Ask anyone who works with climate projections for a living and they'll tell you the same thing: the question they get most often isn't about physics or emissions. It's "so how do you actually know?" Which is fair. When someone tells you the planet will be 3°C warmer in 2100, you deserve to know how that number was produced, what could break it, and how much of it is real versus how much is a very educated guess.
That's what this article is about. Not the scary headlines, not the political fights. Just the machinery: how climate models predict future temperature rise, where they've been wrong, and what that means for the specific years you're probably worried about.
Key Takeaways
- Climate models aren't crystal balls. They're physics engines that simulate how energy moves through the atmosphere, oceans, ice and land.
- They've predicted past warming remarkably well. Where they struggle is with speed and with extremes.
- Every projection comes with a range, not a single number. The range is the honest part.
- The year 2035 matters because it's the deadline for halving emissions to stay near 1.5°C, not because the world ends then.
- What happens between now and 2050 depends far more on what we emit than on what the models get wrong.
How climate models actually predict temperature rise
A climate model is a very large piece of software. We're talking hundreds of thousands of lines of code, run on supercomputers that consume enough electricity to power a small town. Inside it, the planet is sliced into a three-dimensional grid. Each cell — some are 100 km wide, some closer to 25 km in the newer high-resolution models — holds values for temperature, humidity, wind speed, pressure, cloud cover, and a dozen other variables.
Then the model does something almost mundane: it applies the laws of physics, cell by cell, step by step, forward in time.
The physics behind the code
Energy comes in from the sun. Some bounces off clouds and ice. Some is absorbed by the surface. Heat moves through the atmosphere and into the ocean. Water evaporates, condenses, falls as rain. Ice melts and reflects less light, which means more absorption, which means more melting. CO₂ and methane trap outgoing infrared radiation. None of this is mysterious. It's textbook thermodynamics and fluid dynamics, just applied several million times per simulated hour.
The hard part isn't any single equation. It's the feedback loops. Clouds are the classic example. A warmer atmosphere holds more water vapour, and water vapour is itself a greenhouse gas — that's a reinforcing loop. But more water vapour also means more clouds, and clouds can either trap heat or reflect sunlight depending on their altitude, thickness and size. Getting clouds right has been the single biggest source of uncertainty in climate modelling for decades.
Why humans are the variable that breaks everything
Here's the part outsiders rarely appreciate. The physics half of a climate model is relatively settled. The half that causes the most disagreement is the human half.
Models don't predict emissions. They can't. Nobody can predict whether the world will build 400 new coal plants or shut down 400. So modellers do something clever: they run the same physics engine under a handful of different human storylines. One where emissions keep climbing. One where they plateau. One where they fall fast. Each storyline gets a label, and each produces a different temperature curve.
That's why you see projections like "1.5°C to 4.5°C by 2100" rather than a single figure. It's not that scientists are confused. It's that the answer genuinely depends on choices no model can make for us.
How scientists check whether models are worth trusting
This is the question that separates a rigorous projection from an astrology chart. And the answer is more interesting than most coverage lets on.
Backcasting: the real test
The primary validation method is called hindcasting, or backcasting. You take a model, feed it only the data available in, say, 1970, and let it run forward. Then you compare its output to what actually happened. If it can't reproduce the past, you don't trust it for the future.
And here's the thing — they mostly can. When I first dug into the historical model runs out of curiosity, I expected to find a graveyard of failed predictions. Instead I found something closer to a track record. Models built in the 1970s and 1980s, running on computers with less power than your phone, produced warming curves that line up surprisingly well with observed temperatures.
Paleoclimate: the deep-time check
There's a second validation layer that gets almost no public attention. Scientists test models against ancient climates — the last ice age, the mid-Pliocene warm period around 3 million years ago, the aftermath of volcanic eruptions. If a model can reproduce a world with no ice sheets over North America, or a world with CO₂ levels similar to today's, that's strong evidence its physics are sound.
It's slow, unglamorous work. But it's the reason climate scientists are so confident about the direction of change even while debating the exact magnitude.
What the models still get wrong
Honesty matters here. Models have a documented tendency to underestimate certain extremes. Heatwaves in particular have repeatedly come in hotter than the central projections suggested. The 2021 Pacific Northwest heat dome broke local records by margins that startled researchers. Whether that's a model failure or a symptom of natural variability is still being argued.
There's also a known regional bias: some land areas warm faster than the global average, and the grid resolution of older models smoothed those hotspots into oblivion. Newer, finer-grid models are catching them, but the corrections are ongoing.
The specific years you keep hearing about
People don't ask about 2100. They ask about 2030, 2035, 2050. Those are the years that feel real. So let's deal with them properly.
Will 2030 be hotter than 2026?
Yes, and this is one of the more confident predictions in the field. The underlying trend is upward, and it's driven by accumulated greenhouse gases that don't disappear quickly. Year-to-year noise from phenomena like El Niño and La Niña can mask or amplify the trend temporarily, but the direction is not seriously contested. A single cooler year in between wouldn't change that. By 2030 you should expect the average year to be warmer than the average year now. Individual years may surprise you. The direction won't.
What do 97% of scientists believe is causing climate change?
The overwhelming majority of climate scientists attribute recent warming primarily to human activities — chiefly the burning of fossil fuels, which releases CO₂ and other heat-trapping gases. That position rests on multiple independent lines of evidence: the physics of radiation absorption, the isotopic fingerprint of fossil-fuel carbon in the atmosphere, and the fact that observed warming patterns match what human influence should produce while natural factors alone don't account for it. The "97%" figure comes from surveys and literature reviews, and while the exact number varies slightly depending on methodology, the consensus itself is not in serious dispute within the field.
Why is 2035 the year of no return?
It isn't, strictly speaking. The "2035" framing comes from a specific policy benchmark: to have a reasonable chance of holding warming near 1.5°C, global emissions need to fall roughly in half from their peak by around 2035, and reach net zero by mid-century. Miss that window and 1.5°C becomes very unlikely — but "unlikely" isn't "impossible," and it certainly isn't "the world ends." Overshooting 1.5°C doesn't trigger a cliff. It makes the consequences worse and the recovery harder. The framing as a hard deadline is a communication choice, not a physical law.
How hot will the Earth get by 2050?
Under most plausible emission paths, the global average surface temperature by 2050 lands somewhere around 1.5°C to 2°C above pre-industrial levels. The difference between those two numbers isn't trivial — it's the difference between coral reefs mostly surviving and mostly not. But 2050 is close enough that a good chunk of the warming is already locked in by past emissions. The decisions that matter for 2050 are mostly decisions already made. The decisions that matter for 2100 are still open.
Comparing the scenarios side by side
Here's roughly how the main emission storylines translate into temperature outcomes, based on the ranges scientists work with. These aren't exact predictions — they're the ranges the physics supports under each assumption.
| Emission pathway | Roughly what happens by 2050 | By 2100 | What it assumes |
|---|---|---|---|
| Rapid decarbonisation | ~1.5°C | ~1.5–1.8°C | Emissions halve by 2035, net zero by 2050 |
| Moderate action | ~1.7°C | ~2–2.5°C | Slow decline in emissions after 2030 |
| Current trajectory | ~2°C | ~3–4°C | Emissions plateau, then fall slowly |
| High emissions | ~2.2°C | ~4–5°C | Coal and gas expansion continues |
Notice what the table doesn't show: a single number. That's deliberate. Anyone offering you one — in either direction — is selling something.
The gap between models and reality
Reality has been running slightly hotter than the central projections in several recent years. That's worth sitting with, because it cuts against the sceptical argument that models exaggerate. If anything, some of them have been too conservative.
Why this gap exists
Three plausible reasons. First, aerosol pollution. Burning coal and oil releases particles that reflect sunlight and cool the surface. As the world cleans up its air, that cooling mask comes off, and the warming underneath shows through faster than expected. Second, natural variability: we've had a run of El Niño conditions that push temperatures up temporarily. Third, and most unsettling, some feedbacks may be stronger than the central estimates assume.
I'll admit I underestimated that first one for a long time. Aerosols seemed like a footnote. They're not.
What this means for your decisions
If you're planning anything with a 30-year horizon — a farm, a coastal property, a municipal water system, an insurance portfolio — the practical takeaway is to plan against the upper end of the range, not the midpoint. That's not alarmism. It's basic risk management. You don't build a bridge for the average flood.
What still can change
Everything between now and 2100 is conditional. The models don't know what we'll choose. They only know what physics does with whatever we choose.
And that's the part the public conversation keeps missing. Climate modelling is sometimes framed as fatalism — a machine telling us the future is already written. It's the opposite. A model is a way of asking "if we do X, what happens?" That question only has meaning because we get to pick X.
The next time someone declares that the science is settled, ask which part. The physics is settled. The range is settled. What happens next isn't. That's the whole point.
