This study developed a deep-learning-based model to retrieve sea surface hurricane winds from synthetic aperture radar (SAR) imagery. We introduce the essential idea, residual learning, of the Residual Net into the artificial neural network and design a deep cross-layer concatenation network (DCCN). The model inputs include SAR measured physical parameters in backscattering energy, the texture feature represented by the gray level co- occurrence matrix (GLCM), and the morphological hurricane feature. We collected 45 satellite SAR images from Sentinel-1 (S1) over hurricane conditions. These images were divided into 39 and six for model development and independent testing. A total of 16 127 wind samples acquired from 39 SAR images and simultaneously measured by the stepped frequency microwave radiometer (SFMR) were collected as model tuning datasets, among which 80% and 20% were used for training and validation. Our validation results show that the deep-learning-based model achieved a correlation coefficient (CORR) and root-mean-square error (RMSE) of 0.98 and 1.72 m/s for wind speeds up to 75 m/s. We further applied the model to six independent SAR images. The model significantly outperformed two existing geophysical algorithms and one backpropagation neural network (BPNN) algorithm with the RMSE is 2.61 m/s and a CORR of 0.95. Moreover, statistical analysis in different wind speed regimes indicates that our model shows a stable performance improvement than comparable algorithms. The RMSE decreases 10%–70%, especially the reduction of RMSE is more than 45% at high wind speed (>42 m/s). Furthermore, adding an independent rainfall estimate to the deep-learning model further enhanced the wind retrieval algorithm.

The Fusion of Physical, Textural, and Morphological Information in SAR Imagery for Hurricane Wind Speed Retrieval Based on Deep Learning

The existing synthetic aperture radar (SAR) ship datasets have an imbalanced number of inshore and offshore ship targets, and the number of small, medium and large ship targets differs greatly. At the same time, the existing SAR ship detection models in the application have a huge structure and require high computing resources. To solve these problems, we propose a SAR ship detection model named mask efficient adaptive network (MEA-Net), which is lightweight and high-accuracy for imbalanced datasets. Specifically, we propose the following three innovative modules. Firstly, we propose a mask data balance augmentation (MDBA) method, which solves the imbalance of sample data between inshore and offshore ship targets by combining mathematical morphological processing and ship label data to greatly improve the ability of the model to detect inshore ship targets. Secondly, we propose an efficient attention mechanism (EAM), which effectively integrates channel features and spatial features through one-dimensional convolution and two-dimensional convolution, to improve the feature extraction ability of the model for SAR ship targets. Thirdly, we propose an adaptive receptive field block (ARFB), which can achieve more effective multi-scale detection by establishing the mapping relationship between the size of the convolution kernel and the channel of feature map, to improve the detection ability of the model for ship targets of different sizes. Finally, MEA-Net is deployed on the Jeston Nano edge computing device of the 2 GB version. We conducted experimental validation on the SSDD and HRSID datasets. Compared with the baseline, the AP of MEA-Net increased by 2.18% on the SSDD dataset and 3.64% on the HRSID dataset. The FLOPs and model parameters of MEA-Net were only 2.80 G and 0.96 M, respectively. In addition, the FPS reached 6.31 on the Jeston Nano, which has broad application prospects.

/pdf/mea-net-a-lightweight-sar-ship-detection-model-for-t7oc7zw4.pdf

MEA-Net: A Lightweight SAR Ship Detection Model for Imbalanced Datasets

This study developed a deep-learning-based model to retrieve sea surface hurricane winds from synthetic aperture radar (SAR) imagery. We introduce the essential idea, residual learning, of the Residual Net into the artificial neural network and design a deep cross-layer concatenation network. The model inputs include SAR measured physical parameters in backscattering energy, the texture feature represented by the grey level co-occurrence matrix, and the morphological hurricane feature. We collected 45 satellite SAR images from Sentinel-1 over hurricane conditions. These images were divided into 39 and 6 for model development and independent testing. A total of 16,127 wind samples acquired from 39 SAR images and simultaneously measured by the Stepped Frequency Microwave Radiometer were collected as model tuning datasets, among which 80% and 20% were used for training and validation. Our validation results show that the deep-learning-based model achieved a correlation coefficient (CORR) and root-mean-square error (RMSE) of 0.98 and 1.72 m/s for wind speeds up to 75 m/s. We further applied the model to six independent SAR images. The model significantly outperformed two existing geophysical algorithms and one backpropagation neural network algorithm with the RMSE is 2.61 m/s and a CORR of 0.95. Moreover, statistical analysis in different wind speed regimes indicates that our model shows a stable performance improvement than comparable algorithms. The RMSE decreases 10%~ 70%, especially the reduction of RMSE is more than 45% at high wind speed (> 42 m/s). Furthermore, adding an independent rainfall estimate to the deep-learning model further enhanced the wind retrieval algorithm. 

Tropical cyclones (TC) transfer kinetic energy to the upper ocean and thus accelerate the ocean mixed layer (OML) currents. However, the quantitative link between near-surface currents and high wind speeds, under extreme weather conditions, remains poorly understood. In this study, we use multi-mission satellites and drifting-buoy observations to investigate the connections between TC surface winds and currents, including their spatial distribution characteristics. Observed ageostrophic current speeds in the OML increase linearly with wind speeds (for the range 20–50 m/s). The ratios of the ageostrophic current speeds to the wind speeds are found to vary with TC quadrants. In particular, the mean ratio is around 2% in the left-front and left-rear quadrants with relatively small variability, compared to between 2% and 4% in the right-front and right-rear quadrants, with much higher variations. Surface winds and currents both exhibit strong asymmetric features, with the largest wind speeds and currents on the TC right side. In the eyewall region of Hurricane Igor, high winds (e.g., about 47 m/s) induce strong currents (about 2 m/s). The directional rotations of surface winds and currents are resonant and dependent on the location within the storm. Wind directions are approximately aligned with current directions in the right-front quadrant; a difference of about 90° occurs in the left-front and left-rear quadrants. The directional discrepancy between winds and currents in the right-rear quadrant is smaller. Reliable observations of the wind-current relation, including asymmetric features, support published theories developed in idealized numerical experiments to explain the upper ocean response to TCs. 

/pdf/observed-ocean-surface-winds-and-mixed-layer-currents-under-1v0bry3w.pdf

Observed Ocean Surface Winds and Mixed Layer Currents under Tropical Cyclones: Asymmetric Characteristics

Compared with co-polarized (HH/VV) normalized radar cross-section (NRCS) backscattered from the sea surface, there is no saturation phenomenon in cross-polarized (HV/VH) NRCS when wind speed is greater than about 20 m/s, so cross-polarized synthetic aperture radar (SAR) images can be used for high wind speed monitoring. In this work, a new geophysical model function (GMF) is proposed to describe the relation of the C-band cross-polarized NRCS with wind speed and radar incidence angle. Here, sixteen ScanSAR wide mode SAR images acquired by RADARSAT-2 (RS-2) under tropical cyclone (TC) conditions and the matching wind speed data from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the Stepped-Frequency Microwave Radiometer (SFMR) are collected and divided into datasets A and B. Dataset A is used for analyzing the effects of the wind field and radar incidence angle on the reference noise-removed cross-polarized NRCS, and for proposing the new GMF for each sub-swath of the SAR images, while dataset B is used to retrieve wind speed and evaluate the validity of the new GMF. The comparisons between the wind speeds retrieved by the new GMF and the collocated ECMWF and SFMR data demonstrate the excellent performance of the new GMF for wind speed retrieval. To analyze the universality of the new GMF, wind speed retrievals based on 32 Sentinel-1A/B (S-1A/B) extra-wide-swath (EW) mode images acquired under TC conditions are also compared with the collocated wind speeds measured by the Soil Moisture Active Passive (SMAP) radiometer, and the retrieved wind speeds have RMSE of 3.667 m/s and a bias of 2.767 m/s. The successful applications in high wind speed retrieval of different tropical cyclones again supports the availability of the new GMF.

/pdf/development-of-a-new-tropical-cyclone-strip-segment-2da72dpo.pdf

Development of a New Tropical Cyclone Strip Segment Retrieval Model for C-Band Cross-Polarized SAR Data

We present a relatively simple method to estimate tropical cyclone (TC) surface wind structure (34-, 50-, and 64-kt wind radii) and intensity [maximum wind speed (MWS)] from wind fields acquired from the L -band SMAP radiometer and C -band Sentinel-1A/B and RADARSAT-2 synthetic aperture radar (SAR) between 2015 and 2020. The radiometer and SAR-derived wind radii and MWS are systematically compared with the best-track estimates. The root-mean-square errors (RMSEs) of R34, R50, and R64 are 31.2, 21.8, and 17.0 nmi (1 nmi = 1.852 km) for radiometer, and 21.7, 16.5, and 16.3 nmi for SAR, respectively. These error values are smaller than the averaged best-track uncertainty estimates for the three wind radii. Compared with the best-track reports, the bias and RMSE for the MWS estimates are −0.2 m/s and 5.8 m/s for radiometer, and 4.4 m/s and 9.1 m/s for SAR, respectively. These results are for the wind speeds in the range of 17–80 m/s. For the two typical TCs (Lionrock and Noru) in the Northwest Pacific Ocean, our results show that a combination of the radiometer and SAR wind data acquired within a very short time interval has the potential to simultaneously obtain reasonable measurements of the wind radii and intensity parameters. Moreover, for a TC with a long lifecycle, such as Typhoon Noru, we demonstrate that the high-resolution and multitemporal synergistic observations from SAR and radiometer are valuable for studying fine-scale features of the wind field and characteristics of wind asymmetry associated with intensity change, as well as the evolution of TC surface wind structure and intensity.

/pdf/estimating-tropical-cyclone-wind-structure-and-intensity-qsdquzgc1h.pdf

Estimating Tropical Cyclone Wind Structure and Intensity From Spaceborne Radiometer and Synthetic Aperture Radar

Estimation of maximum wind speed associated with tropical cyclones (TCs) is crucial to evaluate potential wind destruction. The Holland B parameter is the key parameter of TC parametric wind field models. It plays an essential role in describing the radial distribution characteristics of a TC wind field and has been widely used in TC disaster risk evaluation. In this study, a backpropagation neural network (BPNN) is developed to estimate the Holland B parameter (Bs) in TC surface wind field model. The inputs of the BPNN include different combinations of TC minimum center pressure difference (Δp), latitude, radius of maximum wind speed, translation speed and intensity change rate from the best-track data of the Joint Typhoon Warning Center (JTWC). We find that the BPNN exhibits the best performance when only inputting TC central pressure difference. The Bs estimated from BPNN are compared with those calculated from previous statistical models. Results indicate that the proposed BPNN can describe well the nonlinear relation between Bs and Δp. It is also found that the combination of BPNN and Holland’s wind pressure model can significantly improve the maximum wind speed underestimation and overestimation of the two existing wind pressure models (AH77 and KZ07) for super typhoons.

Estimating the Key Parameter of a Tropical Cyclone Wind Field Model over the Northwest Pacific Ocean: A Comparison between Neural Networks and Statistical Models

Jie Tang

Papers

Estimating Tropical Cyclone Wind Structure and Intensity From Spaceborne Radiometer and Synthetic Aperture Radar

Estimating the Key Parameter of a Tropical Cyclone Wind Field Model over the Northwest Pacific Ocean: A Comparison between Neural Networks and Statistical Models