# Rapid Aeolus L2B HLOS Wind Retrieval via BP Neural Network

## TL;DR

This paper introduces a fast neural network model to estimate wind data from satellite lidar observations, improving computational efficiency.

## Contribution

A backpropagation neural network is proposed to emulate Level-2B HLOS wind data from Level-1B observations efficiently.

## Key findings

- The model reproduces the statistical characteristics of the L2B HLOS wind product.
- It captures along-track wind patterns effectively across a 0–20 km altitude range.
- The approach provides a computationally fast alternative to traditional retrieval methods.

## Abstract

Wind field information is a key variable in atmospheric science and weather prediction, and spaceborne Doppler wind lidar provides unique global observations of the horizontal line-of-sight (HLOS) wind. This study develops a data-driven model that maps Aeolus Rayleigh-channel Level-1B (L1B) observables to the operational Level-2B (L2B) HLOS wind product. Using the two Rayleigh discriminator responses as inputs, we train a backpropagation (BP) neural network to learn the nonlinear relationship between Rayleigh-channel measurements and the collocated L2B HLOS winds. The proposed approach is intended as a computationally efficient emulation/approximation of the L2B HLOS output from L1B observations, rather than as an independently validated accuracy-improving retrieval. Model performance is evaluated by agreement with the L2B reference across samples spanning July 2019 to May 2020 and an altitude range of 0–20 km. The results show that the proposed model reproduces the main statistical characteristics and along-track HLOS patterns of the L2B product, providing a fast option for generating L2B-like HLOS estimates from Rayleigh-channel inputs.