Extreme Attribution Gets a National Academies Upgrade: But the Weakest Link Problem Remains

The National Academies of Sciences, Engineering, and Medicine has released a new report, Attribution of Extreme Weather and Climate Events and Their Impacts, accompanied by a press release announcing that the science of extreme-event attribution has “advanced considerably” and can now make increasingly confident statements about the influence of human-caused climate change on individual weather events.

That press release is the sales pitch, big on hype, low on science.

The report itself is more cautious. Buried beneath the optimism are acknowledgments of model limitations, sparse observational records, unresolved methodological questions, and an impact-attribution field that remains in its infancy. The press release tells readers the machinery is getting better. The report quietly lists the loose bolts.

There is also an awkward matter of timing. The report appears just as the climate-scenario community is backing away from RCP8.5 and SSP5-8.5, the high-end emissions scenarios that underpinned years of alarming projections and impact studies. Roger Pielke Jr. and his colleagues have argued that these scenarios were poor representations of likely future emissions and should not have been treated as business-as-usual futures.

That does not invalidate every attribution study. But it should encourage a more cautious reading of claims that extend today’s attribution results into tomorrow’s damages, adaptation costs, or litigation.

A Chain of Assumptions

Extreme-event attribution is often described as though it were a single scientific calculation. It is nothing of the sort.

Researchers begin with an observed weather event and then work through a long sequence of choices involving observations, climate models, statistical methods, hypothetical “counterfactual” worlds, and interpretation.

Figure 1. Extreme-event attribution is a chain of observational, modeling, statistical, and interpretive steps. Each stage introduces assumptions and uncertainty that can compound before the final result is communicated.

Every box in this figure represents a scientifically defensible choice. Different choices can produce different answers. That is not misconduct. It is an unavoidable consequence of estimating probabilities in a hypothetical climate that has never existed.

The problem arises when a result produced by this chain is presented as though it were a direct measurement rather than a model-dependent estimate.

Five Reasons for Caution

Rather than examining every link individually, five broad issues deserve particular attention.

1. The Counterfactual Climate Cannot Be Observed

Every attribution study compares today’s weather with a world in which human greenhouse-gas emissions never occurred. That world does not exist.

Researchers must construct it using models, statistical adjustments, or natural-forcing simulations. Different methods produce different counterfactuals, which can change the estimated influence of climate change on the event.

The “no-human-influence” climate is therefore an assumption, not an observation.

2. Climate Models Still Have Important Blind Spots

The National Academies acknowledges that confidence varies substantially depending on the type of event. Confidence is highest for large-scale temperature extremes but considerably lower for severe thunderstorms, tornadoes, and other phenomena driven by processes that global climate models do not adequately resolve.

That admission deserves more prominence than it receives in the press release. A model can reproduce long-term temperature trends while still failing to simulate the circulation patterns, convective processes, or land-atmosphere interactions that produced a specific storm.

Higher-resolution models may improve this in the future. We’ll see.

3. Statistical Choices Matter

Extreme events are, by definition, rare. Researchers frequently estimate probabilities well beyond the length of the observational record using statistical extrapolation and model ensembles.

The mathematics may be sophisticated, but the results remain sensitive to assumptions about event definition, distribution choice, and model behavior. Running thousands of simulations reduces sampling uncertainty. It does not eliminate structural model error.

As the old engineering saying goes, a miscalibrated scale does not become accurate simply because you weigh yourself ten thousand times.

4. Impact Attribution Adds Another Layer of Assumptions

The report also highlights the emerging field of extreme-event impact attribution, which attempts to estimate how much of an event’s economic losses or health impacts are attributable to climate change.

That requires assumptions extending far beyond meteorology. Economic losses depend on wealth, population growth, infrastructure, building practices, emergency response, land use, and adaptation—not simply weather.

Each additional assumption increases uncertainty. Yet these studies are particularly attractive for litigation and policy because they assign dollar values and mortality estimates to climate change.

5. Communication Often Overstates Confidence

Perhaps the most striking contrast is between the report itself and the accompanying press release.

The report repeatedly emphasizes differing confidence levels, unresolved methodological challenges, limited observational records, and the need for further validation. The press release emphasizes scientific progress, operational usefulness, and growing confidence.

Both statements are technically accurate. One simply places much greater emphasis on the caveats.

The Shadow of RCP8.5 Still Looms Large

The report also arrives during a broader reassessment of climate scenarios.

For years, RCP8.5 was widely used as a reference future despite repeated criticism that it represented an implausibly high emissions pathway. Roger Pielke Jr. has documented how the scientific community has now largely moved away from treating RCP8.5 and SSP5-8.5 as realistic baseline futures.

Historical attribution studies do not automatically become invalid because of that change.

But projections of future extremes, damages, and adaptation costs often depend directly on the assumed emissions pathway. If the scenario changes, the projected risks change as well. That deserves more attention than the National Academies report gives it.

The Bottom Line

The National Academies deserves credit for candidly documenting many of the limitations of extreme-event attribution. The report acknowledges that confidence varies by event type, that important model limitations remain, and that impact attribution is still an emerging discipline. The problem is not the report as much as its presentation.

The press release leaves readers with the impression that attribution science has largely solved the methodological challenges standing between weather events and confident claims about climate change.

The report itself says something more nuanced.

Extreme-event attribution is a valuable research tool. But it is also a long chain of assumptions, models, statistical choices, and interpretive judgments. As Figure 1 illustrates, uncertainty enters at every step. That uncertainty does not disappear simply because the final answer is expressed with decimal places or confidence intervals.

Readers should remember that the next time they see headlines claiming that climate change “caused” a particular hurricane, flood, or heat wave.