CMStatistics 2018: Start Registration
View Submission - CMStatistics
Title: A hierarchical max-infinitely divisible process for spatial precipitation modeling Authors:  Gregory Bopp - Pennsylvania State University (United States) [presenting]
Raphael Huser - King Abdullah University of Science and Technology (Saudi Arabia)
Benjamin Shaby - Penn State University (United States)
Abstract: The hazards of atmospheric phenomena such as extreme precipitation are often largely determined by their spatial extent. Different choices of extremal dependence classes can lead to vastly different conclusions about the risk of such hazards. We will present a class of models for spatial extremes that allows for a smooth transition between extremal dependence types. The conditional representation of the proposed model allows for full-likelihood based inference via Markov chain Monte Carlo that scales to large datasets. The model extends a familiar max-stable class to a broader family of max-infinitely divisible processes that allows for more flexible spatial dependence types. Due to a construction in terms of flexible random basis functions that are estimated from the data, straightforward inspection of the predominant spatial patterns of extremes is also possible. The proposed model is applied to extreme precipitation to examine flood risk over hydrologically defined watersheds in eastern North America.