SpatialGEV: Fast Bayesian inference for spatial extremevalue models in R

Extreme weather phenomena such as floods and hurricanes are of great concern due to their potential to cause extensive damage.To develop more reliable damage prevention protocols, statistical models are often used to infer the chance of observing an extreme weather event at a given location (Coles & Casson, 1998;Cooley et al., 2007;Sang & Gelfand, 2010).Here we present SpatialGEV, an R package providing a fast and convenient toolset for analyzing spatial extreme values using a hierarchical Bayesian modeling framework.In this framework, the marginal behavior of the extremes is given by a generalized extreme value (GEV) distribution, whereas the spatial dependence between locations is captured by modeling the GEV parameters as spatially varying random effects following a Gaussian process (GP).Model inference is carried out using an efficient implementation of the Laplace approximation, which produces highly accurate posterior estimates several orders of magnitude faster than Markov Chain Monte Carlo (MCMC) methods.Users are provided with a streamlined way to build and fit various GEV-GP models in R, which are compiled in C++ under the hood.For downstream analyses, the package offers methods for Bayesian parameter estimation and forecasting of extreme events.