Development of Methods and Technologies for Operational Assimilation of Meteorological Observations at the Hydrometcenter of Russia

A brief description of the system functioning at the Hydrometcenter of Russia for operational assimilation of meteorological observations and results of the work on its development are given. The main direction of its development is creating an original ensemble variational data assimilation system using neural network modeling techniques. Originality of the system consists in a new multiscale convolutional analysis. Results of testing the new data assimilation technique for a two-dimensional case are presented. In numerical experiments, the convolutional analysis has demonstrated higher accuracy than the traditional ensemble variational approach. A new multiscale technique is described, and results of its testing are given. The second direction of the development is increasing efficiency of satellite data assimilation. Results of retrieving sea ice concentration using data of AMSR2 and MTVZA-GYa microwave radiometers using machine learning methods are presented. A technique of assimilation of delayed observations is proposed and tested.