A team of scientists from MIT’s Department of Mechanical Engineering has developed a new method using machine learning to correct and enhance prediction accuracy in climate models. These advancements could provide significantly greater insights into the frequency of extreme weather events with more localized precision, improving the ability to plan and mitigate for future climatic changes.
Traditional global climate models, relied upon by policymakers, offer predictions based on decades and even centuries into the future, but are limited by a lack of fine detail. While they may provide predictions for larger geographical areas, exact assessments for smaller locations, such as cities, are elusive and therefore limits effectiveness in forecasting specific events like floods or storms.
The new method developed by the MIT team overlays an algorithm onto the climate model’s simulations, bringing them more in line with actual world conditions. The machine-learning algorithm learns from past data such as recorded global humidity and temperature levels, and uses these learned associations to correct the model’s predictions.
The team applied this new technique to simulations produced by the Energy Exascale Earth System Model – a climate model run by the U.S. Department of Energy. They used eight years of past data to train the algorithm, and found that their new, corrected model produced climate patterns that more accurately reflected recorded observations from the past 36 years.
Notably, the corrected models also successfully predicted the frequency of extreme weather events when combined with a specific, finer-resolution model of tropical cyclones. This indicates a promising future for the accurate forecasting of extreme weather events, enabling better climate mitigation strategies.
Climate change is set to impact every aspect of life, including biodiversity, food security, and the economy, so improving the accuracy and precision of climate models is crucial. This new method has shown the potential to accurately predict how specific weather occurrences, such as intense heat waves, would impact specific locations. The continuing work of the MIT team is now focusing on analyzing future climate scenarios.
This method can be applied to any global climate model and it is anticipated to play a crucial role in determining where and how frequently extreme weather will strike as global temperatures continue to rise. The research was partially funded by the U.S. Defense Advanced Research Projects Agency and has been published in the Journal of Advances in Modeling Earth Systems.