A climate model is not a guess dressed up in maths, and it is not a crystal ball either. It is a set of physical equations, fed with real data, run forward to see what happens. Understanding that middle position — genuinely useful, genuinely uncertain — is what HL questions on this topic are testing.
📚 What you need to know
Climate models are mathematical simulations of the interactions between atmosphere, oceans, land and ice.
Common inputs include greenhouse gas concentrations, solar radiation, volcanic activity and land surface changes.
They work by solving equations for atmospheric circulation, ocean currents and the water cycle, across thousands of variables.
Hindcasting runs a model backwards in time and compares its output with the real record, to test whether it can be trusted.
Uncertainty comes from imperfect input data (especially ancient proxies) and from simplifying complex natural processes.
Models therefore produce a range of outcomes, not one precise forecast — and the range itself is useful information.
What a model actually is
Take everything we know about how air moves, how oceans transport heat, how water evaporates and condenses, and how ice reflects sunlight. Write it as equations. Divide the planet into a three-dimensional grid of boxes, from the deep ocean up through the atmosphere. Then, for each box, calculate what happens over a short time step and pass the result to the neighbouring boxes. Repeat, millions of times, and you have simulated a century of climate.
The important consequence is that a climate model is not fitted to the temperature graph. It is built from physics, and then compared with the graph. That is why it can say something about a future that has not happened yet.
The last output on the list is the one students forget. A model that gave a single number would be pretending to a precision it does not have — the spread is an honest part of the answer.
Testing a model: hindcasting
Here is the obvious problem. If a model predicts the year 2080, how do you check it before 2080 arrives? The answer is hindcasting: you run the model over a period that has already happened, using only the inputs that were true at the time, and compare what it produces with what actually occurred.
If the model reproduces the twentieth century — including the cooling after big volcanic eruptions and the warming since the 1970s — then its physics is capturing something real. If it does not, you can see exactly where it went wrong and improve it. Hindcasting is the reason we have any grounds for confidence in projections at all.
Most of the width of that fan is not scientific uncertainty. It is emissions uncertainty — the models disagree far less about the physics than we disagree about our own future behaviour.
What models predict
Temperature
Models are run under different greenhouse gas emission scenarios, and higher emissions always give greater warming. The usual headline is a rise somewhere between about 1.5 °C and 4 °C by 2100, depending on the pathway taken. The spread reflects emissions choices as much as scientific doubt.
Sea level
Projections combine melting land ice with thermal expansion of warming seawater. Under high-emission scenarios, some models suggest a rise of roughly 0.5 to 1 metre by 2100. Ice sheet behaviour is one of the harder things to model, which is part of why this range is wide.
Precipitation
Rainfall is the hardest of the three, because it depends on regional circulation. The general picture is that wet places get wetter and dry places get drier: some regions face more frequent and more intense rainfall, while others become drier and more drought-prone. Models often show increased rainfall in parts of the northern hemisphere alongside drier conditions in parts of Africa.
The honest summary
confident about direction • less confident about magnitude • least confident about regional detail
Limitations and uncertainty
Two things limit any model, and you should be able to state both.
Imperfect input data. Recent measurements are excellent, but reconstructions of ancient greenhouse gas levels and temperatures come from proxies with their own error ranges. Feed uncertain numbers in, get uncertain numbers out.
Simplification. No model can represent every cloud, eddy or patch of soil. Processes smaller than a grid box have to be approximated, and clouds in particular are genuinely hard: they both reflect sunlight and trap heat.
Because of this, different models give slightly different results for the same scenario. That is not a failure. The spread across many independent models tells scientists how much of the answer is robust and how much depends on modelling choices, which is exactly what planners need to know when deciding how high to build a sea wall.
Notice the trap in “models are uncertain, so we should not act on them”. A model that says the rise is somewhere between one and four metres is telling you very clearly to prepare. Uncertainty cuts both ways — the outcome could be worse than the middle estimate, not just better.
WORKED EXAMPLE
Explain how hindcasting is used to assess the reliability of a climate model. [3]
Step 1 — what is done
The model is run backwards over a past period, using the greenhouse gas concentrations, solar output and volcanic activity that actually applied then.
Step 2 — the comparisonIts simulated climate is compared with the recorded climate for the same period, from instruments and proxies.Step 3 — the conclusion drawnClose agreement suggests the physics is well represented, so its future projections carry more weight; disagreement shows where the model must be improved.3 / 3Saying “it is tested against the past” alone is one mark. The three steps get you all three.
WORKED EXAMPLE
Discuss the value of climate models given their limitations. [4]
Limitation 1
Input data is imperfect, particularly reconstructions of past greenhouse gas levels from proxies.
Limitation 2Complex processes such as cloud formation must be simplified, so different models give slightly different results.Value 1They are validated by hindcasting, and all credible models agree on the direction and rough scale of warming.Value 2 and judgementA range of outcomes still allows governments to plan flood defences, agriculture and emissions targets, so models remain essential despite the uncertainty.4 / 4“Discuss” needs both sides plus a conclusion. Two limitations, two values, one judgement.
💡 Exam tip
Define a model as a mathematical simulation based on physical equations, not as a prediction.
Learn four inputs by heart: greenhouse gases, solar radiation, volcanic activity, land surface changes.
Use the word hindcasting and explain it in three steps: run backwards, compare, judge.
Always give the two sources of uncertainty: imperfect data and simplification.
Quote a range, not a single number, when giving projections. It shows you understand what a model produces.
If a question says “discuss” or “evaluate”, finish with a clear judgement sentence.
⚠️ Common mix-up
“Models are just predictions.” They are simulations of physical processes, tested against the observed record.
Confusing hindcasting with forecasting. Hindcasting looks backwards to test; forecasting looks forwards to project.
Treating uncertainty as a reason to ignore the results. The uncertainty is quantified, and the risk runs in both directions.
Assuming all models must agree exactly. They differ in detail and agree on direction, which is the point.
Forgetting emissions scenarios. Much of the range comes from human choices, not from the science.
Giving one number for 2100. A single figure with no range signals that you have missed the main idea.
Up next: Climate Tipping Points — what happens when the system stops changing gradually and jumps to a new state instead.
Want this explained one-to-one?
Book a free session with an experienced IB ESS tutor and get your trickiest topics made simple.
