Climate change experts are turning to an innovative approach to better predict extreme weather events and the impacts of climate change on specific locations. This new methodology “corrects” global climate models, combining machine learning with dynamical systems theory to bring the models’ simulations much closer to expected real-world patterns. This approach can help policymakers effectively estimate a location’s future risk of extreme weather events more accurately.
Researchers led by Themistoklis Sapsis at MIT have found that relying solely on global climate models leaves significant room for error, as they often only depict a course resolution and cannot provide detailed insights for specific reigons or cities. For instance, a model may be able to predict future climate conditions for the northeastern U.S., but would not be able to provide an accurate prediction specifically for Boston.
To bring the resolution down to a more localized scale, Sapsis and colleagues have developed a scheme that leans on machine learning to correct the predictions of coarse climate models. The machine learning aspect gives an algorithm the ability to learn from previous data, and then uses that data to identify and apply associations within the information that represent fundamental weather dynamics. These learned associations can then be utilized to correct a model’s predictions.
The researchers took simulations from existing coarse models, overlaid them with their correcting algorithm, and were then able to create predictions that more closely represented real-world conditions. Preliminary tests of this approach using the Energy Exascale Earth System Model (E3SM), which was supplied with eight years’ worth of temperature, humidity, and wind speed data, the algorithm produced climate models that better aligned with the observed weather patterns from previous years.
Moreover, when pairing the corrected coarse model with more specific, finer-resolution model of tropical cyclones, the team found that the approach accurately depicted the frequency of extreme storm events at specific locations across the globe.
This advancement in climate change modelling could be crucial in gaining a deeper understanding of the ways in which extreme weather events will manifest in the face of climate change. With climate change poised to impact every facet of human life, this means of accurately predicting the future of our environment could be critical in ensuring we’re adequately prepared and have the right solutions in place.
While the results are promising, Sapsis and his fellow researchers are continuing to delve further into how this modeling approach could be used to understand the ways in which future climate scenarios could come to pass, such as forest fires, heatwaves, and large scale flooding events.
This research has garnered the interest of other experts in the field, notably Pedram Hassanzadeh from the University of Chicago, who expressed keen interest in seeing this new framework implemented into future greenhouse-gas emission scenarios. The research was funded in part by the U.S. Defense Advanced Research Projects Agency.
