Sea ice monitoring is crucial for climate studies, navigation safety, and sustainable management of Arctic regions. The introduction of the AI4Arctic Sea Ice Challenge dataset, which provides standardized multisource remote sensing data, has significantly facilitated the advancement of automated sea ice mapping techniques. This study introduces the Polarization-Aware CrossGate U-Net (PA-CG-U-Net) model with cross attention designed specifically for sea ice classification using this dataset. The proposed architecture integrates dual-polarization SAR images (HH and HV) and coarse-resolution microwave radiometric data (AMSR2) through separate encoders and cross-attention gates to capture polarization-specific features. The PA-CG-U-Net model reduced classification errors compared to the U-Net model and improved the segmentation accuracy for sea ice concentration (SIC), stage of development (SOD), and floe size (FLOE). Quantitative evaluations demonstrated consistent performance gains, with improved robustness indicated by lower variance across multiple training runs. Visual analyses further confirmed that PA-CG-U-Net produced less noisy classifications, which highlights the effectiveness of polarization-specific encoding and attention mechanisms.

This study performs the first multiclass sea ice classification using global navigation satellite system (GNSS) reflectometry (GNSS-R) data from the Fengyun-3E (FY-3E) satellite. A principal component analysis (PCA) based denoising scheme for GNSS-R delay Doppler maps (DDMs) is proposed. The newly proposed processing methods are applied to FY-3E satellite tracks to retain the most significant features of the data. After removing noise from the DDMs using PCA, DDM classification features are calculated using methods established in the literature. An extremely randomized trees (ET) machine learning (ML) classifier was used to classify the sea ice using the denoised classification observables. The classification results are validated using labels derived from National Snow and Ice Data Center (NSIDC) sea ice concentration data. The output classification accuracy was 87.14% with a 0.7834 Kappa coefficient. Per class F1 scores were 89.72%, 85.53%, and 81.34% for first-year ice (FYI), multi-year ice (MYI), and thin ice (TI), respectively. The proposed method produced improvements across all performance metrics compared to classification without PCA based observables and denoising.

Research on estimating sea ice concentration (SIC) from synthetic aperture radar (SAR) data using convolutional neural networks (CNNs) has been widely reported. However, the presence of speckle noise in dual-polarization SAR signals and confusion at ice-water boundaries complicates accurate density estimation, often leading to significant underestimations of SIC. Recently, diffusion models (DMs) have shown significant success in various remote sensing tasks, demonstrating their potential to address these challenges. However, applying DMs directly to SIC estimation leads to singularity issues, hindering the accuracy of results. Additionally, directly incorporating conditional images can cause denoising models to overlook the differences between conditional and noise information. We introduce a conditional denoising diffusion probabilistic model (DiffSIC) that can fundamentally resolve the singularity problem by reweighting the loss function. We designed a U-shaped architecture that integrates conditional information, time steps, and noise information for SIC estimation. Extensive experiments conducted on the AI4Arctic dataset indicate that the proposed DiffSIC framework achieves a coefficient of determination (R2) of 90.959% and a root mean square error (RMSE) of 8.632%, demonstrating the effectiveness and potential of diffusion models in the task of SIC estimation.

This study introduces a method for filling the data gaps in the International Soil Moisture Network (ISMN) by soil moisture (SM) estimated using data from the Cyclone Global Navigation Satellite System (CYGNSS). The estimation process leverages the random forest (RF) algorithm, incorporating CYGNSS-derived products along with soil and surface parameters as input features. This research was conducted based on the daily SM data from the ISMN for the entire years of 2019 and 2020, which served as training and test datasets. Comparison experiments were performed to highlight the limitations of existing methods and SM products for gap filling in ISMN SM data. Subsequently, the optimal retrieval model was deployed to estimate SM for the duration of the study, thereby filling the gaps within the ISMN dataset. The SM results after gap filling showed strong consistency with measured SM, achieving an R-squared ($R^{2}$) of 0.7930 and a root-mean-square error (RMSE) of 0.0492 cm3/cm3. These results indicate that CYGNSS-based SM inversion is a promising approach to enhance the completeness of the ISMN dataset.

This study proposes the use of a per-track normalization scheme and ancillary temporal and temperature data to improve the performance of global navigation satellite system reflectometry (GNSS-R) based sea ice classification. GNSS-R data and sea ice type labels were provided by TechDemoSat-1 (TDS-1) and National Snow and Ice Data Center (NSIDC) datasets, respectively. The delay Doppler maps (DDMs) from TDS-1 were used to compute DDM average (DDMA) observable values to which the proposed per-track normalization scheme was applied. The transformed observables with ancillary time and temperature data were provided to a random forest (RF) machine learning classifier to perform the final sea ice classifications. The purpose of this proposed method was to standardize power measurements across different TDS-1 tracks and to compensate for seasonal changes in sea ice that complicate sea ice classification. When using TDS-1 data within the temporal and geographic scope of this study, the method reported a testing accuracy of 84.86% and per-class F1 scores of 90.11%, 64.20%, and 81.59% for first-year ice (FYI), multi-year ice (MYI), and thin ice (TI), respectively. The proposed method was compared with existing methods of GNSS-R based sea ice classification and demonstrated improved or comparable performance. The proposed method shows promise, as it provides good performance while demonstrating generalizability to geographic and temporal data selection, and requires less data to train to a level comparable to established methods.

Recent advances in Cyclone Global Navigation Satellite System (CYGNSS) data have significantly improved the extraction of monthly surface water fraction (SWF), with neural networks being widely used for large-scale water body mapping based on GNSS-R signals. However, inherent noise in CYGNSS signals, such as multipath effects and interference, presents substantial challenges to the accuracy of SWF estimation. Diffusion models, an emerging class of generative deep learning techniques, have shown remarkable capabilities in capturing complex data distributions. By leveraging an iterative process of noise addition and removal, these models demonstrate significant advantages in processing low signal-to-noise ratio data, offering a novel methodology for precise SWF estimation from CYGNSS data. This study introduces DiffWater, a framework designed to address the unique characteristics of CYGNSS data and systematically explore the applicability of conditional diffusion models for remote sensing tasks. Utilizing a composite reference dataset, which includes the Global Surface Water (GSW) dataset and the Global Surface Water Dynamics (GLAD) dataset as training targets, DiffWater enhances the objectives of conditional diffusion models by integrating advanced conditional feature extractors and implementing multi-level fusion of conditional and temporal features, thereby achieving significant improvements in SWF estimation performance. Comprehensive experimental evaluations on the reference dataset demonstrate that DiffWater achieved the best performance, with a root mean square error (RMSE) of 4.987% and a correlation coefficient (R) of 0.946. Compared to state-of-the-art SWF estimation methods, the proposed approach demonstrated significant improvements in both quantitative and qualitative results.

Accurate global ocean surface wind speed estimation is crucial for weather forecasting and maritime transportation. Traditional retrieval methods based on Global Navigation Satellite System Reflectometry (GNSS-R) data and deep learning techniques often struggle to capture the complex, nonlinear relationships between GNSS-R observables and wind speed. This challenge is further exacerbated by imbalanced data distributions, which lead to significant underestimation under high wind conditions (15 - 30 m/s, here). To mitigate these issues, this study proposes a novel Physics-Informed Attention-Aided Convolutional Neural Network (PA-CNN). The proposed model incorporates an attention mechanism into the CNN architecture to adaptively focus on the most informative features. In addition, geophysical principles related to GNSS-R signal scattering are integrated with data-driven learning, improving both the interpretability and generalization capability of the network. Moreover, a spatial-temporal smoothing post-processing step is designed to enhance consistency in wind speed retrieval. Extensive experiments using Cyclone Global Navigation Satellite System (CYGNSS) datasets demonstrate that PA-CNN with smoothing outperforms existing deep learning approaches, achieving an overall root mean square difference (RMSD) of 1.38 m/s compared with ERA5 reanalysis data and 1.56 m/s against buoy measurements, highlighting the potential of combining physics-informed modeling with advanced deep-learning techniques for improved ocean surface wind speed retrieval.

Global Navigation Satellite System Reflectometry (GNSS-R) plays a crucial role in sea surface wind speed measurement, and convolutional neural networks (CNNs) have been a widely used method for wind speed retrieval from GNSS-R data. However, CNNs have limitations in global feature extraction due to the fixed convolutional kernels. Moreover, current studies on wind speed retrieval from GNSS-R data exhibit significant overfitting at low wind speed and underestimation at high wind speed due to the extremely imbalanced data distribution. To address these issues, a hybrid CNN-Transformer Network (CTN) with a weighted mean square error (MSE) loss is proposed in this study. Specifically, the designed CTN incorporates CNN and transformer encoder blocks to capture local and global features, with a dedicated feature fusion module to integrate these features. In addition, a novel weighted MSE loss function is designed to tackle the issue of imbalanced data distribution and enhance the estimation accuracy at high wind speeds. The proposed method is validated on Cyclone GNSS (CYGNSS) data and demonstrates improved performance compared with other machine learning algorithms. It achieves a root mean square difference (RMSD) of 1.417 m/s compared to the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis data, and 1.620 m/s compared to the buoy data from National Data Buoy Center (NDBC). Notably, the proposed model trained with the weighted MSE loss function shows an improvement of 20.3% in RMSD for samples with wind speeds exceeding 15 m/s, highlighting the effectiveness of the designed loss function in handling high wind speed data.

Accurate estimation of ocean surface wind speed is crucial for marine meteorology, ocean engineering, and navigation safety. In this study, the WSTCNN method, which combines the wavelet scattering transform (WST) and a convolutional neural network (CNN), is proposed to estimate wind speed from X-band marine radar data. The WSTCNN method begins by applying a preprocessing technique to the raw radar images to reduce noise and enhance data quality. Then, WST is applied to the processed radar images to multi-scale, translation-invariant, and noise-robust features that reflect the patterns of wind-driven sea surface motion. These extracted features are then fed into the CNN network, which is trained to establish a mapping between the extracted features and the corresponding wind speed values. The proposed method is evaluated on two radar datasets collected under diverse conditions. The first dataset was collected using a shipborne Decca radar in an open sea region approximately 300 km off the coast of Halifax, NS, Canada, while the second was collected using a shore-based Koden radar in Guadalupe Dunes, CA, USA. Both datasets include radar data obtained in rain-free and rainy conditions, enabling a comprehensive analysis of the method’s robustness under varying environmental influences. To validate the effectiveness of the WSTCNN method, existing wind speed estimation approaches, including support vector regression (SVR) and a traditional CNN model, were applied for comparison. The results demonstrate that WSTCNN achieves superior estimation accuracy under both rainy and rain-free conditions, highlighting its robustness and adaptability across varying environmental scenarios.

Accurate mapping of oil concentrations is essential for effective response to oil spill emergencies. The complexity of microwave scattering over oil-contaminated sea surfaces poses substantial challenges for synthetic aperture radar (SAR) applications, primarily due to the limited understanding of non-Bragg scattering mechanisms. This knowledge gap restricts the development of robust retrieval algorithms for quantifying oil spill concentrations. To address this issue, a novel retrieval approach is proposed based on the co-polarization difference (PD), which is independent of non-Bragg scattering. The influence of oil on the sea surface is attributed to two dominant factors: suppression of short gravity-capillary waves and reduction in the effective dielectric constant. By analyzing SAR imagery of oil spills with varying concentrations, it is found that the damping effect of oil spills on small-scale waves can be predicted using the Marangoni damping model. Once the contribution of wave suppression to PD reduction is isolated, the residual PD variation is attributed to changes in the dielectric constant. Oil concentration is then retrieved by comparing the PD of each pixel within the contaminated area to a theoretical PD lookup table. The proposed method is validated using simulated SAR datasets representing different oil concentrations and subsequently applied to SAR data acquired during the Deepwater Horizon oil spill in the Gulf of Mexico.

Fully supervised deep learning approaches have demonstrated impressive accuracy in sea ice classification, but their dependence on high-resolution labels presents a significant challenge due to the difficulty of obtaining such data. In response, our weakly supervised learning method provides a compelling alternative by utilizing lower-resolution regional labels from expert-annotated ice charts. This approach achieves exceptional pixel-level classification accuracy by introducing regional loss representations during training to measure the disparity between predicted and ice chart-derived sea ice type distributions. Leveraging the AI4Arctic Sea Ice Challenge Dataset, our method outperforms the fully supervised U-Net benchmark in mapping resolution and class-wise accuracy, marking a significant advancement in automated operational sea ice mapping.

An algorithm for extracting sea surface current information from X-band nautical radar image sequences is presented in this paper. The angular frequencies corresponding to the wave numbers are first extracted from the radar image sequence, and then the Doppler shifts corresponding to every wave numbers are calculated. The Doppler shifts corresponding to wavevectors symmetric with respect to the origin of the wave number plane are used to estimate current speed and direction. Test using simulated data shows the root mean square error (RMSE) of current speed is 0.13 m/s and that of current direction is 1.4°. The results illustrate that the method is feasible with comparable accuracy to existing methods.

Global navigation satellite system reflectometry (GNSS-R) is an emerging remote sensing technology for sea surface wind speed measurement. GNSS-R captures the reflected signal from sea surface and generates Delay Doppler maps (DDM). Existing studies have demonstrated the effectiveness of convolutional neural network (CNN) in retrieving wind speed from GNSS-R. However, most of the existing CNN-based methods assign equal importance to each pixel in the DDM and cannot focus on more discriminative pixels. To address this issue, in this study, the attention mechanism is integrated and an attention-aided CNN (Att-CNN) is developed for sea surface wind speed estimation. The performance of Att-CNN is evaluated on the Cyclone GNSS (CYGNSS) data. Compared to conventional CNN without attention module, the proposed Att-CNN obtains a lower RMSD of 1.151 m/s with an improvement of 3.46%.

Due to the high false alarm rate and low positioning accuracy of compact high-frequency surface wave radar (HFSWR), plot-to-track association methods using kinematic parameters alone may not achieve satisfactory performance. In this letter, the course consistency of vessels is analyzed, and a plot-to-track association method using statistical course modeling is proposed. First, vessel course data sequences are obtained by applying a course estimation method to each track, and each course data sequence is modeled using the Incremental Gaussian Mixture Model (IGMM) by considering the measurement uncertainty. Next, the measurements within the association gate of each track are connected with the last plot respectively to determine instantaneous courses, and the membership probability set that the instantaneous courses belong to the established IGMM is obtained using Bayes’ rule. Subsequently, the obtained membership probability values and kinematic parameters of each measurement are incorporated to calculate the association cost, and a cost matrix is obtained for tracks with shared candidate plots in the overlapped region of their association gates. Finally, the Hungarian algorithm is applied to the cost matrix to obtain plot-to-track association results. Plot-to-track association experiments using both simulated and field data were conducted, and experiment results demonstrate that the optimal sub-pattern assignment distance in latitudes and longitudes obtained by the proposed method is 0.001° lower than that of the Nearest Neighbor Data Association method on average and achieves a competitive performance over the Joint Probability Data Association method but with running time being reduced by 0.48 seconds for each frame.

Compared to shore-based high-frequency surface wave radar (HFSWR), shipborne HFSWR can overcome the constraints of a fixed radar site and extend its detection range. However, the radar echo is influenced by the movement of the shipborne platform, which in turn affects the target monitoring performance of the shipborne HFSWR. In this article, the radar echo model for shipborne radar is introduced, and the Doppler frequency shifts for different signals are given. Then, the characteristics of vessel target echoes for shipborne HF radar under various motion conditions are analyzed. Subsequently, the characteristics of spread sea clutter and its impact on target monitoring under different motion conditions are investigated. Moreover, land clutter, which is often neglected for shore-based HFSWR, is also investigated. Considering the combined effect of clutter blind zones caused by sea clutter and land clutter, the target monitoring capability of shipborne HFSWR under different motion conditions is evaluated, and then, a target monitoring scheme is proposed. In the target monitoring scheme, different navigation scenarios are used to adjust the platform motion state depending on different detection targets. Low-speed navigation scenario is appropriate for the monitoring of moving targets, whereas high-speed navigation scenario is suitable for detecting stationary targets or vessel target initially submerged in nonspread sea clutter. Finally, the clutter extraction results from measured data under different motion conditions and their impact on target monitoring are analyzed, and target monitoring results are provided and validated using field data.

Global Navigation Satellite System Reflectometry (GNSS-R) is a novel remote sensing technique for global significant wave heights (SWHs) observation. Previous studies have illustrated the efficacy of deep learning methods in SWH retrieval from GNSS-R data. However, most of these methods rely on convolutional layers to extract features from delay Doppler maps (DDMs), facing the limitations imposed by the fixed receptive field. To address this issue, in this study, a transformer-based network called WaveTransNet is proposed for SWH retrieval from GNSS-R data. Specifically, the transformer encoder block is exploited to capture long-range dependencies from DDMs. In addition, an attention mechanism-aided ancillary parameters feature extraction branch is devised to extract discriminative features from ancillary parameters, including geometry-related and map-related parameters. The developed model is evaluated on the Cyclone Global Navigation Satellite System (CYGNSS) dataset, and the experimental results demonstrate its improved performance. Compared with the European Center for Medium-Range Weather Forecasts (ECMWF) reanalysis data, it achieves a root mean square difference (RMSD) of 0.443 and 0.444 m when National Data Buoy Center (NDBC) buoy data are used for evaluation.

A new method, which integrates multi-variable consisting of Soil Moisture (SM) Active Passive (SMAP)-derived SM and vegetation optical depth, the water seasonality, geolocation, digital elevation model, slope, and biomass as inputs and adopts the technique of Bootstrap Aggregation of Regression Trees (BARTs) is proposed for retrieving monthly surface water fraction (SWF) at a spatial resolution of 0.025° from Cyclone Global Navigation Satellite System (CYGNSS) data. The model is trained using Surface Water Microwave Product Series (SWAMPS) data with a coarser resolution of 25 km and then applied to CYGNSS data with an enhanced resolution of 0.025° to generate high-resolution water maps. The resulting CYGNSS SWF (CSWF) maps are evaluated by comparing them with other water data sources, namely SWAMPS, Global Surface Water (GSW), and Global surface water dynamics (GLAD), as well as ground measurements. A quadruple collocation analysis indicates that the CSWF results exhibit the lowest error variance among the four SWF datasets. Furthermore, additional testing with water level measurements demonstrates a strong correlation with station data and clear seasonal patterns. Notably, the CSWF estimates significantly improve spatial coverage compared to both optical data (GSW and GLAD) with enhanced spatial resolution and the coarser SWAMPS data. This study underscores the effectiveness and efficiency of CSWF estimates, highlighting their potential as a valuable complement to existing microwave- and optical-based surface water products.

This paper presented a novel significant wave height (SWH) estimation method, SWHFormer, which incorporates the Vision Transformer (ViT) to estimate SWH from X-band nautical radar images. Unlike traditional convolutional neural networks, the ViT model treats the input as a sequence, capitalizing on its attention mechanism to capture long-range dependencies, resulting in superior performance in capturing the complex patterns present in sea wave dynamics. The radar data undergo an image denoising routine, followed by patching, flattening, and embedding processes to form a sequence fed into the Transformer encoding module. The outputs from the encoder are then aggregated to derive the final regression result, i.e., SWH estimation. In order to evaluate the performance of SWHFormer, the dataset collected by a Decca radar aboard a free-navigating vessel is analyzed, both buoy and model-based data are utilized as ground truth. In this study, two traditional linear fitting methods, i.e., ensemble empirical mode decomposition (EEMD) and variational mode decomposition (VMD)-based approaches, and a recent deep learning algorithm, convolutional gated recurrent unit (CGRU) network are exploited for comparison with SWHFormer. It is found that the root mean square error (RMSE) of the estimated results using the proposed SWHFormer is decreased from 0.29 m, 0.26 m, and 0.18 m to 0.16 m after the temporal moving average, respectively, compared to the above three methods, when the buoy-measured SWH is served as ground truth. Besides, it is decreased from 0.30 m, 0.28 m, 0.16 m to 0.14 m, respectively, when the model-based SWH is employed as reference.

Hyperspectral remote sensing images contain abundant spectral information and have a broad application in various areas, such as precision agriculture and geological exploration. However, due to the existence of many bands and the correlation between these bands, it is easy to cause the curse of dimensionality. Therefore, how to efficiently extract spectral and spatial features should be addressed. In this paper, a novel and efficient sharpened cosine similarity operation is applied in hyperspectral images classification to enhance the ability of feature extraction. To validate the efficiency of the proposed method, experiments are conducted on a real hyperspectral dataset, e.g., the University of Pavia (UP). Quantitative and qualitative results demonstrate that sharpened cosine distance operation can efficiently extract discriminative features and achieve a better classification performance.

Algal bloom has become a serious environmental problem caused by the overpropagation of planktons in many water-bodies, and effective remote sensing methods for monitoring it are urgently needed. Global Navigation Satellite System (GNSS)-Reectometry (GNSS-R) has been developed rapidly these years. The reflected GNSS signals received by GNSS-R satellites carry information (e.g., the surface roughness) about the specular points. When algal bloom emerges, the water surface will turn smoother, which could be detected by GNSS-R. In addition, meteorological factors also perform a key role in the formation of algal bloom. In this article, a new machine learning aided GNSS-R algal bloom detection method with the auxiliary of meteorological data is established. This work employs the Cyclone GNSS (CYGNSS) data and the fifth generation of European Reanalysis data with the application of Random Under-Sampling Boost (RUS-Boost) algorithm. During the evaluation stage, the test True Positive Rate of 80%, overall accuracy of 84.1% and the Area Under (Receiver Operating Characteristic) Curve of 0.9 were achieved when all the considered GNSS-R observables and meteorological factors are involved. Meanwhile, the contribution of each meteorological factor was also evaluated.

The sensitivity of Cyclone Global Navigation Satellite System (CyGNSS) data to inland water bodies was well documented, however, its advantage over other sensors has seldom been reported. In this work, a semantic segmentation method is adopted for detecting inland water bodies using the CyGNSS data. The widely used LinkNet with the global attention mechanism (GAM) and atrous spatial pyramid pooling (ASPP), namely GA-LinkNet, is equipped to better extract water distributions. The performance comparison with an existing method and other deep networks proved the accuracy and effectiveness of this approach. Satisfactory agreement between the derived and referenced water masks was achieved, with the overall accuracy being 0.959 and 0.976, the mean intersection over union being 0.785 and 0.641, and the F1 scores being 0.879 and 0.781 for the Amazon and Congo regions, respectively. Furthermore, underestimation of water by the reference data was shown during evaluation, which proves the usefulness of the CyGNSS-derived water mask for improving the existing water mask products.

Hyperspectral images (HSIs) provide hundreds of continuous spectral bands and have been widely used for the fine identification of targets with similar appearances. In earlier studies, convolutional neural networks (CNNs) have been an effective method for HSIs classification due to their powerful feature extraction capabilities. Recently, self-attention-based vision transformer (ViT) architecture has been widely explored to fully represent global information. However, most existing transformer-based models primarily focus on global relationships and lack the ability to capture the multi-scale features which are crucial for HSIs classification. This limitation results in inferior performance for transformer-based methods compared to state-of-the-art CNN-based models. To solve this problem, a novel network called multi-scale neighborhood attention transformer (MSNAT) is proposed in this paper. Unlike previous transformer-based models, MSNAT emphasizes the neighborhood pixels within a local window size and extracts multi-scale spatial information by using different local window sizes. In addition, a spatial transformation module is integrated to generate optimized spatial input. The effectiveness of the proposed MSNAT is verified on three real hyperspectral datasets including University of Pavia (UP), University of Houston (UH), and University of Trento (UT). Experimental results demonstrate that the proposed MSNAT method outperforms both CNNs and existing transformer-based models, achieving state-of-the-art classification performance with an overall accuracy of 93.34%, 86.26%, and 96.63% on UP, UH, and UT, respectively. The source code will be available at https://github.com/xinqiao123/MSNAT.

Due to its low transmit power and reduced aperture size of a receiving antenna array, compact high-frequency surface wave radar (HFSWR) suffers from low detection probability, low positioning accuracy, and high false alarm rate. In a multi-target tracking scenario, similar kinematic parameters of adjacent targets raise challenges to the track-to-track association procedure. Taking the measurement uncertainty of compact HFSWR into consideration, a track-to-track association method based on maximum likelihood estimation (MLE) for T/R-R composite compact HFSWR is proposed. Firstly, a multi-target tracking algorithm is applied to plot data sequences acquired by both T/R monostatic and T-R bistatic radars to produce two track sets. Then, the measurement errors of range, azimuth, and Doppler velocity are calculated using the obtained radar track and corresponding AIS track data, and a Gaussian distribution model is derived through probability distribution fitting. Subsequently, likelihood functions are established using the obtained Gaussian distribution model to calculate the association cost of tracks respectively for T/R monostatic and T-R bistatic radars, and a cost matrix is obtained. Finally, the Jonker-Volgenant-Castanon (JVC) assignment algorithm is applied to the cost matrix to determine associated track-track pairs. Track-to-track association experiments using both simulated and field data were conducted, and the association performance of the proposed method is compared with that of Mahalanobis distance-based nearest neighbor (NN) method. Experimental results demonstrate that the proposed method can effectively resolve association ambiguity and achieve correct track-to-track association in track crossing and adjacent multi-target scenarios.

In this paper, sea ice classification on a remote sensing image given just a small number of labeled pixels is investigated. Effective sea ice classification is rendered from two aspects. First, a feature extraction method is developed. It extracts the context feature from a classification map. Second, an iterative learning paradigm is established. The labeled pixels are divided into two training subsets. At each iteration, the context feature for one subset is extracted from the classification map which is obtained subject to the other subset. Therefore, the two subsets mutually guide each other for updating the context feature in an iterative manner, which finally renders effective sea ice classification. The above paradigm is referred to as mutually guided contexts. The advantages of the new paradigm are two-fold. First, the context feature enriches the sea ice image representation in a general manner regardless of the types of raw image data. Second, the two training subsets keep providing different refined classification maps for each other such that the comprehensiveness of the context feature is recursively enhanced. Therefore, the paradigm of mutually guided contexts comprehensively characterizes the sea ice image representation for training and classification even when only small training data are available. Experiments validate the effectiveness of the mutually guided contexts for sea ice classification.

In recent years, kernel-based methods and Deep Learning (DL) models have become the two most successful Remote Sensing (RS) analysis techniques for various Earth observations, particularly hyperspectral images. However, kernel-based methods are generally considered shallow models and intrinsically inconsistent with end-to-end learning. On the other hand, end-to-end learning is one of DL models' essential features as it seems to be responsible for their proven higher performances. Nevertheless, kernel methods are based on rigid mathematical theory and can efficiently cope with high-dimensional data. This paper proposed a hybrid deep kernel model to benefit from both kernel-based methods and DL models. This novel deep kernel model, namely Convolutional Kernel Network (CKN), was applied to two benchmark hyperspectral image datasets. Moreover, the proposed hybrid method was compared to Support Vector Machine (SVM) classifiers with various kernel functions. The experimental results indicated that the CKN's outperforms SVM.

As a typical representative of the next generation SAR mission, the RADARSAT Constellation Mission (RCM) provides three C-band SAR satellites with shorter revisit time and broader spatial coverage, which will be widely used in various earth observation applications including sea ice sensing. The Sentinel-1 mission comprises two C-band SAR satellites with dual-polarized imaging capability, providing open and free data from the European Space Agency (ESA). Sea ice classification results of the two C-band SAR missions with a state-of-the-art convolutional neural network: Normalizer-Free ResNet (NFNet) are compared in this paper. HH, HV and cross-polarization ratio are extracted from the overlapping area of dual-polarized RCM and Sentinel-1 images acquired on similar dates. The sea ice classification results shows that RCM Medium Resolution 50m mode performs better than Sentinel-1 EW GRD Medium Resolution mode for the data used in this study.

In this work, an improvement for the quality of X-band radar images affected by rain is proposed, and a support vector regression (SVR)-based method is designed to further obtain the significant wave height (Hs). The first step is to implement the dehazing algorithm on the radar images influenced by rain. Then, SVR is employed to train the Hsregression model including two features, i.e., gray level co-occurrence matrix (GLCM) and signal-to-noise ratio (SNR) features, extracted from rainless data. Finally, Hscan be obtained from the trained model with these two features extracted from the dehazed images. Besides, two classical Hsestimation methods, i.e., ensemble empirical mode decomposition (EEMD)-and SNR-based regression algorithms are utilized for analyzing the effectiveness of the dehazing algorithm. Experiment results confirm that dehazing algorithm can substantially decrease the root-mean-square-error (RMSE) and biases of the estimated Hsand increase the correlation coefficients (CCs) between the results and buoy data. Furthermore, comparisons with the SNR- and EEMD-based regression algorithms incorporating the dehazing algorithm illustrate that RMSEs obtained from the proposed method are further reduced to 0.44 m from 0.49 m and 0.78 m, respectively.

The inland water bodies are critical components of ecosystems and hydrologic cycles. Thus, the water extent data are crucially important for hydrological and ecological studies. Due to its high temporal resolution, the Cyclone Global Navigation Satellite System (CYGNSS) has the potential for real-time inland water monitoring. In this letter, a high-resolution machine learning (ML) method for detecting inland water content using the CYGNSS data is implemented via the random undersampling boosted (RUSBoost) algorithm. The CYGNSS data of the year 2018 over the Congo and Amazon basins are gridded into$0.01^{\circ }\, \times \, 0.01^{\circ }$cells. The RUSBoost-based classifier is trained and tested with the CYGNSS data over the Congo basin. The data of the Amazon basin that is unknown to the classifier are then used for further evaluation. By only using the observables extracted from the CYGNSS data, the proposed technique is able to detect 95.4% and 93.3% of the water bodies over the Congo and Amazon basins, respectively. The performance of the RUSBoost-based classifier is also compared with an image processing-based inland water detection method. For the Congo and Amazon basins, the RUSBoost-based classifier has a 3.9% and 14.2% higher water detection accuracy, respectively.

This work presents a novel scheme to retrieve soil moisture (SM) from the Cyclone Global Navigation Satellite System (CyGNSS) data, which is accomplished by using a bagged regression trees (BRT) algorithm with the inputs being the CyGNSS-derived products, the corresponding geolocation, and associated climate type. This algorithm is validated with thein situhourly SM data acquired by China’s automatic SM observation stations throughout the year 2018. High consistency between the retrieved SM results and the measured SM is achieved, with a correlation coefficient of 0.86 and a root-mean-square error of 0.05 cm3/cm3. The results obtained in this work indicate that the proposed BRT-based method can effectively estimate SM from CyGNSS data in different scenarios of various station locations and climate types in a near real-time manner.

In the paper, a variational mode decomposition (VMD)-based method is proposed to estimate significant wave heights (Hs) from X-band marine radar images. Firstly, 10 intrinsic mode functions (IMFs) are decomposed from the selected radar sub-images with VMD. Then, a linear fitting method is conducted to estimateHsby using the sum of the amplitude modulation (AM) components extracted from the 6thto 9thIMFs. The radar data were collected from a ship at sea around 300 km from Halifax, NS, Canada. The real-timeHsdata were obtained by drifting Triaxys buoys around the moving vessel. Experiment results show that the proposed VMD-based linear fitting method generates improvement in theHsmeasurements, compared to the typical ensemble empirical mode decomposition (EEMD)-based linear fitting method, by reducing the root-mean-square error (RMSE) from 0.34 m to 0.32 m and increasing the correlation coefficient (CC) from 0.90 to 0.92 after using the moving average.

Spatial–temporal features are extracted from X-band marine radar backscatter image sequences via deep neural networks to estimate sea surface significant wave heights (SWHs). A convolutional neural network (CNN) is first constructed based on the pretrained GoogLeNet to estimate SWH using multiscale deep spatial features extracted from each radar image. Since the CNN-based model cannot analyze the temporal behavior of wave signatures in radar image sequences, a gated recurrent unit (GRU) network is concatenated after the deep convolutional layers from the CNN to build a convolutional GRU (CGRU)-based model, which generates spatial–temporal features for SWH estimation. Both the CNN and CGRU-based models are trained and tested using shipborne marine radar data collected during a sea trial off the East Coast of Canada, while simultaneous SWHs measured by nearby buoys are used as ground truths for model training and reference. Experimental results show that compared to the classic signal-to-noise ratio (SNR)-based method, both models improve estimation accuracy and computational efficiency significantly, with a reduction of RMSD by 0.32 m (CNN) and 0.35 m (CGRU), respectively. It is also found that under rainy conditions, CGRU outperforms SNR and CNN-based models by reducing the RMSD from around 0.90 to 0.54 m.

HF radars have been extensively used for current observation. However, wave measurement with HF radars is challenging mainly due to the limited measurable range of wave height using a single operating frequency. In order to obtain robust wave measurements in complex and various sea states, a wave inversion method is proposed for a multifrequency HF radar. In this method, the nondirectional wave spectrum is directly retrieved from the radar echoes collected at various frequencies (up to four), and then, the significant wave height and the mean wave period are obtained from the integration of the derived nondirectional wave spectrum. Simulation analysis is carried out to evaluate the performance of the proposed method for a case in four various radar frequencies. Then, the proposed method is applied to a three-day observation to validate its advantages by comparing the radar-estimated and WaveRider-measured nondirectional wave spectra. In addition, a 14-day dataset collected with an HF radar operating at 8.267 and 19.2 MHz during a super Typhoon event is selected for further validation via comparisons between the radar-estimated and the buoy-measured wave parameters. The results indicate that the agreement between them is reasonable, and the comparisons also demonstrate that the accuracy of the wave measurement using the proposed multifrequency method is better than that with a single frequency.

In this paper, a new scheme is proposed for simultaneously retrieving soil moisture (SM) and vegetation optical depth ($\tau$), solely from the Cyclone Global Navigation Satellite System (CyGNSS) data. This work is accomplished by employing two pre-trained neural networks as well as a brute-force searching. By adopting the proposed method, the posterior$\text{SM}/\tau$can be estimated merely using the CyGNSS data, free from other auxiliary data. Satisfactory agreements between the retrieved and referred$\text{SM}/\tau$illustrates the capability of CyGNSS as a new independent source for estimating pantropical SM and$\tau$.

In this paper, the accuracy of wave direction estimation from X-band marine radar images under different rain rates is analyzed, and a simple sub-image selection scheme is proposed to mitigate the rain effect. First, each radar image is divided into multiple sub-images, and the sub-images with relatively clear wave signatures are identified based on random-forest-based classification model. Then, wave direction is estimated by performing Radon transform on each valid sub-image. The shore-based X-band marine radar images, simultaneous rain rate data, as well as buoy-measured wave data collected on the West Coast of the United States are used to analyze the rain effect on wave parameters estimation accuracy and validate the proposed method. Experimental results show that the proposed sub-image selection scheme improves the estimation accuracy of wave direction under different rain rates, with reductions of RMSEs by 6.9°, 6.0°, 4.9°, and 1.0° for wave direction under rainless, light rain, moderate rain, and heavy rain conditions, respectively.

Summary Ships and icebergs are similar in size and intensity in SAR images, so it is difficult to distinguish them in remote sensing images. Deep learning is a technique based on neural networks, which has played an important role in image information processing. In order to address the challenge of ship and iceberg classification, we present a convolutional neural network (CNN) based classification method for iceberg and ship discrimination from Sentinel‐1 SAR images with different polarizations and incidence angles. The method is based on the fixed constant false alarm rate (CFAR) detector and the CNN model has three input channels, then the model was trained using parallel algorithm. The CNN is trained using 1443 images and tested using 161 images. The CNN model is also compared with support vector machine (SVM) and k nearest neighbors (kNN) using the same dataset. Comparison shows the CNN‐based method performs the best, and it achieved a validation accuracy of 96%.

Ship detection at the sea surface is important for improving human marine activities. Most existing ship detection methods for high-frequency surface wave radar (HFSWR) are based on peak and constant false alarm rate (CFAR) detection and require a coherent integration time (CIT) of several minutes. However, in such a long period, the target may not be stationary. To account for the nonstationary property, a time-frequency analysis (TFA)-based ship detection and direction finding (DF) method is proposed for HFSWR. Target ridges on the TF representation (TFR) of the echo data are detected first. Next, array snapshots are formed by sampling the extracted ridges and are used to estimate the direction of arrival (DOA). The processing results of the radar data collected at Dongshan, Fujian Province, China, show that the proposed method outperforms the CFAR method with both increased detection rates and decreased DF errors, especially under relatively low signal-to-noise ratio (SNR) scenarios.

In this paper, the sensitivity of Cyclone Global Navigation Satellite System (CYGNSS) data to precipitation is investigated. First, the soil moisture (SM) is estimated from the CYGNSS and Soil Moisture Active Passive (SMAP) data based on a three-layer model. Next, the correlation between the CYGNSS-derived global SM and the precipitation rate is analyzed. The CYGNSS data collected over the land surfaces within ±37° (latitude) during the whole year of 2018 are employed and the CPC Merged Analysis of Precipitation (CMAP) data are adopted. Experimental evaluation proves the sensitivity of CYGNSS-derived SM to precipitation, indicating possible applications of CYGNSS data for detecting rainfall events and estimating precipitation.

In this paper, an approach for estimating soil moisture (SM) from Cyclone Global Navigation Satellite System (CYGNSS) data is developed. Here, a three-layer model of air, vegetation cover, and soil is proposed. In application, the surface reflectivity along with its statistics are derived from the CYGNSS data and the ancillary vegetation opacity data are obtained from Soil Moisture Active Passive (SMAP). These variables are adopted for estimating SM using the devised linear regression function. Through comparing with the reference SM data obtained from SMAP, CYGNSS-derived SM demonstrates its satisfactory accuracy and plausible global coverage. The achieved results prove CYGNSS as an efficient complementary tool for global daily SM sensing.

We explore the use of convolutional neural networks (CNNs) for filling voids in digital elevation models (DEM). We propose a baseline approach using a fully convolutional network to predict complete from incomplete DEMs, which is trained in a supervised fashion. We then extend this to a shadow-constrained CNN (SCCNN) by introducing additional loss functions that encourage the restored DEM to adhere to geometric constraints implied by cast shadows. At the training time, we use automatically extracted cast shadow maps and known sun directions to compute the shadow-based supervisory signal in addition to the direct DEM supervision. At the test time, our network directly predicts restored DEMs from an incomplete DEM. One key advantage of our SCCNN model is that it is characterized by both CNN data inference and geometric shadow cues. It thus avoids data restoration that may violate shadowing conditions. Both our baseline CNN and SCCNN outperform the inverse distance weighting (IDW)-based interpolation method, with the shadow supervision enabling SCCNN to obtain the best performance.

A common source of interference in high-frequency (HF) surface wave (HFSW) radars is radio frequency interference (RFI). Its existence inhibits the detection performance of HFSW radars since its amplitude can mask the sea echoes. On the basis of the analysis of RFI characteristics, a new RFI mitigation algorithm based on inverse temporal windowing and complex empirical mode decomposition (CEMD) is proposed in this letter. In this method, echoes containing RFI are decomposed into a number of intrinsic mode functions (IMFs) via CEMD and then the inverse temporal windowing technique is applied to each IMFs in the time domain. Test results show that the proposed method outperforms the conventional method in simulated and practical conditions and can effectively mitigate RFI without losing sea echoes.

In this article, an unsupervised clustering-based method for identifying rain-contaminated and low-backscatter regions in X-band marine radar images is presented. Rain blurs the wave signatures of radar images, and low-backscatter images caused by calibration errors or too-low wind speed contain little or no wave signatures. In both cases, ocean surface parameter measurement using X-band marine radar will be negatively affected. Four types of features can be extracted based on the distinct difference in texture and pixel intensity distribution between rain-free, rain-contaminated, and low-backscatter regions. Features extracted from each pixel are combined into a feature vector and mapped onto a 10×10-neuron self-organizing map (SOM). Then, the hierarchical agglomerative clustering algorithm is introduced, which clustered those neurons into three types (i.e., rain-free, rain-contaminated, and low-backscatter). The method is validated using the shipborne marine radar data collected on the East Coast of Canada. The good agreement between the pixel-based clustering results and manually segmented reference images indicates that both rain-contaminated and low-backscatter regions can be identified effectively using the proposed method.

Since rain alters the histogram pattern of radar images, rain-contaminated radar data can be identified. In this article, a support vector machine (SVM)-based method for rain detection using X-band marine radar images is presented. First, the normalized histogram bin values for each image are extracted and combined into feature vector. Then, SVMs are employed to classify between rain-free and rain-contaminated images. Radar images and simultaneous rain rate data collected from a sea trial in North Atlantic Ocean are utilized for model training and testing. Comparison with the zero pixel percentage (ZPP) threshold method shows that the SVM-based method obtains higher detection accuracy, with 98.4% for the Decca radar data and 99.7% for the Furuno radar. It is also found that as the total number of bins does not significantly affect detection accuracy, the proposed method can be applied to different radar systems directly with a suitable number of bins. In addition, compared to the ZPP threshold method, the SVM-based method proves to be more robust even with limited training samples.

Small-aperture compact high-frequency surface wave radar (HFSWR) suffers from low azimuth accuracy for target detection due to its wide beamwidth. Multitarget tracking (MTT) algorithms, when applied to the raw target detection data of HFSWR, fail to effectively filter the target azimuths, and thus, resulting in inaccurate target tracks and courses. In this article, a vessel azimuth and course joint re-estimation method by exploring Doppler velocity and the information accumulated from consecutive observations is presented. It begins with applying an MTT algorithm to a measured target states data sequence acquired by HFSWR to establish initial target tracks, from which the measured range, azimuth, and radial velocity data sequences are obtained. Then, the azimuth trend is extracted from the obtained azimuth data sequence as roughly corrected azimuth estimates, with which the target locations are roughly corrected. Subsequently, target speeds and initial courses are estimated based on the roughly corrected location data sequence, followed by a data selection procedure based on proposed control parameter rules to select the qualified data for calculating the projected angles in terms of speed and direction, separately. Eventually, the target azimuth data sequence is further refined using a linear azimuth error model, whose parameters are obtained by minimizing the difference between the projected angles using a constrained optimization method. Experimental results from field data demonstrate that the proposed method can estimate the target azimuths with significantly improved accuracy. The deviations of the corrected target locations are considerably reduced, and the accuracy of course estimation is enhanced.

Compact high-frequency surface wave radar has been widely applied to the measurement of sea surface current, but its accuracy and direction resolution of wave parameter estimation are always limited due to the wide beam of the antenna. In this article, a novel wave-height mapping method based on the second-order harmonic peak (SHP) of radar Doppler spectra is proposed to address this concern. The characteristic of the SHP at the Doppler frequency of$\sqrt {2}$times the Bragg frequency is studied through the theoretical derivation and numerical simulation. A relationship between the ratio ($R$) of the SHP power to the Bragg peak power and significant wave height ($H_{s}$) is derived. Furthermore, the$R$–$H_{s}$model is improved by incorporating influences, such as background noise and antenna beamwidth. With this improved model, a wave-height mapping algorithm based on the direction finding technique is presented. This approach enables the significant wave-height map extraction using a broad-beam radar. Finally, wave-height maps obtained at different sea states are depicted and analyzed, and the wave heights appearing on the maps are compared with buoy data over a one-month experiment to verify the validity and robustness of the algorithm. During this period, the significant wave height varies from about 0.5 to 4.5 m, and the radar measured wave heights at different range/distance bins show an overall root-mean-square error (RMSE) of 0.33–0.77 m and a correlation coefficient (CC) of 0.78–0.94, with respect to the buoy measurements.

Ocean wave height measurement using a wide-beam high-frequency surface wave radar (HFSWR) remains challenging due to its poor spatial resolution, which significantly limits the application of such compact systems. In this article, a novel method for wave height field extraction from the first-order Doppler spectra of a dual-frequency wide-beam radar is proposed. A model relating significant wave height to the ratio of the first-order spectral powers associated with two radar frequencies is put forward and studied numerically. Through theoretical analysis and experimental validation, it is confirmed that the first-order Doppler peaks of two radar frequencies have arisen from an approximately same direction of arrival (DOA), and their amplitudes are also affected by a similar wave directional spreading. Hereby, an algorithm combining beamforming and direction finding is developed to determine the spatial distribution of the first-order spectral power ratio and derive the significant wave height field. Finally, experimental results are given to verify the algorithm. The radar-derived wave height field agrees well with that obtained using a numerical wave model. Furthermore, the radar-measured wave heights are compared with the data collected by two buoys at the distances of 12.7 and 73 km, respectively. The comparison shows that the corresponding root-mean-square errors are 0.3 and 0.5 m and the correlation coefficients are 0.85 and 0.88, respectively.

Compact high-frequency (HF) antenna arrays are convenient to deploy. However, using a small-aperture HF surface wave radar for wind direction measurement is still a challenging problem, since an unsatisfactory array pattern degrades the performance of Bragg ratio estimation. To address this issue, a digital beamforming method based on a superdirective synthesis technique for an HF receiving array that consists of seven elements positioned on a 5-m diameter circle is proposed. This superdirective beamforming method contains a sidelobe constraint. Subsequently, a hybrid superdirective beamforming and direction-finding method is adopted to estimate the wind direction using a multifrequency HF radar based on a circular array (MHF-C). The superdirective beamforming approach, as well as the wind direction estimation method, is presented in detail. The wind direction estimation method has been applied to the raw data sets that were collected with two MHF-C radars installed along the coast of the East China Sea in April 2015 and comparisons between radar-derived and in situ wind directions have been made. Ship-mounted anemometers were used to obtain in situ measurements at six sampling locations within the overlapping coverage of both radars. Another comparison between the radar-derived and anemometer-derived wind directions, which were obtained from June 15, 2015 to August 12, 2015, has also been made. The results indicate that the proposed method is effective for wind direction estimation with root-mean-square differences (RMSDs) between 24.1° and 33.1°, when wind speeds were higher than 5 m/s. The analysis encourages us to recommend a minimum wind speed of 5 m/s for reasonably assessing wind direction measurement performance.

Shore-based coherent microwave radar and high-frequency (HF) surface wave radar are two components of a rapidly emerging set of technologies suitable for ocean wave remote sensing. To investigate their differences, this paper describes and analyzes the relationship between the water particle velocity and the wave height spectrum based on linear wave theory which underpins the algorithms developed for the analysis of data collected by coherent microwave radar. The backscatter mechanism which addresses the interaction of the HF radio waves with the ocean surface waves, as well as the empirical method adopted in our HF radar is also presented. The wave characteristics observed by the shore-based coherent S-band radar [Microwave Ocean Remote SEnsor (MORSE)] are analyzed. A multifrequency HF (MHF) radar based on a circular receiving array, which is capable of sensing waves up to 100-km offshore, is also introduced. An intercomparison of the wave height measurements obtained from the MORSE, MHF radar, and wave buoy is made. The comparison indicates that the wave heights measured by the MORSE and the MHF radar are consistent with the buoy-derived wave heights, with the root-mean-square differences (RMSDs) of 0.27 and 0.37 m, respectively.

Wave-current interaction is a common and important phenomenon in the ocean. As an ocean remote sensing tool, high-frequency (HF) radar can be used to measure currents and wave parameters. In this paper, the possibility of studying wave-current interaction using HF radar is investigated. The first-order spectral power (FSP) of HF radar is used to explore the effect of current on the Bragg wave. By analyzing the FSP change with current (FSP-current distribution), we find that, in deep water, the wave-current interactions mainly belong to 2-D refraction case, while, over a relatively shallow shelf, the interactions are stronger and more complicated. Based on local topography and current field data at Taiwan Strait, the simulation results obtained using the SWAN model confirm the 2-D refraction of the Bragg wave. When the wave-current interaction is stable, we compensate the FSP with radar-measured currents according to the radar extracted FSP-current distribution and achieve a more accurate wind estimation. Comparisons between the original and refined wind fields show the effectiveness and necessity of the current-based compensation.

A support vector regression (SVR)-based method for estimating wind speed from X-band marine radar images is proposed. The dependence of histogram pattern of radar images on wind speed and rain condition is first observed. Then, the feature vectors based on bin values of histograms are extracted and trained using an SVR algorithm. Radar images and anemometer data collected from several periods in a sea trial of the east coast of Canada are used for model training and testing. Experimental results show that compared with the ensemble empirical mode decomposition-based methods, the accuracy of wind speed estimation is improved with a reduction of about 0.14 m/s for rain-free images and 0.11 m/s for rain-contaminated images in root mean square error. Moreover, the proposed method also shows high efficiency by greatly reducing the computational time.

High-frequency (HF) radars have great potential for maritime surveillance, and the multiple signal classification (MUSIC) algorithm is usually used to estimate the direction of arrival (DOA) of targets for a wide-beam radar. However, the performance of the MUSIC algorithm relies on the precision of the antenna pattern, which could be contaminated by nearby electromagnetic interference. Therefore, the actual antenna pattern must be measured and used. In order to remove the requirement of antenna pattern measurement, a new method for target DOA estimation from wide-beam HF radar data using support vector regression (SVR) is proposed in this letter. A system model that relates target bearing and radar data feature is obtained through the SVR-based machine learning using the automatic identification system data and data associated with the vessels successfully detected by the HF radar. Then, such a model is used to determine the DOAs of targets from new data. The field experimental results at two sites demonstrate that the performance of the SVR method is better than that of the MUSIC algorithm.

In this letter, a scheme that uses convolutional neural networks (CNNs) is proposed for sea ice detection and sea ice concentration (SIC) prediction from TechDemoSat-1 Global Navigation Satellite System Reflectometry delay-Doppler maps (DDMs). Specifically, a classification-orientated CNN was designed for sea ice detection and a regression-based one for SIC estimation. Here, DDM images were used as input, and SIC data from Nimbus-7 Scanning Multi-Channel Microwave Radiometer and Defense Meteorological Satellite Program Special Sensor Microwave Imager-Special Sensor Microwave Imager/Sounder sensors were modified as targeted output. In the experimental phase, the CNN output resulted from inputting full-size DDM data (128-by-20 pixels) showed better accuracy than that of the existing NN-based method. Besides, both CNNs and NNs with further processed input data (40-by-20 pixels, and with a fixed position in each image) were evaluated and the performance of both networks was enhanced. It was found that when DDM data are adequately preprocessed, CNNs and NNs share similar accuracy; otherwise the former outperforms the latter. Further conclusion was thus drawn that CNNs were more tolerant to the data format changes than NNs.

A formulation of the relationship between sea-surface roughness and extension of the glistening zone (GZ) of a Global Navigation Satellite System Reflectometry (GNSS-R) system is presented. First, an analytical expression of the link between GZ area, viewing geometry, and surface mean square slope (MSS) is derived. Then, a strategy for retrieval of surface roughness from the delay-Doppler map (DDM) is illustrated, including details of data preprocessing, quality control, and GZ area estimation from the DDM. Next, an example for application of the proposed approach to spaceborne GNSS-R remote sensing is provided, using DDMs from the TechDemoSat-1 mission. The algorithm is first calibrated using collocated in situ roughness estimates using data sets from the National Data Buoy Center, its retrieval performance is then assessed, and some of the limitations of the suggested technique are discussed. Overall, good correlation is found between buoy-derived MSS and estimates obtained using the proposed strategy ($r=0.73$ ).

Wetland classification has always been a challenging task among remote sensing experts. Typically, wetland classes have low accuracies regardless of the applied dataset, as they have many spectral and ecological similarities. In this paper, a method is developed particularly effective for distinguishing spectrally similar classes such as wetlands. In this method, feature selection and object-based classification are not done in one step, but instead several feature selections and classifications are applied, and in each level a target class is classified and masked out. While classifying the target class, other spectrally resembling classes are merged so that feature selection is mainly concentrated on separating two classes only. Object-based features were extracted from several SAR and optical images, including RADARSAT-2, ALOS-1, ALOS-2, RapidEye and Landsat-8 images. 15 and 10 percent improvement was obtained in wetlands' average producer and user accuracies compared to the typical feature selection by using the proposed method.

In this letter, an ensemble empirical mode decomposition (EEMD)-based method is proposed to estimate significant wave height (SWH) from the X-band marine radar sea surface images. First, the data sequence in each radial direction of a radar subimage is decomposed by the EEMD into several intrinsic mode functions (IMFs). A normalization scheme is then applied to the IMFs to obtain their amplitude modulation components. Finally, by adopting a linear model, the SWH is estimated from the sum of the amplitudes from the second to the fifth modes. The method is tested using radar and buoy data collected in a sea trial off the east coast of Canada. The root-mean-square differences with respect to the buoy reference for the SWH estimations using the traditional signal-to-noise-based method, a recent shadowing-based method, and the proposed technique are 0.78, 0.48, and 0.36 m, respectively. The result indicates that the proposed technique produces improvement in the SWH measurements.

In this paper, sea surface wind direction and speed are obtained from X-band nautical radar images. A data control strategy is proposed to distinguish rain-free and rain-contaminated radar data. The radar data are decomposed by an ensemble empirical mode decomposition method into several intrinsic mode functions (IMFs) and a residual. A normalization scheme is applied to the first IMF to obtain the amplitude modulation (AM) component. Wind direction is determined from the residual for the rain-free and high-wind-speed rain-contaminated data, and from the AM portion of the first IMF for the low-wind-speed rain-contaminated data, based on curve fitting a harmonic function. Wind speed is determined from a combination of the residual and the AM part of the first IMF for both rain-free and rain-contaminated data using a logarithmic relationship. Results employing ship-borne radar and anemometer data collected in a sea trial off the east coast of Canada are presented. The root-mean-square differences for wind direction and speed measurements are 11.5° and 1.31 m/s, respectively, compared with reference values from anemometers.

Two ensemble empirical mode decomposition (EEMD)-based methods are presented to retrieve wind direction from rain-contaminated X-band nautical radar sea surface images. Each radar image is first decomposed into disparate intrinsic mode function (IMF) components using 1-D EEMD or 2-D EEMD. Then, the standard deviation of one IMF component or the combination of several IMF components as a function of azimuth is least-squares fitted to a harmonic function to determine the wind direction. Tests of the proposed algorithms are conducted by employing radar and anemometer data collected in a sea trial during rain events off the east coast of Canada. The results show that compared with the 1-D discrete-Fourier-transform-based method, both the 1-D- and 2-D-EEMD-based algorithms improve the wind direction results in rain events, showing a reduction of 7.4° and 8.7°, respectively, in the root-mean-square difference with respect to the reference.

A theoretical mixed-path ionospheric clutter model for bistatic high frequency radar is presented. Based on previous monostatic work, the first-order received electric field for bistatic radar is derived by considering the scattering processes on both the ionosphere and the ocean surface. Then, the first-order received power model is developed by incorporating a vertically polarized pulsed dipole antenna. In order to investigate the power spectrum of this ionospheric clutter model and its relative intensity to that of the ocean clutter, a normalized ionospheric clutter power is simulated. Numerical simulation results are compared with that of monostatic radar looking at the same ocean scattering patch. Subsequently, the simulations show how the bistatic angle and the ionospheric conditions affect the power spectrum for this bistatic mixed-path ionosphere clutter.

In this letter, a new method for estimating wind speeds from X-band nautical radar images is presented. Wind speeds are determined from the wavenumber spectra of radar backscatter using a logarithmic relationship between the spectral strength and the wind speed. The method can be applied to both rain-contaminated and rain-free radar data. The method has been tested using shipborne X-band nautical data collected over the North Atlantic Ocean. A comparison with the anemometer data shows that the root mean square errors of wind speeds estimated from rain-contaminated radar data using the proposed method and that by Lund et al. are 1.6 and 7.5 m/s, respectively. The wind speed estimation accuracy is improved by 5.9 m/s with the new method.

A new method for retrieving wind direction from X-band marine radar images is presented in this letter. The new algorithm investigates radar backscatter in the wavenumber domain and obtains wind direction from the wavenumber spectrum. Different from previous algorithms that detect rain-contaminated images and discard them, the new algorithm could be applied to both rain-contaminated and rain-free images. For rain-contaminated images collected under low wind speeds (i.e., less than 8 m/s), wind directions were retrieved based on spectral components with wavenumbers of [0.01, 0.2]. For rain-contaminated images obtained under high wind speeds and rain-free images, wind directions were retrieved from the spectrum with values at zero wavenumber. The algorithm has been tested using X-band radar images and shipborne anemometer data collected on the east coast of Canada. Comparison with the anemometer data shows that the root-mean-square error of wind directions retrieved from rain-contaminated images collected under low wind speeds is reduced by 25.1°.

In this paper, the retrieval of sea surface current velocity from vertically polarized (V-pol) X-band marine radar data is presented. Three different methods, including the iterative least square approach, the normalized scalar product method, and the polar current shell algorithm, that have been used for horizontally polarized data are employed here. A comprehensive comparison of the performance of the three methods is conducted using the radar images collected by the V-pol radar deployed on the Forschungsplattformen in Nord- und Ostsee No. 3 (FINO3) offshore research platform and the acoustic Doppler current profiler (ADCP) data in the North Sea. The results indicate that all three methods are capable of providing reliable current speed and direction measurements from V-pol data, with similar performance. Based on the experimental data for which the current magnitude is less than 0.5 m/s, the radar-derived results agree best with the ADCP data at a depth of 6-8 m, with the root mean square difference for current velocity x- and y-components being 7.2-8.9 cm/s. The correlation coefficients between the radar-derived and ADCP-measured current velocity components are as high as 0.87-0.93.

In this paper, a self-adaptive 2-D continuous-wavelet-transform-based algorithm for extracting wave information from X-band nautical radar images is presented. After investigating the 2-D continuous wavelet transform and its application for radar image processing, it is found that the wavelet scaling parameters will affect the results of wave field analysis. The relation of the scaling parameters to the minimum distinguishable wavenumber is developed using a calibration factor. Optimal empirical values of such calibration factors are determined from a series of simulation data tests for variable wave conditions. An iterative algorithm is then proposed that enables the system to automatically select the optimal calibration factor without requiring a reference to other instrumentation. The algorithm is evaluated using dual-polarized radar data collected on the east coast of Canada. Results of the proposed algorithm are analyzed and compared with in situ TRIAXYS wave buoy data as well as that obtained from the conventional 3-D fast Fourier transform (FFT)-based method. The impact of signal polarization on the results is explored. The agreement between the buoy and FFT results indicates that the proposed algorithm is practical and effective as an alternative to the classic 3-D FFT-based method for retrieving ocean wave information.

In this paper, a modified intensity-level-selection-based algorithm for sea surface wind retrieval from shipborne X-band nautical radar images is presented. A data quality control process is introduced to improve the accuracy of wind measurements, including the recognition of blockages and islands in the temporally integrated radar images. Moreover, a harmonic function is used to least-squares fit the selected range vector to antenna look directions to improve the wind direction determination. The wind information obtained from the radar data is compared with shipborne anemometer results. The modified algorithm reduces the mean differences between the radar and the anemometer results for wind direction and speed by about 7.4° and 1.2 m/s, respectively.

In this paper, a novel current inversion algorithm from X-band marine radar images is proposed. The routine begins with a 3D-FFT of the radar image sequence, followed by the extraction of the dispersion shell from the 3-D image spectrum. After a polar coordinate transformation, the "polar current shell" is then analyzed to retrieve current information such as the speed and direction of encounter. Particularly, a Grubbs' test is conducted to remove outliers along each radial direction, and a robust sinusoidal curve fitting is applied to the data points along each circumferential direction. For validation, the algorithm is tested against simulated radar PPI images. The results indicate that the proposed procedure, unlike most existing current inversion schemes, is not susceptible to high current speeds and has no direction restriction. Meanwhile, the relatively low computational cost makes it an excellent choice in practical marine applications.

In this letter, a new method is presented to retrieve sea-surface wind fields by least squares (LS) fitting the 2-D simulated global navigation satellite system reflectometry (GNSS-R) delay-Doppler maps (DDMs) to the measured data. Unlike previous methods, all the DDM points with normalized power higher than a threshold are used in the LS fitting. To reduce the computational cost of the fitting process, a variable step-size iteration is employed. Three GNSS-R data sets that were collected at two different sea-surface regions by the UK-Disaster Monitoring Constellation satellite are used to validate the proposed approach. An 18-s incoherent correlation processing is applied to each data set to reduce the noise level, and ad hoc correction is made on the simulated antenna pattern. The retrieved wind results are compared with the in situ measurements provided by the National Data Buoy Center. The results show that an error of 1 m/s in the wind speed and 30° in the wind direction can be obtained with a lower threshold set as 30% to 42% of the peak DDM point.

The first- and second-order monostatic cross sections of swell contaminated seas for high-frequency ground wave pulsed radar operation are derived based on the fundamental electric field equations. The ocean surface, a mixture of both swell and wind wave components, is first represented with a Fourier series. The corresponding equations of the received electric field are then obtained. Fourier transformation of the autocorrelations of the electric field gives the Doppler power spectral density, which ultimately leads to the radar cross sections. Simulated results are also provided and are observed to be highly consistent with real Doppler spectra.

In this paper, an improved algorithm for extracting sea surface currents from X-band nautical radar image sequences is presented. The marine radar images are simulated by first numerically generating current-included ocean surface elevation maps for given sea states and radar parameters. Then the plan position indicator (PPI) radar images are produced via tilt and shadowing modulation processing. From the simulated radar images, current velocity is derived with a modified iterative least square (LS) fitting method in which higher harmonic wave components and aliasing effects are considered. Here, the correct current velocity is found by a proposed iteration-terminating criterion instead of using a fixed number of iterations. Simulated inversion results for a variety of current velocities and sea states validate the algorithm and show that the number of iterations may be reduced while obtaining highly precise measurement. The performance of the algorithm is also tested using real vertically polarized radar data through comparison with the results from other current methods and an acoustic Doppler current profiler (ADCP).

A method for extracting sea surface wind direction information from bistatic high-frequency (HF) radar Doppler spectra is presented. By analogy to the monostatic case, the ratio of the intensities of the positive and negative bistatic Bragg peaks is used to derive the (ambiguous) wind direction. For bistatic operation, the reference is taken with respect to the scattering ellipse normal rather than the radar beam direction. The method is shown to be valid based on simulated bistatic HF radar Doppler spectra. Wind direction is also extracted from the bistatic radar data collected on the Southern China coast. Comparison between the radar-measured wind directions and those obtained from the Advanced Scatterometer shows good agreement.

A new approach to extracting sea surface wind speed from high-frequency radar Doppler spectra is presented. Based on certain appropriate approximations associated with the Doppler region close to the first-order (Bragg) peaks, the second-order radar cross section equation is differentiated. Once Doppler shifts due to ocean currents are removed from the data, an expression that relates the wind speed to the frequency position of the second-order peak is derived. The method is applied to simulated noisy data as well as to field data obtained from a Seasonde (a product of CODAR Oceans Sensors) in Breezy Point, NY. In the latter case, the retrieved wind speeds are then compared to ground truth data measured by an anemometer from a National Oceanic and Atmospheric Administration weather station located in the vicinity of the illuminated patch of ocean. Subject to certain constraints as detailed in the manuscript, the algorithm shows significant promise.

High-frequency (HF) bistatic Doppler cross sections of the ocean surface are examined with respect to their dependency on the bistatic angle. Previously derived results which incorporate a pulsed dipole source and two orders of scatter are considered. It is trivially seen that the first-order result has a linear dependence on the cosine of the bistatic angle. The second-order echo accounts for a double scatter of incident radiation from first-order surface waves - the so-called electromagnetic term - and a single scatter from a second-order ocean wave. The latter, generally referred to as the second-order hydrodynamic term because it originates from coupling between first-order ocean waves, predominates the Doppler continuum in most regions of interest. The analysis presented here verifies that in addition to a cosine-dependent reduction in cross section magnitude with increasing bistatic angle, both components of the second-order scatter tend to zero under the condition of near-forward scatter for bistatic HF radar operation. Of course, this imposes practical limitations on the region over which a bistatically configured HF radar system may be used to remotely sense ocean surface parameters.

High-frequency (HF) radar can be employed to measure sea surface state parameters such as waveheight, wind field, and surface current velocity. This paper describes the application of the HF ground wave radar in remote sensing the surface conditions over the Eastern China Sea in October 2000. The radar, referred to as the OSMAR2000, was developed by Wuhan University. Preliminary wave spectra, waveheights, and wind fields estimated from the collected data are presented and compared with ship-recorded measurements where such are available. The range for wind direction sensing is up to 200 km. Wave information and wind speed can be provided up to a range of 120 km. The mean difference between radar- and ship-measured significant waveheight is 0.323 m; wind direction is measured within 20/spl deg/; and wind speed to within 0.6 m/s. With such agreement being fairly reasonable, the feasibility of the inversion algorithm and the ocean state real-time sensing capability of OSMAR2000 are demonstrated.