Climate scientists have a surprising habit: They often underplay the climate threat. In 2007 a team led by Stefan Rahmstorf compared actual observations with projections made by theoretical models for three key climate variables: atmospheric carbon dioxide, global average temperature and sea-level rise. While the projections got CO2 levels right, they were low for real temperature and sea-level rise. In 2008 Roger Pielke, Jr., found that sea-level rise was greater than forecast in two of three prior Intergovernmental Panel on Climate Change reports. In 2009 a review of hundreds of papers on climate change identified several areas where scientists had lowballed event predictions but none in which they had overestimated them.

In 2013 researcher Keynyn Brysse, then at the University of California, San Diego, along with other colleagues and me, pointed out that these underestimates represent a kind of bias. Scientists tended toward lower projections because they did not want to be accused of making dramatic and exaggerated claims. The articles reporting the underestimates have been widely cited, so one might think that by now scientists would have taken corrective steps.

But recent studies of Arctic warming suggest that the problem may not have gone away. For instance, scientists have long known about how Arctic ice reflects sunlight, redirecting heat away from the planet. But as polar ice melts because of global warming, the Arctic Ocean absorbs more heat, which causes the Arctic to warm even more, which melts more ice, and so on. It should surprise no one, then, that the area is warming fast. Yet scientists have been caught off-guard by just how fast the region is heating up.

A recent study led by Mika Rantanen of the Finnish Meteorological Institute found that, since 1979, the Arctic has warmed nearly four times faster than Earth as a whole. Few climate models have predicted an effect this large.

Model results are typically reported as the averages of many runs of a set of similar models, referred to as ensembles. These new Arctic temperature observations are not only warmer than all the major ensemble averages but, in some cases, outside the whole ensemble envelope. In one very large and highly respected one—the Max Planck Institute Grand Ensemble—the observed warming for 1979–2021 is entirely beyond the results. Some real-world observations are hotter than even the hottest projections.

This has several implications. First, it reminds us that averages can be misleading. Extreme outcomes may be unlikely but do occur and are crucial in assessing risk. Second, it suggests that climate models may be continuing to underestimate key climate effects.

Admittedly, the observations might be wrong; measuring the temperature of the region is notoriously difficult, in part because of sparse sensor coverage over the Arctic Ocean. In addition, scientists may have analyzed different time periods or used conflicting definitions of “Arctic” boundaries. But it may also be that their subconscious bias toward playing things down was playing itself up.

We’ve heard a lot in recent years about subconscious bias in relation to race and gender discrimination. But subconscious bias can be caused by many things, including defensiveness. Even now scientists continue to be accused of exaggerating climate risks by prominent figures who get outsized media attention. Scientists who have internalized this concern may be subconsciously biasing their models to be unrealistically conservative.

If scientists have underestimated Arctic warming, they have likely minimized amounts of permafrost melting and methane release as well. And that could be truly dire because the permafrost holds about 1.5 billion metric tons of organic carbon, twice as much as now in the atmosphere. Were that carbon to be rapidly released, it could cause a worst-case scenario: a runaway greenhouse effect. Whatever the cause, it’s time that scientists looked seriously at whether their models continue to underplay critical aspects of the climate problem. Low estimates can create the false impression that we have more time to fix the problem than we actually do.