Assessing a community’s risk of extreme weather events – such as catastrophic storms and catastrophic flooding – is a key task for policymakers who are trying to prepare for the potential impacts of global climate change. Despite major advancements in technology and modeling techniques, however, these forecasts leave a lot to be desired. Now, a team of researchers from the Massachusetts Institute of Technology (MIT) and the Pacific Northwest National Laboratory (PNNL) have developed a novel approach that combines machine learning with dynamical systems theory to correct large-scale coarse climate models.
Although long-term climate modeling has improved, today’s large-scale models are still constrained by resolution size. These models commonly generalize weather conditions over large 100 kilometers grids, but they miss finer details such as clouds or storms that happen at a smaller scale. These details, although smaller, play a significant role in extreme weather events.
The researchers have introduced an innovative method that overlays an algorithm on a model’s output that guides the simulation toward a result that more closely represents reality. The algorithm uses machine learning, taking past data such as temperature and humidity, learning associations within the data that represent fundamental dynamics among weather features, and then uses these learned associations to adjust a model’s predictions.
The research team initially tested the approach using the Energy Exascale Earth System Model (E3SM), a model used by the US Department of Energy, which maps global climate patterns at a 110 km resolution. The machine-learning scheme was trained using eight years of past data on various weather parameters and the E3SM model. Subsequent application of the trained algorithm resulted in the corrected model generating climate patterns that more closely mirrored real-world observations from the past 36 years.
Going forward, the method can be used to provide a better understanding of how extreme events such as forest fires, floods, and heat waves will look in a future affected by climate change. Notably, this new correction scheme can be applied to other global climate models and should assist policymakers in determining where and how often extreme weather will occur as global temperatures rise. Accurate predictions of extreme weather events are vital for effective preparation and proactive strategies to mitigate the impact of climate change. The team’s findings have been published in the Journal of Advances in Modeling Earth Systems. The project received partial funding from the U.S. Defense Advanced Research Projects Agency.
