Simulation of rain events time series with Markov model

This work presents a rain rate time series model based on a two-level Markov model structure. 'Rain' or 'no-rain' events are generated in two steps: in a first time, the 'rain' or 'inter-rain' duration of the considered event is determined according to the experimental data series. This corresponds to the first level of the model that is a semi-Markov model. In a second time, the rain rate intensities are generated resorting to an N states hidden Markov model based on the conditional probabilities of experimental rain samples (modeling the dependence of the current rain intensity with the previous one). This two-level model produces simulated rain samples whose statistics fit very accurately those of the experimental data without using any stored rain events in the generation step when the model is known.