Sea surface wind (SSW) is a crucial parameter for meteorological and oceanographic research, and accurate observation of SSW is valuable for a wide range of applications. However, most existing SSW data products are at a coarse spatial resolution, which is insufficient, especially for regional or local studies. Therefore, in this paper, to derive finer-resolution estimates of SSW, we present a novel statistical downscaling approach for satellite SSW based on generative adversarial networks and dual learning scheme, taking WindSat as a typical example. The dual learning scheme performs a primal task to reconstruct high resolution SSW, and a dual task to estimate the degradation kernels, which form a closed loop and are simultaneously learned, thus introducing an additional constraint to reduce the solution space. The integration of a dual learning scheme as the generator into the generative adversarial network structure further yield better downscaling performance by fine-tuning the generated SSW closer to high-resolution SSW. Besides, a model adaptation strategy was exploited to enhance the capacity for downscaling from low-resolution SSW without high-resolution ground truth. Comprehensive experiments were conducted on both the synthetic paired and unpaired SSW data. In the study areas of the East Coast of North America and the North Indian Ocean, in this work, the downscaling results to 0.25° (high resolution on the synthetic dataset), 0.03125° (8× downscaling), and 0.015625° (16× downscaling) of the proposed approach achieve the highest accuracy in terms of root mean square error and R-Square. The downscaling resolution can be enhanced by increasing the basic blocks in the generator. The highest downscaling reconstruction quality in terms of peak signal-to-noise ratio and structural similarity index was also achieved on the synthetic dataset with high-resolution ground truth. The experimental results demonstrate the effectiveness of the proposed downscaling network and the superior performance compared with the other typical advanced downscaling methods, including bicubic interpolation, DeepSD, dual regression networks, and adversarial DeepSD.

The world’s first operational early morning-orbit meteorological satellite, Fengyun-3E (FY-3E), launched on 5 July 2021, and has expanded global data coverage for numerical weather prediction (NWP) of 6-h assimilation windows. To determine whether the early morning orbit affects microwave observations and the benefits to NWP, radiance from FY-3E microwave instruments was assimilated into the operational global Yinhe four-dimensional variational assimilation (YH4DVAR) system. The performance of the microwave instruments was first assessed based on the statistics of background departure. The sunlight conditions have little effect on FY-3E microwave observations, and ascending and descending pass observations could be uniformly assimilated for both the Microwave Humidity Sounder-2 (MWHS-2) and the Microwave Temperature Sounder-3 (MWTS-3). The standard deviation of observation minus background (OmB) after variational bias correction increased, indicating slightly reduced performance of MWTS-3 compared with equivalent Advanced Technology Microwave Sounder (ATMS) channels on SNPP/NOAA-20. Based on reduced OmB mean and standard deviation values, FY-3E MWHS-2 performs better than MWHS-2 on predecessor satellites. To test the impact of FY-3E microwave observations on the operational YH4DVAR system, three NWP experiments were performed from June to August 2022. All-sky assimilation of FY-3E MWHS-2 observations in the YH4DVAR system has a neutral to positive impact on forecast skill. Assimilation of FY-3E MWTS-3 under clear-sky conditions provides little benefit to the YH4DVAR system in longer-range forecasts but improves the fit to the background of stratospheric temperature-sounding channels of ATMS and Advanced Microwave Sounding Unit-A (AMSUA) and radiosonde temperature observations over 150–400 hPa in the Northern Hemisphere and humidity observations over 100–850 hPa.

The Geostationary Interferometric Infrared Sounder (GIIRS) on board FengYun-4B (FY-4B), a Chinese second-generation hyperspectral infrared, enables the provision of critical data for forecasting high-impact weather events such as typhoons. To evaluate the reliability of FY-4B/GIIRS data, this study conducted three comparative assimilation trials for both Typhoon Gaemi and Typhoon Doksuri, assimilating observations from the Infrared Atmospheric Sounding Interferometer (IASI), Advanced Microwave Sounding Unit-A (AMSU-A), and FY-4B/GIIRS, respectively. Results demonstrate that the assimilation of GIIRS observations yields more stable forecasts of the wind field at 300 hPa and 500 hPa compared to AMSU-A and IASI, with biases within ±6 m/s relative to NCEP FNL data. However, GIIRS assimilation produces systematic underprediction of vertical velocity, whereas AMSU-A forecasts align more closely with reanalysis. For track forecasts, the GIIRS-assimilated trajectory exhibits closer alignment with observations than AMSU-A and IASI experiments, maintaining biases below 50 km throughout 48 h forecast period of Gaemi. This study provides valuable experience for the application of FY-4B/GIIRS data assimilation.

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.

In order to overcome the shortcomings of traditional spectral filtering method which cannot reflect the local characteristics of the spatial scales, a more efficient noise filtering technique in ensemble data assimilation is needed. In this paper, by introducing the improved threshold value and constraint parameter, a constrained wavelet threshold de-noising method (CWTD-NM) is designed to filter out the noise swamped in different scales. An ideal experiments group is implemented based on the two-dimensional barotropic vorticity equation. The best wavelet basis function (Db11) and the optimal decomposition level (five) are chosen firstly. The numerical results show that, since the wavelet coefficients are constrained by parameter α, the CWTD-NM can produce better filtering results with the smallest root mean square error (RMSE). Moreover, the filtering accuracy of 10 ensemble samples’ variances of the CWTD-NM is equivalent to the result estimated directly from 80 ensemble samples, while the running time is only about 1/7 of the latter. It also shows that Peak Signal to Noise Ratio (PSNR) performs well, and has a good corresponding relationship to the minimum RMSE.