How Do We Know What We Know About the Climate?

Here’s a strange thing: almost every claim you’ve ever heard about climate change—that the planet is warming, that humans are causing it, that the Arctic is melting—rests on information that nobody has ever directly observed. Nobody measured the average temperature of the entire Earth in 1880. Nobody has run an experiment where they doubled the amount of carbon dioxide in the atmosphere to see what happens. And nobody can run that experiment, even if they wanted to.

So how do climate scientists actually know what they claim to know? The answer turns out to be surprisingly complicated—and philosophers have been arguing about whether the methods scientists use are good enough.

What Even Is “Climate”?

Before we can argue about whether the climate is changing, we need to agree on what “climate” means. That might sound simple, but it’s not.

You’ve probably heard someone say that “climate is what you expect, weather is what you get.” The idea is that weather is what’s happening right now (it’s raining), while climate is the pattern over a long period (this place usually gets 40 inches of rain per year). This is the most common definition: climate is average weather, or the statistical distribution of weather, over a long time.

But there’s a problem. If climate is just a description of what actually happened, then “climate change” means something like: the 30-year average temperature changed between the period 1950–1980 and 1990–2020. That’s a perfectly sensible thing to measure. But it doesn’t tell you why it changed, or whether it will change again.

This is why many climate scientists prefer a different way of thinking. They treat the Earth’s climate as a system—the atmosphere, oceans, ice sheets, land surface, and living things all interacting with each other. On this view, climate is a property of that system, and climate change happens when something pushes the system into a different state.

This matters because the two definitions lead to different conclusions. Suppose the 30-year average temperature stays flat for a decade, but the oceans are absorbing lots of heat deep below the surface. Under the “climate is average weather” view, nothing is changing. Under the “climate is a system” view, a lot is changing—it’s just happening somewhere we can’t see. Which definition is right? Philosophers still argue about this.

How Do You Measure a Planet?

Here’s another problem. To know whether the Earth is warming, you need to know the Earth’s average temperature. But nobody has a thermometer that can measure the whole planet at once.

Instead, scientists stitch together millions of individual measurements from thousands of weather stations around the world. But those stations aren’t evenly spread. There are lots in Europe and North America, fewer in the middle of the ocean, almost none in Antarctica and the deep interior of South America. So how do you get a global number from a bunch of local ones?

Scientists do something remarkable: they make the data. They take the station measurements, correct them for known problems (like when a station moves to a new location, or a tree grows next to it and starts shading the thermometer), and then use statistical methods to fill in the gaps. Different scientific groups use different methods, but they all end up with very similar results—the Earth has warmed about 1.1°C since 1850.

But here’s where it gets even stranger. In addition to station measurements, scientists also use data from weather balloons, satellites, ocean buoys, and—most surprisingly—from computer models. These “reanalysis” datasets combine actual observations with physics-based computer forecasts to produce a complete picture of what the atmosphere was doing at every point on Earth, going back decades. Many climate scientists treat these as observations. Others insist they aren’t real measurements at all, because a computer helped produce them.

The Big Question: Are Climate Models Reliable?

This brings us to the most important philosophical issue in climate science. Climate models—the computer programs that simulate the Earth’s climate—are the main tool scientists use to understand the past and predict the future. But how do we know they’re any good?

Climate models work by dividing the atmosphere and ocean into millions of three-dimensional boxes and using equations of physics to calculate what’s happening in each box. A simple model might have boxes 500 kilometers across. A cutting-edge model might have boxes just 10 kilometers across. The smaller the boxes, the more detail you can capture—but the more computing power you need.

Here’s the catch. Even with tiny boxes, many important processes happen at scales smaller than the boxes. Clouds, for instance, are often smaller than a single model box. But clouds have a huge effect on climate—they reflect sunlight and trap heat. So modelers have to parameterize clouds: they use mathematical formulas that estimate, based on what’s happening in the box as a whole, how much cloudiness there should be.

This is where things get philosophical. Modelers take known physics (the equations that describe how air moves), add semi-empirical guesses (how much cloud to expect), and then tune the model—adjusting numbers until the model matches past observations. If you then claim the model can predict the future, you’re essentially saying: “Because this model matches the past really well, it will also work for conditions we’ve never seen before.” But that’s a leap of faith. A model could match the past for the wrong reasons.

Some philosophers argue this is a serious problem. Others say it’s not as bad as it sounds, because models are built on physical understanding, not just curve-fitting. The debate is very much alive.

How We Know Humans Are Causing Warming

One of the strongest arguments that humans are causing climate change doesn’t come from computer models at all. It comes from basic physics.

We’ve known since the 19th century that carbon dioxide traps heat. You can measure this in a lab. You can calculate it with equations. This isn’t controversial. What the computer models do is help scientists ask: how much of the observed warming is due to CO₂, and how much might be due to natural causes like changes in the sun’s output or volcanic eruptions?

To answer this, scientists use a technique called “fingerprinting.” They run climate models with only natural factors changing (sun, volcanoes) and see what pattern of warming emerges. Then they run models with only human factors (greenhouse gases, aerosols) and see what pattern emerges. The human fingerprint looks different from the natural one—it warms the lower atmosphere and the surface, while the upper atmosphere actually cools slightly. Natural factors don’t produce that pattern.

When scientists look at the actual observations, the pattern matches the human fingerprint. The IPCC, the international body that assesses climate science, has concluded it is “extremely likely” (meaning more than 95% probability) that more than half of the observed warming since 1950 is due to human activities. This conclusion has gotten stronger over time as more evidence has accumulated.

When Scientists Disagree

But not everyone agrees. Some scientists—a small minority—argue that the evidence isn’t strong enough. Philosophers have gotten involved in this debate too.

One philosopher, Joel Katzav, has argued that the claim that humans caused most of the recent warming hasn’t passed a “severe test”—meaning, scientists haven’t done enough to try to prove themselves wrong. For example, our estimates of how much the climate varies naturally might be wrong, and if they’re wrong by enough, the human fingerprint might not be as clear as scientists think. Katzav thinks we should be more cautious about what we claim to know.

Other philosophers push back. They point out that even if our estimates of natural variability were three times larger than current ones, the warming would still be detected. They also note that confidence in the human cause of warming doesn’t come from a single study but from many different lines of evidence, all pointing in the same direction.

This is worth taking seriously. When scientists themselves disagree, it’s easy for non-scientists to throw up their hands and say “nobody knows.” But that’s not right either. The fact that the consensus is very strong doesn’t mean it’s wrong—and the fact that a few scientists dissent doesn’t mean they’re right.

The Big Puzzle That Remains

Here’s what keeps philosophers up at night. Climate models are enormously complex—millions of lines of code. They’re built by thousands of people over decades. Nobody fully understands every part of them. And yet we need them to make decisions about the future.

How much confidence should we have in their projections? When multiple models agree on a prediction, does that mean the prediction is more likely to be right? Not necessarily—all the models might share the same blind spot. When models disagree, how do we decide which ones to trust?

These aren’t just academic questions. Real decisions are being made right now—about where to build sea walls, what crops to plant, how much to spend on cutting emissions—based on information from these models. Philosophers are trying to figure out what it means for a model to be “good enough” for these purposes, and whether the science is actually there yet.

The honest answer is that nobody knows exactly how reliable climate projections are. Scientists are working hard to figure it out. Philosophers are helping by asking the right questions. And the rest of us have to live with the uncertainty, make the best decisions we can, and keep paying attention as our understanding evolves.

---

Appendix

Key Terms

Term	What it does in this debate
Climate	A disputed concept: either average weather over a long period, or a property of the interacting Earth system (atmosphere, oceans, ice, land)
Climate model	A computer program that simulates the Earth’s climate using physics equations; the main tool for understanding past and predicting future climate
Parameterization	A mathematical shortcut that estimates what’s happening at scales too small for the model to simulate directly (like clouds)
Tuning	Adjusting a model’s numbers so it matches past observations; a necessary but controversial step
Fingerprinting	Comparing the pattern of observed climate change to the patterns predicted for different causes (human vs. natural)
Internal variability	Natural climate fluctuations that happen even without any external push, like changes in ocean currents
Reanalysis	A dataset that combines actual observations with computer model forecasts to create a complete picture of past weather
Detection	Showing that climate has changed in a way that can’t be explained by natural variation alone
Attribution	Identifying the cause(s) of an observed climate change

Key People

• Svante Arrhenius (1859–1927): Swedish chemist who first calculated, in 1896, that doubling CO₂ in the atmosphere would warm the Earth by several degrees—by hand, without any computer.
• James Hansen (1941–): NASA climate scientist who testified to the US Congress in 1988 that global warming was already happening, bringing the issue to public attention.
• Joel Katzav (contemporary): Philosopher of science who argues that the evidence for human-caused warming might not be as strong as mainstream science claims, because the possibility of large natural variability hasn’t been ruled out.
• Wendy Parker (contemporary): Philosopher who has written extensively about how we should evaluate climate models and whether they can be trusted for different purposes.

Things to Think About

1. If you needed to decide whether to build a sea wall to protect a coastal city, how much confidence would you need in climate projections? Is “more likely than not” good enough, or do you need “virtually certain”? Who gets to decide?

2. Imagine two climate models: one that matches past temperature observations almost perfectly but uses a very simple approach, and another that doesn’t match as well but includes many more physical processes. Which would you trust more for predicting the future? Why?

3. The article mentions that climate models are “tuned” to match past data. Some philosophers say this means we shouldn’t treat their past performance as evidence of future reliability. Others say tuning is fine as long as we understand why the model works. What do you think?

4. When a small number of scientists disagree with a strong consensus, how should the public decide who to believe? Is it rational to base your belief on the number of scientists who agree, or should you examine the evidence yourself? What are the limits of each approach?

Where This Shows Up

• When you hear about a “hiatus” or “pause” in global warming, you’re seeing a real tension between different definitions of climate and different ways of measuring it.
• The debate about whether climate models can be trusted shows up every time there’s a controversy about a specific prediction—like whether a particular hurricane season will be worse because of climate change.
• The question of how to combine data from many imperfect sources is a problem in many sciences, not just climate: epidemiologists trying to track disease spread, astronomers trying to map the universe, and economists trying to predict recessions all face similar challenges.
• Companies and governments that need to adapt to climate change are now wrestling with exactly the philosophical questions about uncertainty and decision-making discussed in this article.