Atmosphere (Feb 2022)

Weight Loss Function for the Cooperative Inversion of Atmospheric Duct Parameters

  • Jie Han,
  • Jia-Ji Wu,
  • Hong-Guang Wang,
  • Qing-Lin Zhu,
  • Li-Jun Zhang,
  • Chao Zhang,
  • Qian-Nan Wang,
  • Hui Zhao

DOI
https://doi.org/10.3390/atmos13020338
Journal volume & issue
Vol. 13, no. 2
p. 338

Abstract

Read online

Low-altitude atmospheric ducts are abnormal atmospheric phenomena in the troposphere, impacting the operation of microwave or ultrashort wave radio systems. Therefore, the real-time acquisition of low-altitude atmospheric duct parameters is essential to ensure the successful operation of radio systems. Remote sensing methods based on deep learning are, presently, the most important tools to infer duct parameters. In a traditional deep learning loss function, different duct parameters adopt the same weight coefficient. This study establishes a weight loss function and proposes a method for determining the weight coefficient based on the extended Fourier amplitude sensitivity test method. Based on Global Navigation Satellite System (GNSS) occultation signals, the cooperative inversion model of atmospheric duct parameters is established. Test results show that our proposed loss function was feasible, effective, and yielded a higher inversion accuracy than the traditional loss function.

Keywords