Multisite rainfall dependence

The spatial structure of precipitation is of great importance for hydrology. Spatial dependence is usually described using bivariate measures such as covariance or rank correlation. These measures cannot capture the multidimensional dependence accurately. In this paper data from pairs and triples of precipitation stations are considered. Models for the reproduction of the bivariate dependences are investigated from the trivariate perspective. Copulas and entropy are used to quantify the three dimensional dependences. In a case study in South-West Germany using 32 daily precipitation series with 30 years of observations it is shown that meta-Gaussian models are unable to reproduce the three dimensional dependence. Possible alternative multivariate models will also be discussed.