To evaluate a community’s risk of utmost weather, policymakers rely first on global climate models that may be run many years, and even centuries, forward in time, but only at a rough resolution. These models is perhaps used to gauge, as an example, future climate conditions for the northeastern U.S., but not specifically for Boston.
To estimate Boston’s future risk of utmost weather reminiscent of flooding, policymakers can mix a rough model’s large-scale predictions with a finer-resolution model, tuned to estimate how often Boston is prone to experience damaging floods because the climate warms. But this risk evaluation is barely as accurate because the predictions from that first, coarser climate model.
“For those who get those fallacious for large-scale environments, then you definitely miss the whole lot when it comes to what extreme events will appear to be at smaller scales, reminiscent of over individual cities,” says Themistoklis Sapsis, the William I. Koch Professor and director of the Center for Ocean Engineering in MIT’s Department of Mechanical Engineering.
Sapsis and his colleagues have now developed a technique to “correct” the predictions from coarse climate models. By combining machine learning with dynamical systems theory, the team’s approach “nudges” a climate model’s simulations into more realistic patterns over large scales. When paired with smaller-scale models to predict specific weather events reminiscent of tropical cyclones or floods, the team’s approach produced more accurate predictions for the way often specific locations will experience those events over the subsequent few many years, in comparison with predictions made without the correction scheme.
Sapsis says the brand new correction scheme is general in form and may be applied to any global climate model. Once corrected, the models may help to find out where and the way often extreme weather will strike as global temperatures rise over the approaching years.
“Climate change could have an effect on every aspect of human life, and each variety of life on the planet, from biodiversity to food security to the economy,” Sapsis says. “If we’ve capabilities to know accurately how extreme weather will change, especially over specific locations, it could make a number of difference when it comes to preparation and doing the fitting engineering to provide you with solutions. That is the strategy that may open the technique to do this.”
The team’s results appear today within the . The study’s MIT co-authors include postdoc Benedikt Barthel Sorensen and Alexis-TzianniCharalampopoulos SM ’19, PhD ’23, with Shixuan Zhang, Bryce Harrop, and Ruby Leung of the Pacific Northwest National Laboratory in Washington state.
Over the hood
Today’s large-scale climate models simulate weather features reminiscent of the common temperature, humidity, and precipitation world wide, on a grid-by-grid basis. Running simulations of those models takes enormous computing power, and with a purpose to simulate how weather features will interact and evolve over periods of many years or longer, models average out features every 100 kilometers or so.
“It’s a really heavy computation requiring supercomputers,” Sapsis notes. “But these models still don’t resolve very necessary processes like clouds or storms, which occur over smaller scales of a kilometer or less.”
To enhance the resolution of those coarse climate models, scientists typically have gone under the hood to try to fix a model’s underlying dynamical equations, which describe how phenomena within the atmosphere and oceans should physically interact.
“People have tried to dissect into climate model codes which were developed over the past 20 to 30 years, which is a nightmare, because you may lose a number of stability in your simulation,” Sapsis explains. “What we’re doing is a totally different approach, in that we’re not attempting to correct the equations but as a substitute correct the model’s output.”
The team’s recent approach takes a model’s output, or simulation, and overlays an algorithm that nudges the simulation toward something that more closely represents real-world conditions. The algorithm relies on a machine-learning scheme that takes in data, reminiscent of past information for temperature and humidity world wide, and learns associations throughout the data that represent fundamental dynamics amongst weather features. The algorithm then uses these learned associations to correct a model’s predictions.
“What we’re doing is attempting to correct dynamics, as in how an extreme weather feature, reminiscent of the windspeeds during a Hurricane Sandy event, will appear to be within the coarse model, versus in point of fact,” Sapsis says. “The tactic learns dynamics, and dynamics are universal. Having the right dynamics eventually results in correct statistics, for instance, frequency of rare extreme events.”
Climate correction
As a primary test of their recent approach, the team used the machine-learning scheme to correct simulations produced by the Energy Exascale Earth System Model (E3SM), a climate model run by the U.S. Department of Energy, that simulates climate patterns world wide at a resolution of 110 kilometers. The researchers used eight years of past data for temperature, humidity, and wind speed to coach their recent algorithm, which learned dynamical associations between the measured weather features and the E3SM model. They then ran the climate model forward in time for about 36 years and applied the trained algorithm to the model’s simulations. They found that the corrected version produced climate patterns that more closely matched real-world observations from the last 36 years, not used for training.
“We’re not talking about huge differences in absolute terms,” Sapsis says. “An extreme event within the uncorrected simulation is perhaps 105 degrees Fahrenheit, versus 115 degrees with our corrections. But for humans experiencing this, that could be a big difference.”
When the team then paired the corrected coarse model with a selected, finer-resolution model of tropical cyclones, they found the approach accurately reproduced the frequency of utmost storms in specific locations world wide.
“We now have a rough model that may get you the fitting frequency of events, for the current climate. It’s rather more improved,” Sapsis says. “Once we correct the dynamics, this can be a relevant correction, even when you could have a unique average global temperature, and it could be used for understanding how forest fires, flooding events, and warmth waves will look in a future climate. Our ongoing work is specializing in analyzing future climate scenarios.”
“The outcomes are particularly impressive as the strategy shows promising results on E3SM, a state-of-the-art climate model,” says Pedram Hassanzadeh, an associate professor who leads the Climate Extremes Theory and Data group on the University of Chicago and was not involved with the study. “It will be interesting to see what climate change projections this framework yields once future greenhouse-gas emission scenarios are incorporated.”
This work was supported, partially, by the U.S. Defense Advanced Research Projects Agency.