Spatial Extreme Value Analysis Using Conditional Modeling Approach

• Activity Number: 264
• Type: Contributed
• Date/Time: Monday, August 10, 2015 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Joanne Lo* and Elizabeth Mannshardt and Montserrat Fuentes and Dongsoo Kim
• Companies: North Carolina State University and North Carolina State University and North Carolina State University and National Climatic Data Center
• Keywords: Spatial Extremes ; Generalized Extreme Value ; Precipitation

Abstract

The 2014 Intergovernmental Panel On Climate Change reports that increasing trends in heavy precipitation events over many areas of the globe are statistically significant. Extreme value methods can be applied in modeling the tail distribution to assess the behavior of such rare events. Due to limited data, an intrinsic problem existing with extremes is the high variability for extrapolation. One way to reduce variability of the estimates is to borrow strength across locations. Various spatial extremes methods show that incorporating spatial methods improves estimates by increasing precision. However, increase in model complexity leads to computational cost and intractable forms for high dimensionality. To capitalize on the spatial information in surrounding sites, we propose methods to incorporate neighboring data in a Conditional Autoregressive (CAR) model framework while preserving the Generalized Extreme Value (GEV) marginals at each location. Using this approach the model gains strength from additional data, computational costs can be reduced, and a tractable solution derived. We apply our model to 12 years of U.S. precipitation data focusing on the North-East region.