What if our Climate Models Are Flawed?

What if our Climate Models Are Flawed?

New research from Deep Sky uses a novel approach for predicting the impact of climate change and finds current climate models could be underestimating the risks we face. 

Why climate models matter

International organizations like the UN and the IPCC use climate models to predict the impact of climate change. The IPCC takes averages of hundreds of climate models to produce its official projections and these are disseminated to governments, the press, and the general public. When governments make plans for how quickly they need to cut emissions, or how they should prepare for increasingly severe storms, they use these projections to understand what is to come. 

Humanity’s understanding of the risks posed by climate change is primarily based on climate models. These models do an impressive job simplifying our entire natural world into computer programs that can make predictions. But as the saying goes: All models are wrong, some are useful. What if our climate models aren’t as useful as we thought? And what if they are wrong in ways that are dangerous for humanity? 

Why the current models are flawed

A scientist acting with caution will suppress their worst fears and wait for more evidence before publishing anything. Risk experts, like actuaries in the insurance industry, are cautious in the opposite way. They start by considering a plausible and serious risk and then they work backwards. What are the worst case scenarios? How likely are they? And what can we do to avoid them?

Scientists’ approach of starting from what can be proven and hypothesis testing has led to a profound misunderstanding of the climate risks we face. The climate crisis is rapidly spiraling out of control and climate projections have not adequately prepared us for what is to come. We need rigorous analysis of the specific risks we face, we need to borrow from industries with expertise in assessing and managing risk, and we need to be clear-eyed about the consequences of our greenhouse gas emissions. 

The scientific method has been a foundation of humanity’s scientific progress. It is essential when society is trying to determine if, say, a phenomenon exists or not. But we are past that point with climate change. We are no longer debating its existence - we are dealing with an existential threat and for that reason we need more risk assessment. 

What the models don’t know

Climate models have an enormous amount of uncertainty baked into them. Take climate sensitivity, for example. This is a measure of how much global warming we should expect for a given ton of emissions. It is a fundamental piece of the climate change puzzle. Yet scientists have astonishing uncertainty about climate sensitivity. 

The plot below shows scientists’ estimates of a common measurement in the scientific community called equilibrium climate sensitivity (ECS) over the past 20 years. The range of uncertainty (shaded red below) has narrowed only slightly over that time period and remains wide enough that every prediction should come with a warning: if ECS is near the top end of the IPCC’s “very likely” range, things will be much worse than predicted.

Data collected with help from Knutti et al. 2017 and Zeke Hausfather. 

In part, this is because measuring ECS relies on climate models that themselves have a huge amount of uncertainty. In fact, much of the research represented in the plot acknowledges that ECS estimates do not properly account for tipping points, and there is evidence from paleontology that ECS changes as temperatures change, increasing at warmer temperatures. If we properly account for feedback loops and critical non-linear relationship between emissions and global temperatures, and somehow manage to accurately account for tipping points, the climate sensitivity estimates pictured above would be far higher. While slowly improving, our knowledge of climate sensitivity is still woefully limited which means our understanding of what’s coming is grossly insufficient. These climate models are highly complex undertakings - they require huge quantities of data, significant computing power, and complex physics-based underpinnings. After all, they are simulations of our entire environment: ocean, atmosphere, rainfall, etc. We should understand some basic facts about these climate models before putting too much faith in their predictions. 

For instance, we are still unable to accurately model cloud systems. The IPCC updates its projections every so often via the Coupled Model Intercomparison Projects (CMIP). The latest update, CMIP 6, made some significant changes to previous projections. These changes were almost exclusively in one direction - that things are worse than we thought. Many scientists attributed the changes to a previous flawed understanding of cloud formations. But experts studying how to model cloud formations say we still lack basic understanding.

This may seem like a minor point. How important can clouds be? Very. The entire greenhouse effect that causes climate change is directly impacted by clouds. The cloud layer radiates solar radiation back to space and also traps some of what makes it through our atmosphere. It has both large cooling and warming effects. We cannot make confident climate projections without a strong understanding of cloud systems. 

Another example is what is called ‘methane hydrates’ in the permafrost and seafloor. As ice melts, methane will be released into the atmosphere. Methane is a much more potent greenhouse gas than CO2. We know that this will happen, and that it will have a significant impact on climate change, but we have very little understanding of where the methane is and what the rate of emission will be. For these reasons it is modeled very poorly, if at all. Some models do not include methane hydrates in projections because of uncertainty.

This raises a concerning fact about these climate models on which the international community has based its understanding of the future of our planet. High levels of uncertainty have not stopped us from making predictions. Averages of hundreds of highly uncertain predictions do not address these shortcomings because the models tend to be limited in the same ways. They borrow from each other. So when our understanding is biased, it tends to be systematically biased. 

You don’t have to look far to see the impact of this uncertainty. Consider just how wrong scientists’ predictions were about temperatures last year. Luckily, scientists at the highest levels are starting to publicly acknowledge the gap. For example, the director of NASA’s Goddard Institute for Space Studies wrote: “the 2023 temperature anomaly has come out of the blue, revealing an unprecedented knowledge gap” and that “we need answers for why 2023 turned out to be the warmest year in possibly the past 100,000 years. And we need them quickly.” In fact, one climate scientist referred to September 2023’s sky-high temperatures as “absolutely gobsmackingly bananas.”

A mantra of risk experts in industries like insurance is to keep models simple when uncertainty is high. Complex models used in situations of high uncertainty will tell you more about the models’ assumptions than the real world they are meant to represent. Today’s climate models violate this rule on both counts – climate modeling is extremely complex and uncertainty is enormous. Our confidence in their predictions should keep this in mind.

What we don’t predict at all

The numerous sources of uncertainty mean we should be wary of putting too much trust in existing climate predictions, but what is perhaps more concerning is what we have no predictions about at all. As previously mentioned, science doesn’t start with the risks; it starts with what it can prove. The scientific community’s insistence on consensus aggravates this issue. In his book Five Times Faster, Simon Sharpe gives an alarming example. In the 2014 IPCC report, one research paper uncovered distressing details about humans’ inability to survive extreme temperatures, and the increasing likelihood of reaching these temperatures as we continue to emit. Such temperatures and levels of humidity would kill people even if they are “out of the sun, in gale-force winds, doused with water, wearing no clothing, and not working.”

The findings were left out of the most important part of the report, the “Summary for Policymakers,” because of a convention that “information could only be included if it was supported by the findings of at least two independent pieces of research.” This meant that policy makers tasked with guiding society through this impending crisis read research from “nine research papers that had considered the impacts of climate change on skiing resorts, and thirteen research papers that had investigated the important topic of climate change risks to grape-growing in Europe” but none that addressed the risks to human life from extreme heat. 

That is not to say the IPCC has never reported on the risks of extreme heat to human health. The AR6 report refers to a wealth of research on the topic in Chapter 7. The concern is whether the findings are being communicated early and prominently enough for us to take appropriate action. 

What do insurance companies know that we don’t?

Climate models have a poor track record of predicting exactly what is most important: what could go really wrong and how to prepare. Global mean temperature does not alone kill people. What kills people is severe drought, wildfires, and hot bulb temperatures. When research is geared towards these kinds of risks, the results are much more alarming. 

Risk experts who have started to analyze climate data themselves have been shocked at the level to which we are understating risks. Our predictions of the financial impacts of climate change are wildly optimistic, and the IPCC’s “overshoot pathways” – where we temporarily overshoot 1.5°C before coming back down again – ignore the overwhelming evidence of “point of no return” tipping points. 

Look at insurance companies’ behavior. Thirty-one states saw double-digit home insurance rate increases in 2022 and major insurance companies like State Farm and Allstate have stopped issuing new policies in California entirely because of the increased risk of wildfires. Expansive restrictions have been enacted in states like Florida and Louisiana as well. The trend is clear. Homes in larger and larger areas in the US are becoming uninsurable. These companies have decades of experience assessing and pricing risk. Their survival relies on accurately predicting risk. In truth, so does ours, but we haven’t started acting like it yet. 

What the new approach tells us

Studying climate risk in the era of climate change requires a new approach. 

Risk assessment can help us predict disasters ahead of time. Scientists use statistical methods that draw from large sample sizes to assess probability. One problem with predicting extreme, rare events like storms or droughts–that are becoming more common and more severe–is that we don’t have past examples to base our analysis on. We have no precedent. So, we need to be able to predict events we have never seen in the past. The lack of precedent means these events are deeply under-researched. Again, the issue is with picking research topics based on what can be proven, rather than what’s important.

Researchers applying a new modeling technique called UNSEEN modeling found that they could predict extreme events that were surprises to everyone else. This technique pools predictions from numerical weather prediction systems to help expand our understanding of what’s possible. When these methods are applied, we find rapidly increasing risks which we have been ignoring.

Deep Sky Research used this new modeling technique to assess the risk of a severe heatwave that would have devastating consequences on American agriculture and food supply. We found that what used to be a once in 100-year heat wave, will now happen every five years. In other words, our greenhouse gas emissions have caused the risk to shoot up 20x. 

New modeling techniques can help us assess the risk of climate disaster in ways conventional climate science has failed thus far. The more of this kind of analysis we do, the more it becomes clear just how precarious our situation is. We need to cut emissions immediately and rapidly scale up permanent carbon removal in order to avert disaster. The risks are too great – we have no other choice.