Engineers at the Massachusetts Institute of Technology have developed an artificial intelligence tool capable of generating plausible extreme weather scenarios without being trained on historical examples of similar disasters. The tool was developed by mechanical engineering graduate student Kai Chang and Professor Themis Sapsis. It creates maps of events that have never been recorded in a particular region but remain statistically possible. Each map includes estimates of the event’s potential intensity, duration and geographical reach. Modelling Extreme Events With No Historical Precedent Sapsis holds the William I. Koch Professorship in Mechanical and Ocean Engineering at MIT. Both researchers are affiliated with the MIT Center for Computational Science and Engineering, while Sapsis also holds an appointment with the MIT Institute for Data, Systems, and Society. The researchers presented their method, called Extreme Event Aware or η-learning, in a paper published in Nature Communications on August 20. Conventional risk models generally rely on datasets containing previous examples of extreme events. Insurers, urban planners and electricity-grid operators use these models to estimate what a once-in-a-century storm could look like in a specific location. Such systems typically identify the conditions that produced earlier disasters and then project comparable patterns into the future. According to the researchers, this limits their ability to model events more severe than anything included in the historical record. The new approach instead seeks to estimate the possible scale and impact of rare disasters that have never occurred before. Combining Statistical Indicators With Spatial Data The algorithm uses two main types of information. The first consists of point statistics that describe how frequently a particular level of intensity occurs in a dataset, such as the maximum rainfall recorded anywhere on a map. The second consists of spatial maps showing how the effects of an event vary across different locations. By learning