Toward Unsupervised Segmentation of Extreme Weather Events

Extreme weather is one of the main mechanisms through which climate change will directly impact human society. Coping with such change as a global community requires markedly improved understanding of how global warming drives extreme weather events. While alternative climate scenarios can be simulated using sophisticated models, identifying extreme weather events in these simulations requires automation due to the vast amounts of complex high-dimensional data produced. Atmospheric dynamics, and hydrodynamic flows more generally, are highly structured and largely organize around a lower dimensional skeleton of coherent structures. Indeed, extreme weather events are a special case of more general hydrodynamic coherent structures. From a physics perspective, understanding such emergent structures in complex systems is an overwhelming challenge for mathematical analysis of the equations of motion, and from a machine learning perspective it is a fundamentally unsupervised problem. Bringing these perspectives together, we present a scalable physics-based representation learning method that decomposes spatiotemporal systems into their structurally relevant components, which are captured by latent variables known as local causal states. The shared coordinate geometry between the observable field and the latent local causal state field makes the learned representation immediately usable for extracting structural features such as coherent structures, and also provides some degree of interpretability. For complex fluid flows we show that our method is capable of capturing known coherent structures, and with promising segmentation results on CAM5.1 water vapor data we outline the path to extreme weather identification from unlabeled climate model simulation data.