Articles | Volume 15, issue 8
https://doi.org/10.5194/nhess-15-1757-2015
https://doi.org/10.5194/nhess-15-1757-2015
Brief communication
 | 
11 Aug 2015
Brief communication |  | 11 Aug 2015

Statistical detection and modeling of the over-dispersion of winter storm occurrence

M. Raschke

Abstract. In this communication, I improve the detection and modeling of the over-dispersion of winter storm occurrence. For this purpose, the generalized Poisson distribution and the Bayesian information criterion are introduced; the latter is used for statistical model selection. Moreover, I replace the frequently used dispersion statistics by an over-dispersion parameter which does not depend on the considered return period of storm events. These models and methods are applied in order to properly detect the over-dispersion in winter storm data for Germany, carrying out a joint estimation of the distribution models for different samples.

Download
Short summary
Here, I discuss and improve the detection and modeling of the over-dispersion of winter storm occurrence using the example of Germany. For this purpose, the generalized Poisson distribution and criteria for the model selection are introduced. Correct statistical model selection ensures the statistical significance of the model, including an over-dispersion. The relation between expectation and variance of a thinned inhomogeneous Poisson process is derived. This is also applied to data.
Altmetrics
Final-revised paper
Preprint