Policymakers usually depend on coarse-resolution global climate models to assess a community’s risk of extreme weather. By looking decades and even centuries into the future, these models can predict large-scale weather patterns but struggle to provide specific data for smaller locations. To estimate the risk of an area such as Boston experiencing extreme weather events like floods, policymakers have to combine the broad-scale predictions from these models with finer-resolution models tailored to predict localised weather patterns. However, these estimates are only as precise as the larger, coarser-model predictions.
To improve the accuracy of these climate models, a team of scientists from MIT’s Department of Mechanical Engineering, led by Professor Themistoklis Sapsis, developed a method to “correct” the models’ predictions using a combination of machine learning and dynamical systems theory. According to Sapsis, their technique “nudges” simulations from climate models toward more realistic patterns. When used in combination with smaller-scale models, the method produces more accurate predictions regarding the frequency of extreme weather events in specific locations. The correction scheme can be applied to any global climate model and can help policymakers determine where and how often extreme weather will hit as global temperatures continue to rise.
The approach doesn’t aim to rectify the underlying dynamical equations in climate models—interactions between atmospheric and oceanic events— which has been the norm. Instead, it seeks to correct the output or results from the models. It overlays an algorithm that nudges the model’s simulations closer to real-world conditions based on machine-learning.
As a first trial, the algorithm corrected simulations from the Energy Exascale Earth System Model (E3SM), a climate model run by the U.S. Department of Energy. Using eight years of historical data, the machine-learning algorithm corrected the simulation from E3SM to produce realistic climate patterns that mirrored real-world observations from the past 36 years.
By pairing the large-scale model with a specific model for tropical cyclones, the researchers found that the method could accurately reproduce the frequency of extreme storms in select locations worldwide. Even slight adjustments in predicting temperature, such as the difference between 105 degrees Fahrenheit and 115 degrees in the uncorrected simulation, can significantly affect the lived experiences of humans in such climates.
Sapsis says that their approach leads to a much-improved coarse model and it could be beneficial for understanding the impacts of climate change on different natural disasters, such as forest fires, heatwaves, and flooding. The next phase of their research involves analysing future climate scenarios. The Defense Advanced Research Projects Agency in the U.S partially supported the research.
