Studies find climate models likely underestimate future extreme heat

Recent heatwaves have not just broken previous temperature records, they have shattered them by wide margins. The 2003 European heatwave and the 2010 Russian heatwave both saw week-long temperature maxima exceed previous records by more than two standard deviations, a statistical measure of how unusual an event is compared to typical variation, and both caused tens of thousands of heat-related deaths (Source 1).

A study using large ensembles of climate simulations, sets of many runs of the same climate model started from slightly different conditions, projects that events breaking records by three or more standard deviations will become far more common. In high-emission scenarios, such record-shattering week-long heat extremes are projected to be 2 to 7 times more likely in 2021 to 2050, and 3 to 21 times more likely in 2051 to 2080, compared with the past three decades. By the later period, the study estimates such events would occur roughly every 6 to 37 years somewhere across the northern mid-latitudes (Source 1).

In one illustrative simulation, a hottest week of temperatures over Central North America exceeded the previous 175-year record in that simulation by more than five standard deviations. What would have been considered a 1-in-1000-year event for the period 1951 to 2019 becomes roughly a 1-in-100-year event around 2020, and a 1-in-40-year event by the mid-2020s, once the model's warming trend is taken into account (Source 1).

A separate analysis examined how well 19 climate models, compared against two reanalysis datasets (statistical reconstructions of past weather built from observations and models, called ERA5 and MERRA-2), reproduce a specific measure: the gap between a location's hottest day in a month and that month's average temperature. This gap, called the monthly high temperature anomaly, captures how far extreme heat departs from typical conditions, separate from the overall warming trend (Source 2).

The models tend to underestimate this gap. Across 552 combinations of region and month, models underestimated the 22-year average version of this anomaly in 64% of cases, and underestimated the 22-year maximum, a proxy for rarer, more extreme events, in about 87% of cases. This bias appeared consistently across two separate multi-decade periods between 1980 and 2023, suggesting it is a persistent feature of the models rather than a one-off error (Source 2).

The study also found that projected future extremes vary sharply by region. Higher-latitude regions, especially in winter, are projected to see this heat anomaly shrink over time, a pattern the researchers link to loss of sea ice and snow cover and other changes tied to a warming Arctic. Lower and middle-latitude regions, including South Africa, Australia, and South America, are instead projected to see the anomaly grow, potentially due to drier surface conditions under warming (Source 2).

Both studies point to the same underlying problem from different angles. The large-ensemble study shows that record-shattering events are physically plausible in current climate models and become sharply more likely as warming accelerates, driven by a combination of extreme anomalies in atmospheric circulation, soil moisture, and incoming sunlight that are individually unremarkable but rare in combination (Source 1). The bias study shows that the models used to make such projections may themselves be starting from an underestimate of how extreme monthly heat anomalies really get, since they undercount the size of the gap between hot extremes and the surrounding average in most regions studied (Source 2).

The large-ensemble study frames its results in terms of warming rate rather than the overall global warming level, meaning that how fast the climate warms, not just how much it warms in total, is what drives the odds of a record-shattering event (Source 1). Neither study claims a single mechanism explains all cases; the first attributes its example event to a combination of ordinary drivers reaching extreme values together, not to a distinct process unique to record-shattering extremes (Source 1).

NASA's overview of extreme weather states that record-breaking heat waves, droughts, floods, and wildfires are becoming more frequent and more intense as greenhouse gases trap more heat, citing the Intergovernmental Panel on Climate Change's Sixth Assessment Report as attributing this shift to human-caused emissions (Source 3). A 2021 overview of climate impacts noted that a heat dome that broke previous temperature records in the US Pacific Northwest and Canada's west coast, killing hundreds of people, would have been virtually impossible without human-caused warming, according to attribution scientists (Source 4).

The authors of the large-ensemble study caution that record-shattering events of the size they simulate have not been observed, since the observational record is too short to confirm or rule them out; they describe their results as showing such events cannot be ruled out as implausible, rather than proving they will occur (Source 1). The bias study notes that a model's ability to match past climate says only weakly how much future change it will predict, because errors in different parts of a model's physics can offset each other and hide problems that only appear in extreme projections (Source 2). Both studies point toward the same practical concern: infrastructure and heatwave preparedness plans built around historical records, or around current model projections, may be designed for a narrower range of extremes than what a warming climate can actually produce (Source 1, Source 2).