is not “ensembles failed, therefore ensembles are useless”. It is that uncertainty inside the ensemble did not capture every possible model limitation. Scientific uncertainty has more than one source.

Prediction is not the same as observation

Reality Lab Vol No.016 distinguishes a simulation from an observation. Forecasting adds another layer: a model output is tied to a future time and will later meet reality.

The most useful cycle is:

observe now → initialise model → predict future → communicate uncertainty → observe later → compare prediction with reality → improve the system.

This is scientific inquiry operating in the world.

Tempting reasoning that fails

“The app says 30°C, so 30°C will happen.” Forecast is not observation.
“The forecast changed, so forecasting is unscientific.” Updating after new evidence is a scientific strength.
“Twelve ensemble members are twelve independent experiments.” They are related model runs, not twelve future observations.
“Wide spread means the model knows nothing.” It means the modeled future is less tightly constrained for that quantity and time.
“Narrow spread means the forecast must be correct.” Shared model limitations can affect many members together.
“Uncertainty makes a forecast useless.” Decisions often improve when uncertainty is communicated rather than hidden.

What evidence would strengthen a forecast claim?

recent observations entering the forecast system;
several model approaches giving similar results for understandable reasons;
small ensemble spread when the forecast system is known to be well calibrated for that situation;
strong past verification for similar lead times and conditions;
agreement with independent observations as the forecast time approaches;
clear communication of update time and uncertainty.

What would weaken it?

a long lead time with rapidly increasing spread;
an old forecast after important new observations became available;
a public headline that hides a very wide range;
a local process the model represents poorly;
many ensemble members that share the same unrecognised dependency;
a claim of certainty that the forecasting system itself does not make.

PSLE-style transfer case

A student heats water for five minutes and records 24°C, 28°C, 32°C, 36°C and 40°C at one-minute intervals. The student predicts 44°C at six minutes.

Is 44°C an observation?

No. It is a prediction based on the observed pattern. The prediction may be reasonable if the same conditions continue, but it has not yet been measured. If the heater switches off automatically after five minutes, the trend will not continue. This connects to the existing owner How to Predict Beyond the Tested Range in PSLE Science Without Pretending the Trend Must Continue.

Practice

1. A forecast app shows 30°C. The ensemble members range from 29°C to 31°C. What can you say?

Answer: The model runs are relatively tightly grouped around 30°C, but 30°C is still a forecast rather than a guaranteed future observation.

2. Another forecast also shows 30°C, but its members range from 23°C to 36°C. Is the same single number carrying the same information?

Answer: No. The second forecast has much wider model spread, so the same displayed centre hides greater uncertainty.

3. A forecast issued yesterday says 28°C. A new forecast issued this morning says 31°C after fresh observations. Which one should be treated as more current?

Answer: The newer forecast, while still checking the same target time, source and uncertainty information.

4. All ensemble members agree closely, but the observed value later falls outside their range. What have you learned?

Answer: Ensemble agreement did not capture every source of error or model limitation. Narrow spread is not proof of accuracy.

Delayed independent return

The next time you see one future number—temperature, rainfall, water level or another forecast—ask three quiet questions: When was this forecast made? How far ahead is it? What range or uncertainty is hidden behind the single display? Then return after the event and compare the forecast with the observation. That final return is what turns prediction into tested scientific learning.

Where to go next

Teaching guide for parents and tutors

Give the learner two cards: observation and prediction. Read statements such as “the thermometer reads 29°C now”, “the app predicts 31°C tomorrow” and “the model predicts water will reach 42°C after six minutes”. Ask the learner to classify each sentence before discussing whether it is accurate.

Next, give two forecasts with the same central value but different ranges. Ask which forecast communicates greater model agreement. This separates the headline number from the uncertainty around it.

Finally, return after the predicted event. Compare what happened with what was predicted. The teaching goal is not to make children distrust forecasts. It is to help them understand prediction as a testable scientific object that becomes more informative when uncertainty, updates and later observations remain visible.

Authoritative sources

The quiet return

A forecast is science speaking before reality has answered. The strongest scientific habit is therefore neither blind belief nor automatic doubt. It is to preserve the prediction, preserve the uncertainty, wait for the observation—and then let the world reply.

Small Group Tutorials