Tropospheric Refractivity Estimation Using DeepONet

Abstract

The problem of tropospheric tomography is considered. The task is formulated as finding an inverse operator that maps electromagnetic field measurement data to the spatial distribution of the refractive index. The DeepONet architecture and an automatically differentiable parabolic equation are used to construct the inverse operator. Training data is generated using the parabolic equation method. Unlike previous works, the proposed method does not require prior information about the spatial distribution of the refractive index profile. The effectiveness of the proposed method is demonstrated for typical profiles: evaporation duct, surface duct, and trilinear profiles. A comparison with the adjoint method and genetic algorithm was carried out. It is shown that the proposed method allows performing inversion with an accuracy of up to 1 M-unit in real time.