A stochastic rainfall model with intensity dependent autocorrelations.

Abstract

The space-time behaviour of precipitation is very complex. The knowledge of the dependence structures in space and time is very important for the assessment of flood risks. There are many different models available for stochastic simulations of precipitation time series. Most of the models are constructed such that the simulated time series match the autocorrelation structure of the observations in time along with the reproduction of spatial correlations. However, both auto and spatial correlations are value dependent i.e., if it is the upper or the lower tail. High and low intensity values have different dependence structures which have a significant influence on simulated extremes in space. In this presentation, first indicator correlations are introduced to show the intensity dependence of precipitation both in space and time at various resolutions. Then, a stochastic simulator based on gradual change of the correlations for the values in different parts of the distributions is introduced. The idea is that value dependent correlation is changed in such a way that the overall values remains same as that of a reference but not when considering values in different sections of the distributions alone. The model is applied to a large number of German catchments with hourly temporal resolution. The results are carefully analysed and compared to classical approaches.