To better predict the risks of extreme weather events due to climate change, scientists at MIT have developed a method that refines the predictions from large, coarse climate models. The key to this approach is leveraging machine learning and dynamical systems theory to make the climate models’ large-scale simulations more realistic. By correcting the climate model’s predictions, the researchers can obtain more accurate forecasts of how often specific regions will experience extreme weather events over the coming decades.
The new correction method is universal and can be applied to any global climate model. Once the models are corrected, they can provide valuable information about where and how frequently extreme weather events, such as floods or tropical cyclones, will occur as global temperatures increase. This in-depth knowledge can help with preparations and creating engineering solutions to mitigate the impact of these events.
Today’s coarse climate models simulate global weather parameters and require immense computing power. However, these models still struggle to capture fine-scale phenomena such as storms and clouds. Traditionally, scientists have attempted to refine these models by adjusting the underlying dynamical equations. In contrast, the new approach by the MIT team does not aim to adjust the equations but to amend the model’s output.
Utilizing a machine learning-based algorithm, the scientists take the model’s simulation and gently modify it to create a result that aligns more closely with real-world observations. The algorithm learns associations within the data, such as the relationship between temperature and humidity, and corrects the model’s forecasts based on these learned dynamics.
Preliminary tests of this method used the Energy Exascale Earth System Model (E3SM) simulations, and they found that the corrected version produced climate patterns that more accurately matched real-world data from the past 36 years. In addition, the corrected model, when combined with a finer-resolution model on tropical cyclones, accurately reproduced the frequency of extreme storms in different locations globally.
According to Themistoklis Sapsis at MIT, the new method could generate more accurate predictions about occurrences of forest fires, flooding, and heat waves in a future climate, making it valuable for addressing climate change impacts. However, the research team will perform further studies on future climate scenarios. Pedram Hassanzadeh at the University of Chicago, who was not involved in the study, commented that this correction method showed promising results with the E3SM model and it would be interesting to see it applied to future greenhouse-gas emission scenarios.
This research was partially funded by the U.S. Defense Advanced Research Projects Agency (DARPA) and was published in the Journal of Advances in Modeling Earth Systems. The team included researchers from the Pacific Northwest National Laboratory and MIT.
