Meteorological satellite remote sensing is important for numerical weather forecasts, but its accuracy is affected by many things during observation and retrieval, showing that it can be improved. As a standard way to measure wind from space, atmospheric motion vectors (AMVs) are used. They are separate pieces of information spread out in the troposphere, which gives them more depth than regular surface or sea surface wind measurements. This makes rectifying problems more difficult. For error correction, this research builds a deep-learning model that is specific to AMVs. The outcomes show that AMV observational errors are greatly reduced after correction. The root mean square error (RMSE) drops by almost 40% compared to ERA5 true values. Among these, the optimization of solar observation errors exceeds 40%; the discrepancies at varying atmospheric pressure altitudes are notably improved; the degree of optimization for data with low QI coefficients is substantial; and there remains potential for enhancement in data with high QI coefficients. Furthermore, there has been a significant enhancement in the consistency coefficient of the wind’s physical properties. In the assimilation forecasting experiments, the corrected AMV data demonstrated superior forecasting performance. With more training, the model can fix things better, and the changes it makes last for a long time. The results show that it is possible and useful to use deep learning to fix errors in meteorological remote-sensing data.
Accurate initial conditions are crucial for improving numerical weather prediction (NWP). Variational data assimilation relies on a static background error covariance matrix (B), yet its variance estimation is often inaccurate, affecting assimilation and forecast performance. This study introduces DRL-AST, a deep reinforcement learning-based adaptive variance rescaling strategy that dynamically adjusts the variances of B to optimize forecast skill through improved assimilation performance. By formulating variance rescaling as a Markov Decision Process and employing an actor–critic framework with Proximal Policy Optimization, DRL-AST autonomously selects spatio-temporal rescaling factors, enhancing assimilation and forecast accuracy without additional computational cost. As a new paradigm for adaptive variance tuning, DRL-AST demonstrates competitive improvements in forecast skill in experiments with the Lorenz-96 model by generating initial states that better conform to model dynamical consistency. Given its adaptability and efficiency, DRL-AST holds great potential for application in high-dimensional NWP models, where deep learning-based dimensionality reduction and reinforcement learning techniques could further enhance its feasibility and effectiveness in complex assimilation frameworks.
Atmospheric motion vectors, which can be used to infer wind speed and direction based on the trajectory of cloud movement, are instrumental in enhancing atmospheric wind-field insights, contributing notably to wind-field optimization and forecasting. However, a widespread problem with vector data is their inaccuracy, which, when coupled with the mediocre effectiveness of existing correction methods, limits their practical utility in forecasting, often falling short of expectations. Deep-learning techniques are used to refine atmospheric motion vector data from the FY-4A satellite, notably enhancing data quality. Post-training data undergoes a thorough analysis using a quality evaluation function, followed by its integration into a numerical weather prediction system in order to conduct forecasting experiments. Results indicate a marked improvement in data quality post-error correction by the model, characterized by a significant reduction in root mean square error and a notable increase in correlation coefficients. Furthermore, refined data demonstrate a considerable enhancement in the accuracy of meteorological element forecasts, particularly for Asian and Western Pacific regions.