Chaogang Guo

Tropical cyclones (TCs) characterized by extreme wind speeds present severe hazards to human life and infrastructure. Spaceborne synthetic aperture radar (SAR) has emerged as a critical observational tool for TCs, owing to its all‐weather, high‐resolution imaging capabilities. Nevertheless, retrieving sea surface winds from satellite SAR remains challenging, as radar backscatter is modulated not only by local wind dynamics but also by non‐Bragg scattering mechanisms. Heavy precipitation within TCs further exacerbates errors in normalized radar cross section measurements. In this study, multi‐scene Sentinel‐1 SAR observations from multiple TCs during 2016–2020 were collected and collocated with stepped‐frequency microwave radiometer measurements to construct a dual‐polarization scattering feature data set. To address class imbalance, resampling was applied exclusively within the training data of each cross‐validation fold. Feature selection was conducted using correlation analysis combined with maximum relevance–minimum redundancy, also implemented in a storm‐independent manner. Three machine learning models were trained for comparison using a random 7:3 data split, among which the XGBoost model exhibited superior performance and was designated as the C‐band dual‐polarization wind speed model (CDP_WS1). It achieved a root mean square error (RMSE) of 2.65 m/s and 3.44 m/s on the training and test sets, respectively, maintaining stable performance across wind speeds up to 70 m/s. Furthermore, to rigorously assess storm‐independent generalization, a Leave‐One‐Storm‐Out (LOSO) cross‐validation was conducted across 14 TCs, yielding a RMSE of 4.10 m/s. Independent validation on two representative hurricanes further confirmed its robustness and accuracy.

Spaceborne synthetic aperture radar (SAR) is widely used in ocean wave monitoring due to its high resolution and all-weather observation capability. Rainfall distorts sea surface wave spectra and SAR imagery, yet current significant wave height (SWH) retrieval models do not account for the influence of rainfall. To address this gap, we construct a collocated dataset of 9,245 Sentinel-1 WV mode images and Copernicus Marine Environment Monitoring Service (CMEMS) SWH data. A dual-branch convolutional neural network (DB-CNN) is proposed for SWH retrieval. An Inception v3 model is employed to classify rainfall intensity grades from SAR images, which are then fed as auxiliary input to the DB-CNN model. Experimental results show that incorporating rainfall correction reduces the root mean square error (RMSE) of SWH retrieval from 0.48 m to 0.43 m (10% improvement). Independent validation using Jason-3 and SARAL data confirms the robustness of the model, it significantly reduces the estimation bias.

Spaceborne synthetic aperture radar (SAR) is gradually being applied to hurricane observation because of its all-weather, high-resolution observation capability. In particular, the retrieval of rain rate using SAR images holds significant scientific and practical importance. However, accurately retrieving rain rate over the sea surface, particularly for high rain rate events under hurricane conditions, remains a significant challenge. The study proposes a new method for rain rate retrieval from hurricane SAR images. We've developed a cascaded feedforward neural network (CFNN) model that based on Sentinel-1's double-polarized C-band SAR images of 46 hurricanes to retrieve rain rate under hurricane conditions. In order to overcome the problem of local optimal solution of neural network, genetic algorithm is employed for optimized model parameter selection. Preliminary results indicate that this approach not only enhances the neural network's iteration speed but also improves its prediction accuracy. Compared with the rain rate of the Stepped-Frequency Microwave Radiometers (SFMR), the RMSE of retrieved rain rate is 3.05 mm/hr and the correlation coefficient is 0.88. Furthermore, we independently verify rain rate during hurricane Douglas and compared with Global Precipitation Mission (GPM) 2ADPR rain rate product, the results demonstrate that our model can effectively retrieve rain rate in the range of 0∼60 mm/hr under hurricane conditions. The encouraging results prove the feasibility of the method in SAR rain rate retrieval.

The technology of retrieving sea surface wind field from spaceborne synthetic aperture radar (SAR) is increasingly mature. However, the retrieval of the sea surface wind field related to the precipitation effect is still facing challenges, especially the strong precipitation related to extreme weather such as tropical cyclone will cause the wind speed retrieval error to exceed 10 m/s. Semantic segmentation and weak supervision methods have been used for SAR rainfall recognition, but rainfall segmentation is not accurate enough to support the correction of wind field retrieval. In this article, we propose to use deep learning to classify the rainfall grades in SAR images, and combine the rainfall correction model to improve the retrieval accuracy of sea surface wind speed. To overcome the challenge of limited training samples, the transfer learning method in fine-tune is adopted. Preliminary results demonstrate the effectiveness of this deep learning methodology. The model classifies rain and no-rain images with an accuracy of 96.2%, and classifies rainfall intensity grades with an accuracy of 86.2%. The rainfall correction model with SAR rainfall grade identified by convolution neural network reduces the root-mean-square error of retrieved wind speed from 3.83 to 1.76 m/s. The combination of SAR rainfall grade recognition and rainfall correction method improves the retrieval accuracy of SAR wind speed, which can further promote the operational application of SAR wind field.

Sea surface wind streak is one of many geophysical phenomena in synthetic aperture radar (SAR) images, which is often used to obtain sea surface wind direction. At present, the recognition of wind streaks mainly depends on artificial experience, and the recognition efficiency and accuracy are not high. In this study, the transfer learning based convolutional neural network architecture of Inception v3 was introduced to the recognition of sea surface wind streaks. Four categories of geophysical phenomena imaged by gaofen-3 (GF-3) SAR from 2019 to 2020 were chosen for retraining of the full pre-retrained Inception v3 model. Then, we use the retrained model to identify the wind streak of GF-3 in 2018 and use it to retrieve the sea surface wind direction. The results show that the transfer learning method is effective. The recognition accuracy of the model can reach 92.0% and 95.2% after data augmented. Compared with the reanalysis data of european centre for medium-range weather forecasts (ECMWF), the root mean square error of the retrieved wind direction is 9.12, which further verifies the ability of the training model to identify wind streaks.