Classification of sea ice types in Sentinel-1 synthetic aperture radar images

Abstract

Abstract. A new Sentinel-1 image-based sea ice classification algorithm using a machine-learning-based model trained in a semi-automated manner is proposed to support daily ice charting. Previous studies mostly rely on manual work in selecting training and validation data. We show that the readily available ice charts from the operational ice services can reduce the amount of manual work in preparation of large amounts of training/testing data. Furthermore, they can feed highly reliable data to the trainer by indirectly exploiting the best ability of the sea ice experts working at the operational ice services. The proposed scheme has two phases: training and operational. Both phases start from the removal of thermal, scalloping, and textural noise from Sentinel-1 data and calculation of grey level co-occurrence matrix and Haralick texture features in a sliding window. In the training phase, the weekly ice charts are reprojected into the SAR image geometry. A random forest classifier is trained with the texture features on input and labels from the rasterized ice charts on output. Then, the trained classifier is directly applied to the texture features from Sentinel-1 images operationally. Test results from the two datasets spanning winter (January–March) and summer (June–August) seasons acquired over the Fram Strait and the Barents Sea showed that the classifier is capable of retrieving three generalized cover types (open water, mixed first-year ice, old ice) with overall accuracies of 87 % and 67 % in winter and summer seasons, respectively. For the summer season, the classifier failed in distinguishing mixed first-year ice from old ice with accuracy of only 12 %; however, it performed rather like an ice–water discriminator with high accuracy of 98 % as the misclassification between the mixed first-year ice and old ice was between them. The accuracy for five cover types (open water, new ice, young ice, first-year ice, old ice) in the winter season was 60 %. The errors are attributed both to incorrect manual classification on the ice charts and to the semi-automated algorithm. Finally, we demonstrate the potential for near-real-time service of the ice map using daily mosaicked Sentinel-1 images.

Presentation and evaluation of the Arctic sea ice forecasting system neXtSIM-F

“…Other alternatives for assimilating extent are MASIE (Multisensor Analyzed Sea Ice Extent: Fetterer et al, 2010) and IMS (Interactive Multisensor Snow and Ice Mapping System: Helfrich et al, 2007), as done by the GOFS 3.1 system for example. Automatic ice type classification from SAR is also possible now -for example the algorithm developed by Park et al (2020) and upgraded by Boulze et al (2020) will become operationally distributed by CMEMS in 2021.…”

“…Arctic sea ice has been in great decline in the last number of years (Meier, 2017). Perovich et al (2018) report that in 2018, the summer extent was the sixth lowest and the winter extent was the second lowest in the satellite record . Moreover, surface air temperatures in the Arctic continued to warm at twice the rate relative to the rest of the globe, and Arctic air temperatures for the past 5 years (2014)(2015)(2016)(2017)(2018) have exceeded all previous records since 1900 (Overland et al, 2018), which will also contribute to future sea ice decline if it continues.…”

Automatic Selection of Relevant Attributes for Multi-Sensor Remote Sensing Analysis: A Case Study on Sea Ice Classification

“…The pertinence of different attributes for sea-ice characterization has been investigated for both SAR [4], [7]- [14] and optical imagery [15], [16]. In some studies, e.g., the entropy was found to be well suited for separating sea ice types [13]. In other studies, however, the same parameter was found to be less relevant for sea ice classification [8], [11], and less useful for detection of leads in the ice [9], [12].…”

Development of a Dual-Attention U-Net Model for Sea Ice and Open Water Classification on SAR Images

“…A series of studies have been devoted to classifying sea ice and open water on SAR images, including threshold-based methods [4], expert systems [5], and machine learning (ML) methods. Since the 21st century, ML has become the mainstream method, such as the neural network (NN) model [6], [7], support vector machine (SVM) [8], [9], and random forest (RF) classifier [10].…”

“…A typical procedure of these CNN-based models includes This work is licensed under a Creative Commons Attribution 4.0 License. For more information, see https://creativecommons.org/licenses/by/4.0/ two steps: 1) inputting an SAR image chip (e.g., with the size of 13 × 13, 45 × 45, or 50 × 50 [10], [17]- [20]) into several stacking pairs of CNN + max-pooling layers to extract downscaled feature using FC NN layers to connect the downscaled feature maps and output the class of the central pixel of the input chip. The drawback is that we need manually construct training/testing samples for each pixel of an image, increasing the workload and the storage burden of memory.…”