B1749
Title: Physics-informed max-stable spatial processes for inference in regions with no-observations
Authors: Jose Blanchet - Stanford University (United States) [presenting]
Ali Hasan - Duke University (United States)
Vahid Tarokh - Duke University (United States)
Abstract: Max-stable distributions are used for inference about extreme future realizations based on relatively limited observations. These multivariate distributions represent statistical laws that can be calibrated based on observations collected in a given geographical region. We explore the use of physical laws (encoded via partial differential equations) to extend the inference from areas in which observations are collected to areas in which no observations are collected, but for which physical principles can be applied. This leads to new definitions of solutions to PDEs with random input, which are appropriate for extremes. We illustrate the method in several applications, including extreme heat estimation.
