A model represents the world as reservoirs connected by fluxes.
Reservoir A store of matter or energy in a system model.
Step 1: Something to Notice
Watch first. The explanation comes later.
Two models can fit the same historical record equally well and disagree completely about the next century.
How do you tell which one to believe?
Step 2: Find Out
Isolate the level of detail and note what each zoom leaves out.
Step 1 — Predict
A model reproduces the past record exactly. What does that establish?
Choose what you think will happen. You cannot see the experiment until you do — guessing first is what makes it worth watching.
Step 3: So Here Is Why
Now the explanation, after you have seen it happen.
Every model leaves things out
A model including everything would be as complex as the thing itself.
Fitting the past is a minimum, not a test
Enough adjustable parameters will match almost any record.
The test is predicting what was held back
A later period, a different region, or a variable the model never saw.
And the output should be a range
Its width says how much the answer depends on assumptions nobody can pin down.
Matching the past is not yet a prediction
A model with enough adjustable parameters can be tuned to reproduce any history. Confidence comes from predicting data the model was not fitted to, which is a much harder test.
Uncertainty is quantified, not hidden
Good models report a range and say where it comes from — measurement error, unknown parameters, or scenario choices. A single number with no range is the less trustworthy output.
Step 4: A Common Mistake
Lots of people think
“A model that reproduces past data must be right.”
Step 5: Why Weather and Climate Models Differ in Kind
The same idea somewhere new.
A weather model predicts a specific state — rain in this place on Thursday — and becomes useless within about two weeks, because tiny errors in the starting conditions grow. A climate model predicts statistics: the distribution of temperatures over decades, not any particular day. That is why "they cannot forecast next week, so why believe a century" compares two different claims. Predicting the average of many rolls is easier than predicting one roll, not harder.
Step 6: Think It Through
Practice makes it stick.
A Perfect Fit
Problem 1 of 2
A model reproduces the past record exactly. What does that establish?
Next Week and Next Century
Problem 2 of 2
Why can climate be projected further ahead than weather?
What a Model Shows
1 of 5
What is a reservoir in a system model?
2 of 5
Why does a model leave things out?
3 of 5
How is a model genuinely tested?
4 of 5
What does the width of a projected range tell you?
5 of 5
Weather and climate models differ how?
Step 7: Quick Check
Show what you know.
Question 1 of 1
A model reproduces the past perfectly. Is it right?
Step 8: Explain It in Writing
Claim, evidence, then reasoning.
The question
Explain how an Earth system model should be tested, and why fitting past data is not enough.
Fill in all three boxes. The reasoning box is the one that matters most.
What You Found Out
- Models represent the world as reservoirs connected by fluxes.
- Every model leaves things out, and that is what makes it useful.
- Fitting past data is a minimum requirement, not evidence.
- A model is tested by predicting what it was not fitted to.