We can no longer rely on historical data to predict extreme weather

Climate change is dramatically shifting how we look at the future.

Floods and other dangerous weather extremes are only getting more intense and more frequent as our climate warms. Historically, we’ve always been able to predict these extremes by looking at how often they occurred in the past. But a new study published Wednesday in Science Advances reveals just how many of those forecasts actually fall short. In just a decade, the findings suggest, the climate has shifted so drastically that the frequency of past extreme events is no longer a reliable predictor.

These predictions help us draw floodplain maps and design infrastructure so that it can withstand even intense events. But if our predictions are wrong, that means we can no longer plan new housing, roads, and bridges based on the storms of the past. Increasing extremes—such as tropical cyclones, heat waves, and heavy storms—will force us to change our plans and design structures that can endure these changes.

It’s hard to understand the influence human-caused warming has on extreme events. The atmosphere is chaotic by nature, and record-setting extremes are by definition rare, giving scientists few data points by which to understand them. Noah Diffenbaugh, an earth system scientist at Stanford University, and a team of climate scientists incorporated a record of extreme hot, wet, and dry weather events from 1961 to 2005 into a climate prediction framework. That framework incorporates both historical event-based predictions and climate models, which incorporate projected future warming into their estimations.

In the following decade, however, humans continued to burn fossil fuels and record-setting weather events hammered regions worldwide. Seven of the 10 hottest years on record hit between 2006 and 2017, and huge storms like Hurricane Harvey in 2017 caused greater destruction than ever before.

Given the extra warming, Diffenbaugh wanted to test how well historic data could predict recent extreme events. He used the data between 1961 to 2005 to develop probabilities of hot, wet, and dry extremes across the Northern Hemisphere between 2006 and 2017. Separately, Diffenbaugh also used climate models to compare real-world record heat waves, storms, and droughts against their frequency in the past, as well as to climate model-based projections.

Diffenbaugh’s predictions based on historical data fared poorly for that decade. The results underestimated extreme events, especially the hot and wet ones. Compared to the projection based on historical data, actual extreme hot days increased by at least 50 percent in Europe and East Asia. And observed wet extremes were also 50 percent more frequent in the United States and Europe relative to the historically-based prediction. “I was very surprised,” says Diffenbaugh. “I had a sinking feeling that the framework my group had been developing over the last several years had some flaws.”

But that wasn’t necessarily the problem. The framework does fine when predicting extreme events that occured in the latter part of the 20th century. But in this most recent decade, the extra warming we’ve generated is so significant that extreme weather patterns are diverging from those of the past. Meanwhile, the climate models Diffenbaugh tested were able to accurately predict the frequency of record-setting events between 2006 and 2017. “The climate models for the near-term future encompass what actually happened,” says Diffenbaugh. “Even though they were future predictions at the time.”

“The paper by Noah Diffenbaugh is innovative and combines both models and observations to demonstrate that the odds of extremes are rapidly changing due to global warming,” says Erich Fischer, a climate scientist at ETH Zurich, who was not involved in the study. “The paper has implications for risk management.”

Diffenbaugh’s findings have huge implications for designing new infrastructure for climate change and updating existing structures. It seems we now can’t estimate the probability of a 500-year flood event simply based on past flooding, which is how we’ve tended to do our hazard planning in the past. These findings show we need to use a combination of historic information and climate modeling to design for the future under climate change.

That’s because while climate models are good at predicting changes across large regions, they’re not going to tell an urban planner how often a particular river running through their city might flood. Those local events are simply too hard to predict right now. Some states are already trying to cope. In California, the 2016 Climate-Safe Instructure bill was enacted to develop a planning process for new roads, bridges, and other structures so that they can weather a changing climate. Diffenbaugh is part of the working group established by the bill. He says we need to use the results of climate models together with local historic data to better prepare for future extremes. Adds Fischer, “The climate and thereby the odds of an extreme to occur are different today than they used to be 10 or 20 years ago.” To prepare, we need to use both past observations and future predictions to quite literally keep our cities above water.

Ula Chrobak

Ula Chrobakis a freelance science writer and editor. You can check out more of her work at her website.