M.G. Matveev - Dr. Sc. (Eng.), Professor, Research Scientist, Russian Air Force Military Educational and Scientific Center «Zhukovsky-Gagarin Air Force Academy» (Voronezh). E-mail: mgmatveev@yandex.ru
S.L. Kirnosov - Ph. D. (Eng.), Dr. Sc. Candidate, Russian Air Force Military Educational and Scientific Center «Zhukovsky-Gagarin Air Force Academy» (Voronezh). E-mail: slk_met@mail.ru

In this article in order to obtain information about the dynamics of the spatial and temporal distribution of air temperature was built prognostic meteorological model. Building a model based on the raw data taken from the three-dimensional grid and is a scalar fields in the space temperature of Earth's atmosphere. In order to exclude the influence of the underlying surface land air temperature information is considered higher than 300 meters. One of the main aspects of construction of the above model was assumption of the existence of correlation processes of change in the temperature of adjacent grid nodes. In this paper made a formalization of tasks performed multivariate modeling of the dynamics of the vector field temperature on the basis of using the notion of autoregression based on the physical features of atmospheric processes. This accounting treatment is done to separate the source of statistical data on the classes in which the stationarity conditions for the occurrence of meteorological process. Construction of the model performed when used at each grid as a feature of classification - the vector length and direction of the temperature gradient. Also in the work done keeping statistical significance parameters have a direct influence on the temperature. For this purpose, the transition is made from the direct use of the numerical values of air temperature deviations from their respective expectations. As the operator, which determines the result of the model chosen artificial neural network. Built in the adjusted predictive model differs from the existing fact that artificial neural network with the same structure has been used previously, but without analyzing the criteria for classification statistics, analysis of the importance of parameters and without deviations from expectation in the class of homogeneous statistics. To assess the potential effectiveness of the calculated coefficient of determination and the mean error of approximation, numerical values indicate increased relevance and quality of the updated model. In this paper the main conclusion was made that the built-adjusted prognostic meteorological model of spatial-temporal distribution of temperature based on the synthesis of autoregression and artificial neural network is acceptable for use in solving practical problems in the aviation meteorological support.