Identifying Similar Thunderstorm Sequences for Airline Decision Support Using Optimal Transport Theory

We propose a new method to identify similar thunderstorm spatial-temporal sequences to support airline operations based on the optimal transport theory. Different from existing geometric methods, which often suffer from over-approximation of the covering geometric objects, our method models each thunderstorm as a probability distribution supported by the observed weather data. The core of our approach lies in measuring the similarity between thunderstorm sequences through the Wasserstein distance of their respective probability distributions. By setting different time weights and filter functions, this method can also incorporate the temporal features of the thunderstorms and consider the weather impact on key airspace/airport infrastructures. Furthermore, we apply a clustering algorithm within the probability distribution space of thunderstorms to categorize common patterns of archived thunderstorms in a given airspace region. We illustrate the effectiveness of this new method with our results with real-world weather data in the Dallas Fort Worth airspace.