Filtering weather dynamics to find locations of change

Weather can be highly volatile, making extreme events difficult to predict and model. With a special set of data filters, Yusuke Imoto and Tomoo Yokoyama developed a method for inferring where and when a highly sensitive system is likely to change.

“In some system states, small perturbations hardly affect the eventual outcome. In other states, however, even a tiny perturbation can dramatically change the system’s future,” said Yokoyama. “Our method focuses on this exact sensitivity to identify the tipping points where a small perturbation can change the system’s fate.”

Their approach relies on visualizing how safe or risky a specific state is and identifying when and where this changes. Applied retroactively as a case study to the severe tropical storm Dolphin, which occurred off the coast of Japan in 2020, the researchers were able to identify the specific times and locations where the storm’s path was the most sensitive and likely to change due to tiny fluctuations.

“This technique holds promise in its ability to visualize how much time we have left to change a situation and the minimum impact required to make that change during extreme weather events,” said Yokoyama. “For example, it allows us to quantitatively understand that past a certain point in time, small interventions are no longer enough to change the future outcome.”

Imoto and Yokoyama are working on expanding their method to be more flexible — for example, by extending the theory to analyze phenomena through quantitative evaluation, rather than just a binary metric. With these adjustments, they hope their technique will be more applicable to a broader range of real-world systems, including extreme weather applications, pointing out when a specific state takes a negative turn.

Source: “Filtrations indexed by attracting levels and their applications,” by Yusuke Imoto and Tomoo Yokoyama, Chaos (2026). The article can be accessed at https://doi.org/10.1063/5.0305367 .