Abstract:
|
In the univariate framework, extreme events are classically defined as block maxima or excesses above a high threshold. These approaches yield, however, two main drawbacks. First, working with maxima or excesses implies that a lot of observations are completely disregarded: the dataset is artificially shopped down into two pieces, namely large observations and the rest. This necessarily imposes different statistical models for each piece. Second, it raises a non-trivial and very practical difficultly: how to choose the threshold, or equivalently the block size, which correctly discriminates between extreme and non-extreme data. In this work, we propose a statistical model for rainfall data, which is in compliance with Extreme-Value Theory on its upper and lower tails, while avoiding a threshold selection. Inference is performed using the fast method of probability weighed moments (PWMs). The flexibility of this new model and the efficiency of PWMs estimators are illustrated on simulated and real precipitation data.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.