About the Author

Linking Extreme Weather to Climate Change

By Matthew R. Francis

As the world’s climate changes, the warming atmosphere and oceans produce heavier rainfalls and more hurricanes, snowstorms, and other instances of extreme weather. Climate models predict the change in frequency of these events as a result of human-driven global warming. However, scientists and non-scientists alike are interested in whether climate change is responsible for specific weather events — such as Hurricane Maria, which devastated Puerto Rico in 2017.

“The kosher answer to this used to be that we can never say that climate change causes a specific event,” statistician Claudia Tebaldi of the University of Maryland’s Joint Global Change Research Institute said. “This has actually changed over time, because a few recent events were so extreme that the probability of observing them without climate change would have been practically zero.”

In other words, scientists and science communicators are growing increasingly confident about linking specific weather to global changes, a subfield of climate science and meteorology known as “event attribution.” Researchers calculate the probability of a particular event’s occurrence with or without climate change by considering a combination of factors, including human activity and variations that are independent of human contribution. Event attribution is a relatively recent discipline; scientists first used it to link climate change to the 2003 European heat wave [4], which killed thousands of people.

Communication to both the public and policymakers is a major part of event attribution. “The idea is to make this type of analysis—and the communication that ensues—a part of water-cooler conversation, the same sort of thing as weather forecasts,” Tebaldi, who formerly worked at the National Center for Atmospheric Research, said.

After all, people already informally attribute weather events to climate change, or disingenuously use specific weather events as ammunition for climate change denial. Since these conversations happen anyway, event attribution can steer them in more productive directions. For instance, hurricanes occurred long before human records existed, but some hurricanes only result from increased greenhouse gases produced by human activity. In short, certain weather events simply could not have transpired in human history before the rise of industrialization in Europe.

Going to Extremes

Because climate affects weather, changes in climate impact the type and severity of local weather. Higher temperatures drive evaporation, which can increase or suppress precipitation depending on specific conditions (for example, higher humidity prevents overnight cooling). Droughts therefore occur more frequently, and relatively more rain or snow falls during severe storms.

While these statements are well understood and uncontroversial, the challenge lies in linking individual weather events to climate and demonstrating that they could only occur—and/or be as severe—because of human-induced changes to the atmosphere. Researchers examine both human-driven and human-independent factors to assign a probability indicating that a particular weather event can be credited to climate, drawing on multiple climate models and local weather data.

Since this method is not fruitful for every thunderstorm or dry August, event attribution research focuses on extreme weather, which requires that one define “extremity” in a consistent way. Transparency when selecting which events to study is therefore essential, both to avoid any appearance of bias and make the process as automatic and replicable as possible.

The major criteria that define such extreme events include duration—storms localized in time simply have less available data—and potential or actual damage to people. As a result, not all extreme weather measured by rainfall or windspeed will trigger event attribution analysis. For instance, a hurricane that never makes landfall is not a good candidate for event attribution, though it can help improve statistics for determining storm frequency.

An additional complication arises when researchers are confident about linking increases in certain types of weather to climate change, but models for describing specific storms are less robust. This is especially true for events characterized by significant air circulation, such as hurricanes and typhoons. Researchers can link general surges in the frequency of these storms to climate, thanks to measurements of ocean surface temperatures and other factors; however, blaming a particular hurricane on global warming is difficult. Scientists still do not fully understand non-tropical cyclones such as nor’easters (essentially winter hurricanes at high latitudes), so event attribution can neither link them to climate change nor dismiss such a link (see Figure 1). Similarly, a connection between global warming and extreme tornado activity might exist, but models are not yet good enough to demonstrate that relationship.

Figure 1. A qualitative plot illustrating scientists’ confidence in generally connecting weather types with climate change (x-axis), and their confidence in specific event attribution (y-axis). Temperature-driven weather, such as warmer winters and hotter summers, carry the most confidence, while atmospheric circulation weathers have the lowest confidence levels. Figure courtesy of [3].

However, instances of extreme weather that are primarily temperature-driven are good candidates for event attribution, since the link between extreme temperatures and climate change essentially provides the basis for many models. For this reason, event attribution is most successful for heat waves, heat-associated droughts, and decreased cold spell durations.

Researchers have only recently begun to identify events with excessive rain or snowfall that can be credited to climate change. For instance, while current winter temperatures are generally higher than in the past, this is not true for heavy snowstorms, including snowfalls at very low temperatures, which are more common now than in the historical record. “In the case of precipitation, the system’s variability is so high that it was really hard to detect a new trend,” Tebaldi said. “But more and more, scientists are detecting precipitation changes that are consistent with an increase in warming climate.”

Event attribution is necessarily model-dependent because researchers have to estimate the probabilities of events occurring with and without the influence of climate change. The field suffers from many of the same difficulties that plague all climate research, namely that humanity is collectively running an uncontrolled and unreplicable experiment on Earth’s climate. Though we only have one time series of data (that is, history) to go on, this history guides the model-building process. For instance, researchers have used past data to successfully model hypothetical temperatures for various places without greenhouse gas emissions, and then compared these models to actual observations (see Figure 2) [3]. Also, while Earth represents only one “experiment” at the present time, researchers often employ multiple models and simulations to produce many different outcomes for comparison with reality.

Figure 2. Comparison of the probabilities of temperatures during a Russian heat wave in the 1960s versus the 2010s. This plot shows that climate change is clearly responsible for hotter summers; dangerous heat waves are some of the clearest cases for event attribution research. Figure courtesy of [3].

Scientists use these models—with appropriate uncertainties—to estimate the probability of a weather event with \((P_1)\) and without \((P_0)\) climate change. Event attribution then utilizes two related quantities to decide if an incident is attributable. The fraction of attributable risk (FAR) and risk ratio (RR) are defined as follows:

\[FAR = \frac{P_1-P_0}{P_1} \:\:\:\:\:\:\:RR = \frac{P_1}{P_0}.\]

These values are easily interpretable for the purpose of public discussion. For FAR values close to 1 and large RR values, we can confidently state that a given event could only occur because of climate change. Smaller values require more nuanced language.

The common approach obtains probabilities by solving the linear equation

\[ \boldsymbol{y = Xa + u}\]

for \(\boldsymbol{a}\), where \(\boldsymbol{y}\) is the vector containing observed quantities, \(\boldsymbol{u}\) is the effect of climate without human influence (typically represented by a Gaussian random vector), and \(\boldsymbol{X}\) is the matrix that represents all of the weather system’s estimated responses to climate change [2]. The vector \(\boldsymbol{a}\) contains the linear regression amplitudes that map each response onto the observed quantity in \(\boldsymbol{y}\). One can calculate the probability that a particular response \(f\)—represented by the column \(X_f\)—has an influence on the weather by using the probability that the corresponding amplitude \(a_f\) is positive [1-2].

Language and Accuracy Matter

Apart from the scientific problem, people often have trouble grasping the meaning of probabilities, which makes it challenging to communicate event attribution to policymakers and the general public. To mitigate this issue, World Weather Attribution—an international collaboration between universities and other organizations—rapidly analyzes weather events. Affiliated researchers use peer-reviewed methods to produce comprehensible reports on weather, in addition to more detailed analyses to inform policy.

Event attribution also helps clarify unclear terminology in common usage. For example, reports stating that a storm occurs “once every 100 years” (or those that make similar assertions) are often predicated on pre-climate-change assumptions. By comparing the real world to an imaginary world without global warming, event attribution provides alternative language for public communication.

While this approach is effective for Europe and North America, weather models and data are less robust for large parts of the world. Much of Asia, Africa, and the Pacific islands are both poorly modeled and extremely vulnerable to the effects of climate change. Event attribution research must take this into account before it can claim true global success.

However, the urgency of climate change demands global and non-parochial responses. Just as researchers developed event attribution to fill the need for clear analysis and communication, the next phase must encompass the whole planet. After all, we’re all in this together.

[1] Hannart, A., & Naveau, P. (2018). Probabilities of causation of climate changes. J. Clim., 31, 5507-5524.
[2] Hegerl, G., & Zwiers, F. (2011). Use of models in detection and attribution of climate change. Wiley Interdiscip. Rev.: Clim. Change, 2, 570-591.  
[3] National Academies of Sciences, Engineering, and Medicine. (2006). Attribution of Extreme Weather Events in the Context of Climate Change. Washington, D.C.: National Academies Press.
[4] Stott, P., Stone, D., & Allen, M. (2004). Human contribution to the European heatwave of 2003. Nature, 432, 610-614.

Matthew R. Francis is a physicist, science writer, public speaker, educator, and frequent wearer of jaunty hats. His website is BowlerHatScience.org.