Application of machine learning to improving numerical weather prediction

Abstract

In this paper are presented a brief overview of trends in numerical weather prediction, difficulties and the nature of their occurrence, the existing and perspective ways to overcome them. The neural network architecture is proposed as a promising approach to increase the accuracy of the 2m temperature forecast by COSMO regional model. This architecture allows predicting errors of the atmospheric model forecasts with their further corrections. Experiments were conducted with different prehistories of regional model errors. The number of epochs was determined after which the increase of the so-called retraining of the network had place. It is shown that the proposed architecture makes it possible to achieve an improvement of 2m temperature forecast in approximately 50 % of cases.