Analysis of Annual Maxima Daily Precipitation Data Using a Max-Stable Spatial Process: Quantifying the Cost of Assuming Spatial Independence Among Extreme Data

Assumptions of inter-site independence are unrealistic for non-localized extreme events, potentially resulting in miscalculations of risk. We quantify the cost of assuming inter-site independence by applying a max-stable spatial process to model a set of pointwise annual maxima precipitation data in Oregon (OR). A max-stable model is a spatial process extension of extreme value theory which also explicitly accounts for the dependence among the extreme data. By virtue of modeling the joint distribution, more complex areal-based assessments of risk can also be estimated with a max-stable process. Following a comprehensive assessment of spatial dependence of the OR pointwise AM daily precipitation dataset, we outline a max-stable process deployment. Pointwise return levels are compared from trend surfaces fitted with and without a treatment of the observed dependence among the extreme precipitation data. We also compare areal-based exceedance calculations derived from our fitted general max-stable process with a comparable general max-stable process whose dependence parameterization enforces an effective inter-site independence throughout the model domain.