Pattern-based analysis of probabilistic and statistical characteristics of extreme precipitation

Precipitations are the key parameters of hydrological models; so, research related to precipitation processes is necessary for solving various applied problems. The paper demonstrates a violation of the Markov property for precipitation observed in essentially different climatic regions — in the cities of Potsdam and Elista. Such information about the data, along with previously studied properties, represents the basic information which is necessary for the further correct construction of probabilistic models, in particular, for probability distribution of the volumes of extreme precipitation. For the analysis of the probabilistic behavior of the precipitation process and the construction of forecasts, it is suggested to use chains of events (patterns) extracted from the data. At the same time, statistical procedures are automated using the software tools of the MATLAB package. Neural networks were used as an alternative forecasting tool based on patterns, and the best results were demonstrated via the architecture that takes into account a seasonality, has two hidden layers of neurons and a sigmoid activation function. The ideas for further research in this field are suggested.