Assessing the performance of the independence method in modeling spatial extreme rainfall

Abstract

Spatial statistical methods are often employed to improve precision when estimating marginal distributions of extreme rainfall. Methods such as max-stable and copula models parameterize the spatial dependence and provide a continuous spatial representation. Alternatively, the independence method can be used to estimate marginal parameters without the need for parameterizing the spatial dependence, and this method has been under-utilized in hydrologic applications. This paper investigates the effectiveness of the independence method for marginal parameter estimation of spatially dependent extremes. Its performance is compared with three spatial dependence models (max-stable Brown-Resnick, max-stable Schlather, and Gaussian copula) by means of a simulation study. The independence method is statistically robust in estimating parameters and their associated confidence intervals for spatial extremes with various underlying dependence structures. The spatial dependence models perform comparably with the independence method when the spatial dependence structure is correctly specified; otherwise they exhibit considerably worse performance. We conclude that the independence method is more appealing for modeling the marginal distributions of spatial extremes (e.g., regional estimation of trends in rainfall extremes) due to its greater robustness and simplicity. The four statistical methods are illustrated using a spatial data set comprising 69 subdaily rainfall series from the Greater Sydney region, Australia.