Rapid Aeolus L2B HLOS Wind Retrieval via BP Neural Network

Area of Science:

• Atmospheric science
• Remote sensing
• Meteorology

Background:

• Spaceborne Doppler wind lidar offers unique global horizontal line-of-sight (HLOS) wind observations.
• Accurate wind field information is crucial for weather prediction and atmospheric science.

Purpose of the Study:

• To develop a data-driven model mapping Aeolus Rayleigh-channel Level-1B (L1B) observables to the Level-2B (L2B) HLOS wind product.
• To create a computationally efficient emulation of the L2B HLOS wind product from L1B measurements.

Main Methods:

• A backpropagation (BP) neural network was trained using two Rayleigh discriminator responses as inputs.
• The model learned the nonlinear relationship between Rayleigh-channel measurements and collocated L2B HLOS winds.
• Model performance was evaluated against L2B reference data from July 2019 to May 2020, covering altitudes from 0-20 km.

Main Results:

• The developed model successfully reproduces key statistical characteristics of the L2B HLOS wind product.
• The model captures the along-track HLOS wind patterns observed in the L2B data.
• The approach provides a fast option for generating L2B-like HLOS wind estimates.

Conclusions:

• The data-driven model offers an efficient approximation of Aeolus L2B HLOS wind products from L1B data.
• This method serves as a valuable tool for generating rapid HLOS wind estimates, supporting atmospheric studies.
• The model demonstrates the potential of machine learning in processing and interpreting complex atmospheric data from spaceborne instruments.