What Climate Models Can Tell Us and What They Don't

Read...

I've hinted we want to use these models to predict the future. And we'll get to that soon. But how do we know they are accurate? After all, if weather forecasts can’t always get it right two weeks out, how on Earth could a climate model possibly forecast characteristics of the climate 100 years from now? And I’ll say right now—that’s a fair question!

But here’s the thing: it’s not exactly a fair comparison. Comparing weather models to climate models is a bit of an “apples and oranges” situation. To produce accurate short-term weather forecasts, we need to get very specific about the details—high- and low-pressure locations, wind directions, exact temperatures, areas of precipitation, and so on, to start the forecast. These details are the initial conditions of the atmosphere, and to produce a perfect forecast, a model would need to start with a perfect set of initial conditions, i.e., a perfect picture of the atmosphere’s state (exact wind, temperature, humidity, etc.) everywhere on Earth at the same time. Because it’s impossible to measure every inch of the atmosphere at all times, achieving perfect accuracy is out of reach. So, while weather forecasts are generally good in the short term (a few days out), small errors in our “starting” conditions mean the forecast gradually diverges from reality when we try to look a week or more ahead.

The goal of a weather model is to predict the very specific evolution of a single weather feature, like a hurricane. Climate models cannot do this, so they target the "statistics" of weather features -- like hurricanes!

Credit: Ernesto 2006 Model Spread by Richard J. Pasch, Mike Fiorino, and Chris Landsea from Wikimedia (Public Domain).

On the other hand, climate models aren’t attempting to predict day-to-day weather patterns. They’re not designed to tell you if a hurricane will be in the Gulf of Mexico on August 14, 2067, or if the winter of 2078 will be snowier than usual. Instead, climate models focus on projecting large-scale climate trends (like changes in temperature, melting ice, and average rainfall) over decades. These projections are much less dependent on today’s exact weather. In other words, how warm the climate might be 100 years from now has very little to do with today’s specific wind direction or temperature at a given location. That means the small errors in initial atmospheric conditions that affect short-term forecasts don’t have the same impact on climate models.

Why? Because climate models are about predicting statistics and probabilities, not precise weather events. We’ve talked about dice-rolling a lot in this class, but imagine flipping a coin. You and I can’t predict accurately if the next flip will land heads (or tails), but we don’t need a computer model to know that over 1,000 flips, we’ll end up close to a 50/50 split between heads and tails. That’s how climate models work: they’re not trying to predict every “flip” but rather the long-term statistics, e.g., averages and trends, of important quantities.

Take some time, blow off some steam, and play with the coin flipper below.

Those statistics are what we are interested in with climate modeling. For instance, they enable us to estimate the likelihood of certain outcomes, such as the probability of having a hotter-than-average year or the frequency of extreme weather events like hurricanes or droughts. Should people living along the Gulf of Mexico expect more or less heavy rainfall? Do we think the amount of snow falling over the ski resorts of the Rockies will increase, stay the same, or decrease moving forward?

We’ve already seen a figure like the one below, and I want to emphasize that this is exactly what a climate model is trying to predict. There are two probability curves: the gray one represents the current climate (let’s say the year 2020), while the black one represents a future climate (let’s say 2080). The horizontal axis is temperature, and the vertical axis shows how likely a particular temperature is to occur. There aren’t specific numbers on the axes, so feel free to imagine whatever region you like, anywhere from Miami to Fairbanks! While we can’t predict the exact temperature on, say, July 17th, 2080, the climate model gives us many “coin flips” or “dice rolls” for the days around that time. With enough of these simulations, we can build a distribution of likely temperatures.

In this case, the model shows us that, on average, days will be warmer. The coldest days won’t be as cold, and the hottest days will be even hotter. There’s still some overlap between the two curves, but the entire distribution shifts to higher temperatures. It’s a clear indication of how even small changes in the average climate can lead to noticeable shifts in extreme events.

Future Climate Shift

Credit: Future Climate Shift by US Climate Change Science Program / Southwest Climate Change Network (Public Domain)

Now, let's discuss limitations. Don’t let me oversell things—climate models aren't perfect, but they’re still invaluable tools. British statistician George Box said it best: "All models are wrong, but some are useful."

One significant source of uncertainty in climate models is the complexity of ocean processes. Oceans are Earth’s largest carbon reservoir, absorbing about half of the carbon dioxide emissions humans have generated so far, which has helped limit atmospheric warming. However, as the oceans take in more CO₂, their capacity to absorb it diminishes. The rate at which this happens depends on intricate interactions within the ocean that models don’t fully capture. Despite these uncertainties, ocean processes—and their immense heat capacity—are critical in shaping how quickly atmospheric temperatures rise in the future.

Another challenging area for climate models is clouds and water vapor. Clouds have a dual role: they cool the Earth by blocking incoming solar radiation during the day but also warm it by trapping infrared radiation. Which effect dominates depends on the types of clouds that form. As the planet warms and evaporation increases, the atmosphere will hold more water vapor—the most abundant greenhouse gas—likely leading to more cloud cover. But what kinds of clouds will dominate? Will the cooling or warming feedbacks of clouds prevail overall? These unanswered questions make cloud behavior one of the largest sources of variability in future climate projections.

Even with these uncertainties (and others), climate models remain essential tools for exploring potential future climates. They provide valuable insights by simulating how the climate system responds to different factors, even if some details are harder to predict. For instance, while the exact impact of increased cloud cover remains uncertain, models consistently predict continued warming if greenhouse gas emissions remain high. This agreement across multiple models and scenarios bolsters our confidence in the overall projections of climate change.

Moreover, climate models are constantly improving. Advances in satellite monitoring, ocean buoys, and other observational tools provide higher-quality data that enhance model accuracy. Researchers are also refining the mathematical representations of complex processes, such as ocean-atmosphere interactions and cloud dynamics, to reduce uncertainties. When scientists look back 20 years, they’re astounded by how much progress has been made—and we can only hope this trend continues!