Sea Ice Classification Research Articles

Angular variation of SAR polarimetric parameters over multiyear ice

ABSTRACT Backscatter and derived parameters from Synthetic Aperture Radar (SAR) may vary across the swath as the incident angle increases from the near to the far range, especially in the ScanSAR mode. Previous studies characterize this angular variation in the form of linear regression for commonly used ice types, namely first-year (FYI) and multi-year (MYI). The present study took advantage of the appearance of three grounded multi-year ice floes, viewed at different radar incidence angles, in a series of 22 RADARSAT-2 fully-polarimetric images, acquired over the Resolute Passage in the Canadian Arctic, during the period from 21 October to 28 December 2017. The ice properties remained invariant during the observation period; hence, variation of radar parameters of MYI can be attributed to change of incidence angle only. Quantification of angular variation of the selected SAR parameters was performed. They include conventional orthogonal backscatter coefficients, derived parameters from the incoherent decomposition of Cloude-Pottier, Yamaguchi, and Touzi, as well as a set of parameters from the compact polarimetry mode. Results specify the parameters that reveal variation with incidence angle and present the linear regressions equations. The explanation for the behaviour of each parameter, in terms of showing or not showing an angular trend, is offered. Results will be useful when using the polarimetric parameters in the sea ice classification scheme.

Investigating coincident L- and S-band ASAR imagery over Arctic sea ice

Our investigation into coincident L- and S-band ASAR (airborne synthetic aperture radar) imagery explored Arctic sea ice separability in the Beaufort Sea, particularly in light of the imminent launch of the NASA-ISRO Synthetic Aperture Radar (NISAR) mission. Our research has revealed an improved capability to separate i) multi-year and first-year sea ice at the S-band imagery, as well as ii) a higher separability within first-year sea ice classes at L-band imagery. We have also reported that wind-roughened melt ponds show a distinct signature in the S-band. Importantly, our machine learning algorithm has achieved higher accuracy in sea ice classification at the S-band than the L-band. These findings have significant implications for the future of sea ice research and operations using SAR imagery from the NISAR mission.

IceGCN: An Interactive Sea Ice Classification Pipeline for SAR Imagery Based on Graph Convolutional Network

Monitoring sea ice in the Arctic region is crucial for polar maritime activities. The Canadian Ice Service (CIS) wants to augment its manual interpretation with machine learning-based approaches due to the increasing data volume received from newly launched synthetic aperture radar (SAR) satellites. However, fully supervised machine learning models require large training datasets, which are usually limited in the sea ice classification field. To address this issue, we propose a semi-supervised interactive system to classify sea ice in dual-pol RADARSAT-2 imagery using limited training samples. First, the SAR image is oversegmented into homogeneous regions. Then, a graph is constructed based on the segmentation results, and the feature set of each node is characterized by a convolutional neural network. Finally, a graph convolutional network (GCN) is employed to classify the whole graph using limited labeled nodes automatically. The proposed method is evaluated on a published dataset. Compared with referenced algorithms, this new method outperforms in both qualitative and quantitative aspects.

Dynamic Thresholding Fully Automated sea ice extraction and classification methods based on multi-source remote-sensing data in the Yellow sea and Bohai sea regions

This paper presents the extraction and classification of sea ice in the Yellow and Bohai seas using multi-source remote sensing products, including Landsat 8, HY-1C, and MODIS. We propose a Dynamic Thresholding Fully Automated (DTFA) method for sea ice extraction and classification. The accuracy of sea ice extraction and classification is evaluated using four metrics: confusion matrix accuracy rate, overall accuracy, user accuracy, and kappa coefficient. The DTFA method showed higher extraction and classification accuracy compared to the SVM and k-means++ methods. HY-1C has the most effective extraction, followed by Landsat 8. Landsat 8 is able to extract new ice, grey ice, and grey-white ice with the best effect, while HY-1C has a better effect in extracting white ice. MODIS data has advantages in coverage and data updating. Finally, the fusion of multi-source remote sensing images was used to obtain the sea ice extent of the entire Chinese offshore. This method is also applicable to remote sensing monitoring of sea ice in other sea areas.

Deep learning techniques for enhanced sea-ice types classification in the Beaufort Sea via SAR imagery

This study proposes a dual-branch encoder U-Net (DBU-Net) deep learning model to classify sea ice types based on synthetic aperture radar (SAR) images in the Beaufort Sea. The DBU-Net can segment multi-year ice (MYI), first-year ice (FYI), open water (OW), and leads on SAR images. We design a dual-branch encoder to fuse the polarization and the grey-level co-occurrence matrix (GLCM) information of SAR images to improve the model's classification capability. The model is subsequently fine-tuned using lead samples to identify leads. 24 Sentinel-1 SAR images acquired in the Beaufort Sea are utilized for model training and testing. The accuracy (Acc), mean intersection over union (mIoU), and kappa coefficient (Kappa) are employed as evaluation metrics. Experiments show that DBU-Net achieves 91.83%/0.841/0.849 in Acc/mIoU/Kappa in classifying MYI, FYI, and OW, significantly outperforming three traditional models based on support vector machine, random forest, or convolutional neural network. Compared with the original U-Net, the dual-branch encoder and the GLCMs improve 1.45%/4.4%/2.8% in Acc/mIoU/Kappa in MYI, FYI, and OW. Acc/mIoU/Kappa metrics of leads detection is 99.49%/0.801/0.754. Besides, 454 Sentinel-1 SAR images are fed into the optimal DBU-Net to generate 80 m sea ice products in the Beaufort Sea for winters 2018–2022. As the MYI draws wide attention and the FYI and MYI are complementary in the area during the Winter, we discuss the variation of MYI based on the generated sea ice products and explore the relationship between MYI's variation and the Beaufort High. We found that the MYI export in the 2018/19 winter was due to large summer sea ice remains and the abnormal sea ice motion caused by the southeast shifting Beaufort Atmospheric Pressure High (Beaufort High). The MYI import in the 2020/21 winter was due to a strong northward MYI import caused by the powerful Beaufort High.

Arctic Thin Ice Detection Using AMSR2 and FY-3C MWRI Radiometer Data

Thin ice with a thickness of less than half a meter produces strong salt and heat fluxes which affect deep water circulation and weather in the polar oceans. The identification of thin ice areas is essential for ship navigation. We have developed thin ice detection algorithms for the AMSR2 and FY-3C MWRI radiometer data over the Arctic Ocean. Thin ice (<20 cm) is detected based on the classification of the H-polarization 89–36-GHz gradient ratio (GR8936H) and the 36-GHz polarization ratio (PR36) signatures with a linear discriminant analysis (LDA) and thick ice restoration with GR3610H. The brightness temperature (TB) data are corrected for the atmospheric effects following an EUMETSAT OSI SAF correction method in sea ice concentration retrieval algorithms. The thin ice detection algorithms were trained and validated using MODIS ice thickness charts covering the Barents and Kara Seas. Thin ice detection is applied to swath TB datasets and the swath charts are compiled into a daily thin ice chart using 10 km pixel size for AMSR2 and 20 km for MWRI. On average, the likelihood of misclassifying thick ice as thin in the ATIDA2 daily charts is 7.0% and 42% for reverse misclassification. For the MWRI chart, these accuracy figures are 4% and 53%. A comparison of the MWRI chart to the AMSR2 chart showed a very high match (98%) for the thick ice class with SIC > 90% but only a 53% match for the thin ice class. These accuracy disagreements are due to the much coarser resolution of MWRI, which gives larger spatial averaging of TB signatures, and thus, less detection of thin ice. The comparison of the AMSR2 and MWRI charts with the SMOS sea ice thickness chart showed a rough match in the thin ice versus thick ice classification. The AMSR2 and MWRI daily thin ice charts aim to complement SAR data for various sea ice classification tasks.

Assessment of Sea-Ice Classification Capabilities during Melting Period Using Airborne Multi-Frequency PolSAR Data

Sea-ice mapping using Synthetic Aperture Radar (SAR) in the melt season poses challenges, primarily due to meltwater complicating the distinguishability of sea-ice types. In response to this issue, this study introduces a novel method for classifying sea ice during the Bohai Sea’s melting period. The method categorizes sea ice into five types: open water (OW), gray ice (Gi), melting gray ice (GiW), gray–white Ice (Gw), and melting gray–white Ice (GwW). To achieve this classification, 51 polarimetric features are extracted from L-, S-, and C-band PolSAR data using various polarization decomposition methods. This study assesses the separability of these features among different combinations of sea-ice type by calculating the Euclidean distance (ED). The Support Vector Machine (SVM) classifier, when employed with single-frequency polarimetric feature sets, achieves the highest accuracy for OW and Gi in the C-band, GiW in the S-band, and Gw and GwW in the L-band. Remarkably, the C-band features exhibit the overall highest accuracy when compared to the L-band and S-band. Furthermore, employing a multi-dimensional polarimetric feature set significantly improves classification accuracy to 94.55%, representing a substantial enhancement of 9% to 22% compared to single-frequency classification. Benefiting from the performance advantages of Random Forest (RF) classifiers in handling large datasets, RF classifiers achieve the highest classification accuracy of 95.84%. The optimal multi-dimensional feature composition includes the following: L-band: SE, SEI, α¯, Span; S-band: SEI, SE, Span, PV-Freeman, λ1, λ2; C-band: SE, SEI, Span, λ3, PV-Freeman. The results of this study provide a reliable new method for future sea-ice monitoring during the melting season.

A Multiscale Dual Attention Network for the Automatic Classification of Polar Sea Ice and Open Water Based on Sentinel-1 SAR Images

A Multiscale Dual Attention Network for the Automatic Classification of Polar Sea Ice and Open Water Based on Sentinel-1 SAR Images

Partial Label Learning With Focal Loss for Sea Ice Classification Based on Ice Charts

Partial Label Learning With Focal Loss for Sea Ice Classification Based on Ice Charts

A Multi-Task Network With Dynamic Segmentation for Sea Ice Classification in Arctic Shipping Route Optimization

A Multi-Task Network With Dynamic Segmentation for Sea Ice Classification in Arctic Shipping Route Optimization

MFDA: Unified Multi-Task Architecture for Cross-Scene Sea Ice Classification

MFDA: Unified Multi-Task Architecture for Cross-Scene Sea Ice Classification

Arctic shipping trends during hazardous weather and sea-ice conditions and the Polar Code’s effectiveness

The Arctic’s extreme environmental conditions and remoteness make it a complex and dynamic environment for maritime operators. We find that Arctic shipping has grown by 7% per year over the past decade, despite the hazardous weather and sea-ice conditions that pose risks to vessels operating in the region. As a result of a strong increase in winter sailing, the time ships operate in these extreme conditions has even tripled. To mitigate maritime risks, the Polar Code has been introduced. Among other things, it regulates Arctic shipping by specifying hazardous conditions with a sea-ice classification scheme and design temperature threshold. However, we argue that the Polar Code needs refinement through the integration of maritime warning systems and a broader description of hazardous conditions. This is supported by an analysis of shipping activity patterns in severe sea-spray icing conditions and a discussion of a recent sea-ice induced incident along the Northern Sea Route.

Enhancing sea ice segmentation in Sentinel-1 images with atrous convolutions

ABSTRACT Due to the growing volume of remote sensing data and the low latency required for safe marine navigation, machine learning (ML) algorithms are being developed to accelerate sea ice chart generation, currently a manual interpretation task. However, the low signal-to-noise ratio of the freely available Sentinel-1 Synthetic Aperture Radar (SAR) imagery, the ambiguity of backscatter signals for ice types, and the scarcity of open-source high-resolution labelled data makes automating sea ice mapping challenging. We use Extreme Earth version 2, a high-resolution benchmark dataset generated for ML training and evaluation, to investigate the effectiveness of ML for automated sea ice mapping. Our customized pipeline combines ResNets and Atrous Spatial Pyramid Pooling for SAR image segmentation. We investigate the performance of our model for: i) binary classification of sea ice and open water in a segmentation framework; and ii) a multiclass segmentation of five sea ice types. For binary ice-water classification, models trained with our largest training set have weighted F1 scores all greater than 0.95 for January and July test scenes. Specifically, the median weighted F1 score was 0.98, indicating high performance for both months. By comparison, a competitive baseline U-Net has a weighted average F1 score of ranging from 0.92 to 0.94 (median 0.93) for July, and 0.97 to 0.98 (median 0.97) for January. Multiclass ice type classification is more challenging, and even though our models achieve 2% improvement in weighted F1 average compared to the baseline U-Net, test weighted F1 is generally between 0.6 and 0.80. Our approach can efficiently segment full SAR scenes in one run, is faster than the baseline U-Net, retains spatial resolution and dimension, and is more robust against noise compared to approaches that rely on patch classification.

Multi-Featured Sea Ice Classification with SAR Image Based on Convolutional Neural Network

Sea ice is a significant factor in influencing environmental change on Earth. Monitoring sea ice is of major importance, and one of the main objectives of this monitoring is sea ice classification. Currently, synthetic aperture radar (SAR) data are primarily used for sea ice classification, with a single polarization band or simple combinations of polarization bands being common choices. While much of the current research has focused on optimizing network structures to achieve high classification accuracy, which requires substantial training resources, we aim to extract more information from the SAR data for classification. Therefore we propose a multi-featured SAR sea ice classification method that combines polarization features calculated by polarization decomposition and spectrogram features calculated by joint time-frequency analysis (JTFA). We built a convolutional neural network (CNN) structure for learning the multi-features of sea ice, which combines spatial features and physical properties, including polarization and spectrogram features of sea ice. In this paper, we utilized ALOS PALSAR SLC data with HH, HV, VH, and VV, four types of polarization for the multi-featured sea ice classification method. We divided the sea ice into new ice (NI), first-year ice (FI), old ice (OI), deformed ice (DI), and open water (OW). Then, the accuracy calculation by confusion matrix and comparative analysis were carried out. Our experimental results demonstrate that the multi-feature method proposed in this paper can achieve high accuracy with a smaller data volume and computational effort. In the four scenes selected for validation, the overall accuracy could reach 95%, 91%, 96%, and 95%, respectively, which represents a significant improvement compared to the single-feature sea ice classification method.

Region-Based Sea Ice Mapping Using Compact Polarimetric Synthetic Aperture Radar Imagery with Learned Features and Contextual Information

Operational sea ice maps are usually generated manually using dual-polarization (DP) synthetic aperture radar (SAR) satellite imagery, but there is strong interest in automating this process. Recently launched satellites offer compact polarimetry (CP) imagery that provides more comprehensive polarimetric information compared to DP, which compels the use of CP for automated classification of SAR sea ice imagery. Existing sea ice scene classification algorithms using CP imagery rely on handcrafted features, while neural networks offer the potential of features that are more discriminating. We have developed a new and effective sea ice classification algorithm that leverages the nature of CP data. First, a residual-based convolutional neural network (ResCNN) is implemented to classify each pixel. In parallel, an unsupervised segmentation is performed to generate regions based on CP statistical properties. Regions are assigned a single class label by majority voting using the ResCNN output. For testing, quad-polarimetric (QP) SAR sea ice scenes from the RADARSAT Constellation Mission (RCM) are used, and QP, DP, CP, and reconstructed QP modes are compared for classification accuracy, while also comparing them to other classification approaches. Using CP achieves an overall accuracy of 96.86%, which is comparable to QP (97.16%), and higher than reconstructed QP and DP data by about 2% and 10%, respectively. The implemented algorithm using CP imagery provides an improved option for automated sea ice mapping.

Fine Resolution Classification of New Ice, Young Ice, and First-Year Ice Based on Feature Selection from Gaofen-3 Quad-Polarization SAR

A new method of sea ice classification based on feature selection from Gaofen-3 polarimetric Synthetic Aperture Radar (SAR) observations was proposed. The new approach classifies sea ice into four categories: open water (OW), new ice (NI), young ice (YI), and first-year ice (FYI). Seventy parameters that have previously been applied to sea ice studies were re-examined for sea ice classification in the Okhotsk Sea near the melting point on 28 February 2020. The ‘separability index (SI)’ was used for the selection of optimal features for sea ice classification. Full polarization parameters (the backscatter intensity contains the horizontal transmit-receive intensity (σhh0), Shannon entropy (SEi), the spherical scattering component of Krogager decomposition (Ks)), and hybrid polarization parameters (horizontal receive intensity(σrh0), hybrid-pol Shannon entropy (CPSEi), the correlation coefficient (ρrh−rv) between the σrh0 and σrv0, and the surface scattering component of m − α decomposition αs) were determined as the optimal parameters for the different work modes of SAR. The selected parameters were used to classify sea ice by the random forest classifier (RFC), and classification results were validated by manually interpreted ice maps derived from Landsat-8 data. The classification accuracy of OW, NI, YI and FYI reached 95%, 96%, 98% and 85%, respectively.

Sea-Ice Classification Using Conditional Generative Adversarial Networks

In this letter, we develop a conditional generative adversarial network (cGAN) semantic segmentation model to detect and classify sea-ice in scenes captured onboard of a ship. Two datasets are utilized to train the cGAN model. The images in the first dataset capture four classes: sea-ice, open water, sky, and vessel. The images in the second dataset capture first year sea-ice, new sea-ice, and grey sea-ice in addition to the open water, sky, and vessel classes. Data augmentation operations are applied to both datasets to change the camera mounting angle and location and enhance the datasets' size. Results illustrate that the cGAN model outperforms the visual geometry group (VGG-16), fully convolutional network, and pyramid scene parsing network models in both datasets. Results also show that the cGAN model provides remarkable performance with rainy weather images, in which the performance of the other models degrades due to the raindrop.

Sea ice classification of TerraSAR-X ScanSAR images for the MOSAiC expedition incorporating per-class incidence angle dependency of image texture

Abstract. We provide sea ice classification maps of a sub-weekly time series of single (horizontal–horizontal, HH) polarization X-band TerraSAR-X scanning synthetic aperture radar (TSX SC) images from November 2019 to March 2020, covering the Multidisciplinary drifting Observatory for the Study of Arctic Climate (MOSAiC) expedition. This classified time series benefits from the wide spatial coverage and relatively high spatial resolution of TSX SC data and is a useful basic dataset for future MOSAiC studies on physical sea ice processes and ocean and climate modeling. Sea ice is classified into leads, young ice with different backscatter intensities, and first-year ice (FYI) or multiyear ice (MYI) with different degrees of deformation. We establish the per-class incidence angle (IA) dependencies of TSX SC intensities and gray-level co-occurrence matrix (GLCM) textures and use a classifier that corrects for the class-specific decreasing backscatter with increasing IAs, with both HH intensities and textures as input features. Optimal parameters for texture calculation are derived to achieve good class separation while maintaining maximum spatial detail and minimizing textural collinearity. Class probabilities yielded by the classifier are adjusted by Markov random field contextual smoothing to produce classification results. The texture-based classification process yields an average overall accuracy of 83.70 % and good correspondence to geometric ice surface roughness derived from in situ ice thickness measurements (correspondence consistently close to or higher than 80 %). A positive logarithmic relationship is found between geometric ice surface roughness and TSX SC HH backscatter intensity, similar to previous C- and L-band studies. Areal fractions of classes representing ice openings (leads and young ice) show prominent increases in middle to late November 2019 and March 2020, corresponding well to ice-opening time series derived from in situ data in this study and those derived from satellite synthetic aperture radar (SAR) and optical data in other MOSAiC studies.

Classification of Arctic Sea Ice Type in CFOSAT Scatterometer Measurements Using a Random Forest Classifier

The Ku-band scatterometer called CSCAT onboard the Chinese–French Oceanography Satellite (CFOSAT) is the first spaceborne rotating fan-beam scatterometer (RFSCAT). A new algorithm for classification of Arctic sea ice types on CSCAT measurement data using a random forest classifier is presented. The random forest classifier is trained on the National Snow and Ice Data Center (NSIDC) weekly sea ice age and sea ice concentration product. Five feature parameters, including the mean value of horizontal and vertical polarization backscatter coefficient, the standard deviation of horizontal and vertical polarization backscatter coefficient and the copol ratio, are innovatively extracted from orbital measurement for the first time to distinguish water, first-year ice (FYI) and multi-year ice (MYI). The overall accuracy and kappa coefficient of sea ice type model are 93.35% and 88.53%, respectively, and the precisions of water, FYI, and MYI are 99.67%, 86.60%, and 79.74%, respectively. Multi-source datasets, including daily sea ice type from the EUMETSAT Ocean and Sea Ice Satellite Application Facility (OSI SAF), NSIDC weekly sea ice age, multi-year ice concentration (MYIC) provided by the University of Bremen, and SAR-based sea ice type released by Copernicus Marine Environment Monitoring Service (CMEMS) have been used for comparison and validation. It is shown that the most obvious difference in the distribution of sea ice types between the CSCAT results and OSI SAF sea ice type are mainly concentrated in the marginal zones of FYI and MYI. Furthermore, compared with OSI SAF sea ice type, the area of MYI derived from CSCAT is more homogeneous with less noise, especially in the case of younger multiyear ice. In the East Greenland region, CSCAT identifies more pixels as MYI with lower MYIC values, showing better accuracy in the identification of areas with obvious mobility of MYI. In conclusion, this research verifies the capability of CSCAT in monitoring Arctic sea ice classification, especially in the spatial homogeneity and detectable duration of sea ice classification. Given the high accuracy and processing speed, the random forest-based algorithm can offer good guidance for sea ice classification with FY-3E/RFSCAT, i.e., a dual-frequency (Ku and C band) scatterometer called WindRAD.

SAR and Passive Microwave Fusion Scheme: A Test Case on Sentinel‐1/AMSR‐2 for Sea Ice Classification

AbstractThe most common source of information about sea ice conditions is remote sensing data, especially images obtained from synthetic aperture radar (SAR) and passive microwave radiometers (PMR). Here we introduce an adaptive fusion scheme based on Graph Laplacians that allows us to retrieve the most relevant information from satellite images. In a first test case, we explore the potential of sea ice classification employing SAR and PMR separately and simultaneously, in order to evaluate the complementarity of both sensors and to assess the result of a combined use. Our test case illustrates the flexibility and efficiency of the proposed scheme and indicates an advantage of combining AMSR‐2 89 GHz and Sentinel‐1 data for sea ice mapping.