Spatial Dependence Modeling of Precipitation Extremes Using a Conditional Approach

• Activity Number: 87
• Type: Invited
• Date/Time: Sunday, August 3, 2014 : 8:30 PM to 10:30 PM
• Sponsor: Section on Statistics and the Environment
• Author(s): Lina Lin*+ and Peter Guttorp
• Companies: University of Washington and University of Washington
• Keywords: Spatial ; Precipitation ; Multivariate Extremes ; Conditional modelling

Abstract

Prediction of extreme precipitation events is important for flood protection and emergency planning. Rainfall often exhibits strong spatial patterns, which must be taken into account for reliable prediction. Current tools for multivariate modelling of extremes can be rather limiting as they only capture simultaneous extremes. With rainfall, different regions may exhibit different extremal patterns with respect to one another in a complicated manner that is influenced by the underlying topography of the area of interest, among other factors; thus, simultaneous extremes across all sites is a rare event and thus of little practical interest. In their 2004 paper, Heffernan and Tawn presented an alternative semiparametric approach which involves modelling the joint distribution of the components conditional on one component being extreme while making certain assumptions on its form. Here, we provide a modest extension to their method by building a spatial dependence structure into the assumed asymptotic form of the conditional distribution. We apply the modified method to Norwegian precipitation data spanning years 1961-2000.