Modelling daily precipitation occurrence with Markov chain - the role of the limitness of the learning sample