Abstract

In numerous dynamic systems, significant environmental challenges, including severe weather events and abrupt climate changes, have become prevalent. In order to fully understand the underlying mechanisms and enhance informed decision-making, a flexible model capable of accommodating extremes is necessary. The existing dynamic spatio-temporal models exhibit limitations in capturing extremes when assuming Gaussian error distributions, whereas the current models for spatial extremes are focused on joint upper tails at two or more locations while assuming temporal independence in the copula-based modeling framework. Here, we introduce a novel class of dynamic spatio-temporal models capable of accommodating both high and low extremes through dimension reduction. We use a mixture of stable distributions with varying tail indices in the lower dimensional latent space, and re-project unto the physical space using a redistribution kernel embedded in the hierarchical construct. Our model can describe complex advective and diffusive dynamics with relatively few parameters and characterize differing levels of same-tail and opposite-tail extremal dependence which are non-stationary across space and time. We demonstrate the effectiveness of our methods by applying them to turbulence flow observations that are chaotic and highly irregular.