Climate Models Are More Accurate Than People Think

In 1970, the climate scientist Syukuro Manabe published a paper that made one of the most remarkable predictions in the history of climate science. Building on earlier work with Richard Wetherald, Manabe's climate model forecasted that global temperatures would rise by 0.57 degrees Celsius from 1970 to 2000. Actual temperatures went up by a very close 0.54 degrees Celsius over that period, and Manabe went on to win the 2021 Nobel Prize in Physics for his stunning prediction.

Climate models are obviously not perfect. They're models, after all, and like all models they need to make simplifying assumptions about the world. Furthermore, there are still many issues in climate science that are not fully settled, from the impact of clouds on global temperatures to other broader issues related to estimating climate sensitivity. But for all their imperfections, climate models get far more right than their detractors acknowledge.

Contrary to widespread misperceptions, it has been convincingly demonstrated by now that the majority of climate scientists in the 1970s believed that the Earth was headed for a period of warming, not cooling, in the decades ahead. The cooling narratives of the 1970s were largely a media phenomenon, amplified by a small number of scientists who wanted to make a lot of noise. If you've heard about this kerfuffle and as a result developed the impression that climate scientists got it wrong back in the 1970s, the results above should quickly dispel any such notions. With that out of the way, let's examine the performance of major climate models and try to understand why they've become so good.

The Performance of Climate Models

In 2020, Zeke Hausfather and his collaborators published a famous study that evaluated the performance of several major climate models over the past few decades. The research group found that most of the models considered in the study made fairly accurate predictions of global mean surface temperatures (GMST).  They compared the predictions of 17 sophisticated models developed from 1970 to 2007. Some of these models were actually used in prior IPCC assessment reports. Figure 1 below shows a high-level summary of their results. According to the research group:

Climate models published over the past five decades were skillful in predicting subsequent GMST changes, with most models examined showing warming consistent with observations, particularly when mismatches between model‐projected and observationally estimated forcings were taken into account.

Crucially, there was no evidence that the models collectively overpredicted or underpredicted actual global temperatures, as the results shown below demonstrate.

Figure 1: Results from the 2020 paper. At the top, the researchers show the temperature change per decade from actual observations and predicted model results. The bottom half show actual vs. predicted results for change in temperature after accounting for changes in radiative forcing.

How have climate models posted these impressive numbers? To get a basic sense of the answer, it's worth noting that climate models have changed considerably over time. Advances in computational power have allowed them to increase their spatial and geographic resolution, which means that the latest models can now more accurately capture the behavior of phenomena that unfold over small spatial scales, like cloud formation and oceanic convection. The most powerful General Circulation Models (GCMs) are absolute monsters that run on supercomputers. GCMs break up the world into 3-D grid cells and model oceanic, atmospheric, and terrestrial dynamics as an integrated system through coupled equations. They are some of the most powerful scientific models that humans have ever devised, for any purpose. But despite their impressive performance and improvement over time, some people are not convinced that these models are accurate at all.

What the Contrarians Get Wrong

Let's now look at the many flawed arguments and methodologies coming from the most prominent climate denialists and naysayers, who argue that climate models are mostly useless and inaccurate. Roy Spencer and John Christy are two of the more notable climate contrarians. They have become famous on the right for peddling ridiculous arguments about global warming.

Spencer and Christy first gained notoriety for a 1995 study which claimed that the lower troposphere was actually cooling, not warming. But a later paper proved that this 1995 study contained numerous major errors, and the same is also true of their later work. For example, the original 1995 study failed to take into account the effects of orbital decay. If a satellite is going to measure an accurate average temperature, it must pass over the same point on Earth at the same time in each day. But in reality, the time at which the satellite passes over a particular point on any given day varies because the orbit of the satellite drifts over time, and this orbital decay changes the speed of the satellite as it goes around the Earth. The new orbital speed affects the time at which the satellite records the data. That means the data collected needs to be adjusted to avoid spurious relationships and confounding effects. The denialist duo made several other mistakes, including not properly merging data from different satellites that relied on different sensors. Once these mistakes were fixed, the results completely flipped and showed that the troposphere had indeed been getting warmer.

In recent times, Christy generated controversy once again by creating a chart that supposedly shows a huge divergence between the average predicted bulk atmospheric temperature of roughly 100 climate models and the average of actual observed temperature data from a small number of balloon and satellite datasets (see Figure 2 below). This chart has been thoroughly debunked by now, so let's review some of its major problems.

Figure 2: Misleading chart produced by John Christy, purporting to show that climate models are inaccurate.

First, the chart aims to show results from the mid-troposphere. And yet, we do not live on Mount Everest or high up in the sky; we live on the surface of the planet, and thus bulk atmospheric temperature readings from the surface to 50,000 feet are largely irrelevant. When confined to surface temperature predictions, climate models actually do quite well, as Hausfather and his collaborators have shown. Second, Christy's chart has absolutely no uncertainty estimates at all, which is a huge flaw. In scientific studies, it's common to attach a range of uncertainty to model predictions. Christy doesn't bother with any of that. As a result, one simply doesn't know if the observed average temperatures fall within the range of uncertainty of the model predictions. Luckily, others have created those uncertainty estimates and have shown that they cover the observed temperatures quite well.

Third, Christy relies heavily on satellite data for the observed bulk temperatures. But as climate scientists have known for a long time, satellite data is often unreliable, for some of the reasons mentioned above. For example, satellites experience orbital drift and decay over time, which compromises the integrity of their measurements. Furthermore, unlike what many denialists seem to believe, satellites do not record unbiased and direct temperatures of the global atmosphere. Instead, satellites measure the emission spectra of molecules in the atmosphere and then use various scientific formulas and statistical approximations to derive a figure for temperature. It's not like measuring someone's temperature with a common thermometer. Satellites need to use theoretical approximations to infer a particular temperature. This is another potential source of error in satellite datasets. For these and other reasons, climate scientists rely on a variety of additional sources when measuring surface and atmospheric temperatures, such as ships, buoys, weather stations, weather balloons, and so on. Christy's limited and arbitrary datasets seem to have been cherry-picked for the sole purpose of delivering his favorite outcome. Fourth, Christy does the comparison over a 40-year period: 1975 to 2015. That's a pathetically short amount of time by which to judge the adequacy of climate models, which is why climate scientists typically benchmark the models with paleodata, meaning data from very ancient climates. Fifth, averages obscure distributions. We simply don't know about the level of disagreement and variation among the climate models tracked in his chart because he just shows a simple 5-year moving average. There are many models in that group which no doubt accurately track the biased datasets he has collected. I'll pause right here, but it's worth noting that these are not the only major problems with the chart. 

Spencer and Christy are not the only contrarians on the right who have questioned the central results of climate science. The Canadian consultant Patrick Moore is another denialist who has made headlines in the media. In 2014, Moore issued a statement to the Senate Environment and Public Works Committee that questioned the scientific consensus about the causes of global warming, claiming the following: 

When modern life evolved over 500 million years ago, CO2 was more than 10 times higher than today, yet life flourished at this time. Then an Ice Age occurred 450 million years ago when CO2 was 10 times higher than today. There is some correlation, but little evidence, to support a direct causal relationship between CO2 and global temperature through the millennia. The fact that we had both higher temperatures and an ice age at a time when CO2 emissions were 10 times higher than they are today fundamentally contradicts the certainty that human-caused CO2 emissions are the main cause of global warming.

There are many factual errors and conceptual mistakes in this statement. First, Moore presupposes that carbon dioxide and greenhouse gases are somehow only allowed to affect the atmosphere in complete isolation from everything else. The reality is that many causal factors can converge in different ways to mediate the impact of greenhouse gases on global temperatures, such as volcanic or solar activity. For example, in the Ordovician period to which Moore is referring, the Earth received much less sunlight than it does today, perhaps about 4-5% less. That can obviously have a significant cooling effect on global temperatures, even in the presence of higher greenhouse gas concentrations in the atmosphere.

Second, Moore is completely wrong in his assertion that carbon dioxide levels 500 million years ago were more than 10 times what they are now, which was about 400 ppm in 2014 when he issued the statement. He evidently paid attention to some of the older studies on the issue but neglected more recent studies that had already been published when he gave this statement. Those older studies looked at carbon dioxide concentrations at intervals of 10 million years, which is not fine-tuned enough for many events, like certain critical periods in the Late Ordovician glaciation. The newer studies analyzed sharper time intervals and indicated that carbon dioxide levels were almost certainly much less than 8 times those of current levels after collecting and studying additional ancient rock samples throughout North America. Other studies have also shown that atmospheric carbon dioxide levels declined sharply leading up to the Late Ordovician glaciation, which is consistent with the notion that carbon dioxide is a greenhouse gas that can have a powerful impact on global temperatures.

Moore also claimed that greater warming would be beneficial for humanity:

We live in an unusually cold period in the history of life on earth and there is no reason to believe that a warmer climate would be anything but beneficial for humans and the majority of other species. There is ample reason to believe that a sharp cooling of the climate would bring about disastrous results for human civilization.

Moore is exhibiting an unfortunate yet common misunderstanding of what's going on with this misguided remark. The fundamental problem we're facing is not that global temperatures are rising or falling, but that they're changing at an extremely rapid rate that will end up destabilizing the biosphere and human civilization more broadly. If average global temperatures were to change by 2 degrees over 2 million years in a roughly smooth and linear fashion, those changes would have virtually no impact whatsoever on civilization. But changing global temperatures by 2 degrees or more in 200 years is an entirely different proposition altogether. The latter scenario, which is what we're actually facing, involves the sudden emergence of massive thermodynamic instability throughout the biosphere, radically reconfiguring energy flows and disrupting or destroying ecosystems around the planet. If this rate of warming continues or accelerates over the next century, the consequences for human civilization are likely to be completely disastrous.

Moore also questioned whether global warming had been accelerating at all and whether we should attribute our climate observations to human economic activity:

The increase in temperature between 1910-1940 was virtually identical to the increase between 1970-2000. Yet the IPCC does not attribute the increase from 1910-1940 to human influence. They are clear in their belief that human emissions impact only the increase since the mid-20th century.

This is a highly arbitrary and useless comparison because it only covers a part of the observed warming trend since the Industrial Revolution, not all of it. For example, the rate of warming from 1970 to 2000 was much faster than the warming rate from 1850-1900. Setting this point aside, the latest data makes it absolutely clear the rate of warming is increasing as time rolls forward. The 2023 Global Climate Report from NOAA states:

The year 2023 was the warmest year since global records began in 1850 at 1.18°C (2.12°F) above the 20th century average of 13.9°C (57.0°F). This value is 0.15°C (0.27°F) more than the previous record set in 2016. The 10 warmest years in the 174-year record have all occurred during the last decade (2014–2023). Of note, the year 2005, which was the first year to set a new global temperature record in the 21st century, is now the 12th-warmest year on record. The year 2010, which had surpassed 2005 at the time, now ranks as the 11th-warmest year on record.

Moore concludes by arguing that we are largely clueless and cannot predict what will happen next:

We do not know whether the present pause in temperature will remain for some time, or whether it will go up or down at some time in the near future. What we do know...is that the climate is always changing, between pauses, and that we are not capable, with our limited knowledge, of predicting which way it will go next.

This claim is patently false. There are still many improvements that need to be made to climate models, but it's downright silly to claim that we have no idea where the climate is going next. It's painfully obvious that it will continue to get warmer, but how much warmer exactly is still up in the air.

The Long Road Ahead

The records keep falling. 2024 was the hottest year on record, breaking the previous record from 2023. In general, climate models have correctly anticipated that warming would accelerate in the 21st century relative to the 20th, but the recent surge in global warming has raised some alarm bells that climate models might be underestimating what's to come. It's still not fully clear what's causing this recent warming burst, with several factors likely at work, including the removal of most sulfur dioxide from marine fuel after 2020, which reduced sulfur dioxide concentrations in the atmosphere and therefore reduced a key atmospheric compound that bounces incoming sunlight back into space.

Whatever the causes, there's a lot more work to be done, though recent political developments in the United States do not bode well for the short-term future of climate science. Especially at this particular time, it's critical that the public understands that our tax dollars for climate research are going to one of the most impressive, productive, and critical scientific projects of our time. There are a million things we need to change about our economic and political systems to ensure the long-term survival of global civilization, but to adequately plan for any potential future, we absolutely need reliable climate models that can inform actionable policy agendas.