Нейронная сеть в задачах краткосрочного прогнозирования по результатам микроволнового радиометрического зондирования атмосферы

Аннотация

The article addresses matters of including a neural network in results’ processing procedure of VHF radiometric sounding of the atmosphere for the task of short-term forecasting of status change of the ground layer. There was demonstrated difference between approaches to using data of VHF radiometric remote sensing in major studies of atmospheric processes by solving inverse problems of estimating meteorological parameters and for the task of short-term forecasting of meteorological parameters’ changes in the ground layer in the assigned area as per operation results of tri-band VHF radiometric system by integrating a neural network trained on daily measurements’ results. Information on technical issues of VHF radiometric sounding is presented. It is shown that receiving radiothermal atmospheric radiation was done onto a common aperture of the 2400 mm radius mirror antenna in three frequency bands  (7.5 GHz, 10 GHz and 22 GHz central wavelengths) with sequential signal frequency selection of signals in the antenna feed in a two-mode reception to ensure levelling  of background noise effect received via dispersive region of the antenna pattern. The features of training neural network are examined to solve the problem of forecasting meteorological parameters - temperature, humidity and precipitation intensity of the ground layer. It is pointed out that to do this, data was collected from weather stations in cities located near the ground projection of antenna sight line within 200 km range from base location of VHF radiometric system. When performing teaching procedure there were formed three measurement arrays of varying lengths of time for VHF radiometric system with 39, 103 and 159 readings and testing was performed in the time interval from the completion of the training array up to July 2023. There are presented training and testing results of neural network operation for predicting three meteorological parameters - temperature, precipitation intensity and humidity in the specified cities. The results obtained made clear the prospects of incorporating a neural network into VHF radiometric systems for short-term weather forecasts.