Nonparametric Spatial Models for Extremes: Application to Extreme Temperature Data.

Area of Science:

• Environmental statistics
• Extreme value theory
• Spatial modeling

Background:

• Estimating extreme temperature event probabilities is challenging due to limited historical data and the need for spatial extrapolation.
• Characterizing uncertainty in these probability estimates across different locations is also a significant issue.
• While univariate extreme value statistical models are established, spatial extreme data modeling remains an active research area.

Purpose of the Study:

• To introduce a novel nonparametric model for spatial extreme events.
• To develop a flexible alternative to existing parametric copula models for extreme value analysis.
• To apply a Bayesian framework for robust inference, accounting for data and model uncertainties.

Main Methods:

• Development of a Dirichlet-based copula model for spatial extremes.
• Implementation of a Bayesian statistical framework for model fitting and uncertainty quantification.
• Application to annual maximum temperature data in the east-south-central United States.

Main Results:

• The proposed Dirichlet-based copula model offers a flexible approach to spatial extreme value modeling.
• The Bayesian framework effectively incorporates various sources of uncertainty.
• Successful application to real-world temperature data demonstrates the model's utility.

Conclusions:

• The new nonparametric spatial extreme value model provides a valuable tool for environmental risk assessment.
• The Bayesian approach enhances the reliability of extreme event probability estimations.
• This research contributes to advancing the statistical methods for analyzing spatial extreme temperature data.