Accurate retrieval of atmospheric relative humidity (RH) profiles is essential for improving our understanding of atmospheric thermodynamics and climate change. Nevertheless, it remains challenging, as traditional models rely exclusively on vertical brightness temperature (BT) observations. Here, we present a novel retrieval algorithm called AngleNet, a groundbreaking deep-learning model that leverages multi-angle BT observation from ground-based microwave radiometers (MWRs). The innovative “multi-angle-aware” module in AngleNet effectively exploits previously underutilized oblique scanning angle data by accurately capturing these nonlinear relationships between BT and RH profiles, and precisely characterizes its vertical fine structure. Based on the 7-year (2018–2024) in situ measurements from Beijing, Nanjing, and Shanghai, validation results reveal that AngleNet achieves substantial improvements, with an average R 2 of 0.71 and a root mean square error (RMSE) of 10.39%, surpassing conventional models such as LGBM (light gradient boosting machine) and RF (random forest) by over 10% in both metrics, and demonstrating a remarkable 41% increase in R 2 and a 10% reduction in RMSE compared to the previous BRNN method (batch normalization and robust neural network). Moreover, additional independent validation results demonstrate that AngleNet exhibits excellent stability and retrieval accuracy during periods without radiosonde measurements. Feature analysis and evaluations of the “multi-angle-aware” module indicate that optimal RH retrieval performance is achieved by combining zenith-angle BTs with oblique angles at 30° and 19.2°. AngleNet breakthrough performance is especially notable in consistently capturing complex RH profile features, which are critical for accurate numerical weather forecasting and climate monitoring.
Yang et al. (Wed,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: