Climate models play a central role in how scientists understand Earth’s future climate. They inform policy discussions, risk assessments, and long-term planning.
Yet despite decades of research and rapidly increasing computing power, climate models still do not fully agree on key outcomes, such as how sensitive global temperatures are to rising greenhouse gas concentrations.
This disagreement is often portrayed as a failure. In reality, it is an expected and normal feature of modelling complex systems.
A climate model is not a prediction machine. It is a numerical simulation of physical processes, based on:
Conservation of energy
Fluid dynamics of the atmosphere and oceans
Radiative transfer of energy
Chemistry of greenhouse gases and aerosols
These equations are solved over a three-dimensional grid covering the Earth, step by step through time.
Because the planet is vast and processes occur at many different scales, models must approximate many phenomena rather than calculate them directly.
The main reason climate models disagree is not disagreement about basic physics.
All modern models agree on foundational points:
Carbon dioxide absorbs infrared radiation
Increasing greenhouse gases increases Earth’s energy retention
Global temperatures rise as a result
Where they differ is in how strongly various feedback mechanisms respond.
A feedback is a process that amplifies or dampens warming. The most important ones include:
Water vapor feedback
Cloud formation and cloud height
Ice and snow reflectivity (albedo)
Ocean heat uptake and circulation
Clouds are a particularly difficult example. They can both cool (by reflecting sunlight) and warm (by trapping heat), depending on altitude, thickness, and location.
Small differences in how models represent clouds can lead to meaningfully different long-term outcomes.
A commonly discussed metric is climate sensitivity — how much the global average temperature rises if atmospheric carbon dioxide is doubled.
Despite decades of research, credible estimates still span a range rather than a single value.
This is not because scientists are guessing, but because:
Feedbacks interact non-linearly
Some processes are below model resolution
Long-term observational data is limited
Different models, using equally valid physics but different assumptions, therefore produce different sensitivities.
It is tempting to assume that faster computers will eliminate disagreement. They help — but only up to a point.
Many uncertainties are not computational, but structural:
How should unresolved processes be approximated?
Which parameterizations best reflect reality?
How do small-scale processes scale upward?
More precision does not automatically create more certainty when the system itself is complex and sensitive to initial assumptions.
In everyday language, disagreement suggests confusion. In science, it often means active investigation.
Model spread allows researchers to:
Identify which processes matter most
Test sensitivity to assumptions
Compare simulations against real-world observations
Improve future models iteratively
Agreement too early can be more dangerous than disagreement, as it may hide shared blind spots.
Even with differences, climate models are reliable at several levels:
Directional trends (warming vs cooling)
Large-scale patterns (land warms faster than oceans, Arctic amplification)
Relative comparisons between scenarios
They are less precise at:
Exact temperature values decades ahead
Regional precipitation changes
Short-term variability
Understanding this distinction is key to using models responsibly.
Climate models still disagree because they simulate one of the most complex systems humans have ever studied.
That disagreement does not undermine climate science — it reflects its realism.
Uncertainty is not a flaw to be hidden, but a boundary to be acknowledged.
In science, progress often comes not from forcing agreement, but from understanding why agreement has not yet been reached.
