From extreme heat waves, wildfires, and rainfall, we’ve seen a lot of this extreme weather over the past several years throughout the world. With 2023 already having a 50% chance of becoming the hottest year on record, and 2024 potentially being even warmer due largely to El Niño, these events will very likely only continue to impact most people globally.
Now forecasters and atmospheric modelers are attempting to understand and forecast these events with new methods and data in an attempt to keep pace with their seemingly never-ending increase in frequency.
For extreme heat waves, a concept called “ensemble boosting” essentially runs a typical set of ensemble model predictions in climate models. The use of ensemble modeling, which runs forecasts many times through the same model with only small changes in initial conditions that are nearly always within the margin of error for such conditions, has been around for some time in weather forecasting models. This allows forecasters to see a possible plausible range of outcomes for future weather, taking into account these margins of error for current observations. If the range of possibilities is small, the forecaster can be fairly certain about how the atmosphere may evolve; if the range is large, there is higher uncertainty in the forecast. Running these ensembles on extreme heat events in climate models gives an idea whether even more extreme heat events may be possible than what the climate models may predict, even though they wouldn’t give indication about when exactly those events would occur.
That said, a published June 2023 study by Risbey et al found that climate change was not necessarily outpacing our ability to forecast extreme heat waves. Although events such as the 2021 Pacific Northwest heat wave, examined in detail in their research, may indicate differently given the margin by which records were broken, they found through climate modeling simulations that the extreme temperature for any given location was affected by random weather patterns more than many may realize, even though the overall trend of global heating also does have its place. In other words, while climate change likely increased the change of extreme temperatures during the 2021 Pacific Northwest heat wave, the specific weather patten itself was also needed to shatter records by as much as they were broken.
Often, though not always, wildfire risk can increase with increasing temperatures. Michal Aibin, a visiting associate professor at Northeastern University in Vancouver, British Columbia, is working on predicting wildfires before they even occur. Essentially, this involves classifying forecasts based on their risk for wildfires, identify those most at risk, and direct wildfire prevention strategies based on the fire risk of individual forests. The goal of this is also to help reduce the spread of any fires once they begin, if prevention measures are already in place.
Once wildfires do occur, there is of course the risk of wildfire smoke and its effects on human health. Thus, forecasting air quality is important to many. Most air quality forecasts are made from either a machine learning (artificial intelligence, or AI) model or a chemical transport model. The machine learning model looks at past data and “learns” patterns, such as increased pollutants over industrial cities or freeways during rush hour, and predicts that those times are when air quality will be worse, but they don’t take into account the current atmospheric conditions or current known sources of pollution, such as wildfires. Chemical transport models do take these into account, but many of the variables involved, such as clouds, have a lot of uncertainty. Thus, it was found that using both of these types of models together results in more accurate air quality forecasts.
Extreme rainfall events constitute another type of extreme weather event that can be very difficult to forecast. These are often very localized events and while a general location and probability of extreme rainfall can be forecasted in advance, more exact locations and timing can often only be pinpointed once the event has began and when it may be too late for those in the area. A research team led by Dr. Assaf Hochman and doctoral student Tair Plotnik at Hebrew University have developed a mathematical tool, using an extensive database to examine all extreme rain events in the past 40 years or so in Israel. This tool can improve forecasting of extreme rainfall by looking in detail at the factors that led to these extreme rainfall events and subsequently the conditions that develop in advance of such events. They plan to begin working with forecasting operations to implement the use of this method of extreme rain forecasting.
For extreme rainfall events, Columbia University scientists have also developed an algorithm that they say allows for more accurate predictions of extreme weather. They implement cloud organization into climate models; that is, how close together clouds may be for example, which improves predictions of rain intensity. Most climate models have underestimated the variation in ways rain falls, leading to a favoring of light rain – which in turn can greatly underestimate how the probabilities or amounts of heavy rain may fall in the future with global climate change. Climate models, however, naturally have low resolution, so that they can forecast many years in advance across the globe. Thus, they naturally can’t account for cloud organization that exist on small scales. These scientists have developed this algorithm to help bridge that gap.
Overall, extreme weather events are notoriously challenging to forecast. The fact that they occur so rarely – or in the case of a warming world, some levels of extreme events have never occurred at all in record-keeping – makes it very difficult to know what contributes to their occurrence. In the case of events that have never occurred at all, it is of course impossible to look back through meteorological data to know what has caused it before. They are also the events that cause the most deaths, injuries, or other impacts to people and industries all over the world. Different methods, algorithms, and other tools must continue to be developed to try to better understand and forecast these events in the future.