Nabil Panchi

Ship-based sea ice analysis algorithms rely on optical images captured in optimal weather conditions with high visibility. However, Arctic imagery is often affected by weather-related degradation due to haze, snow, and rain, impacting the efficacy of deep learning tasks for sea ice analysis, such as segmentation and classification. This paper introduces and evaluates two strategies to address weather-induced degradation in optical sea ice images (RGB). Strategy 1 employs a two-step pipeline: first, removal of weather degradation using deep learning based de-weathering algorithms, then analysis of images as a part of sea ice segmentation/classification tasks. Strategy 2 proposes a ’weather as augmentation’ training approach to create all-in-one weather-resilient segmentation and classification models. Furthermore, we introduce the first open-source ice image dataset (SeaIceWeather) with paired images - one clean and one weather-degraded. Such a dataset allows for training and validation of supervised deep learning based de-weathering algorithms. Using this dataset, we show that the proposed strategies are effective against weather-degraded images, achieving parity with the segmentation and classification performance on clean images. Additionally, we demonstrate de-weathering models capable of removing degradations due to four different weather conditions, including rain, haze, snow, and raindrops on a camera lens with a single set of weights. The presented strategies lay the foundation for robust ship-borne sea ice analysis systems resilient to adverse weather conditions. Furthermore, we hope that the dataset introduced in this study ignites further interest in the analysis of weather-degraded sea ice images. The dataset can be downloaded from our project web page https://sites.google.com/view/seaiceweather/home.

Icebreaking assistance in ice-covered waters involves the use of icebreaker(s) to create an ice channel for the assisted vessel(s), which facilitate the vessel(s) navigation in ice conditions. However, assisted vessels that follow the icebreaker may become beset in the ice channel. In April 2021, vessel A navigated in the Baltic Sea with the assistance of icebreakers B and C, during which vessel A experienced three instances of besetting within the ice channel. This research endeavors to examine the underlying causes of vessel besetting within the ice channel by analyzing automatic identification system (AIS) data and ice condition data. The preliminary findings indicate that the primary factors that contributed to that besetting were the severe ice conditions and considerable distances between the icebreaker and the vessel. Evidence to suggest that the channel radius influenced the besetting have not been found, but it is hypothesized that the curvature of the ice channel may have impacted the maneuvering capabilities of icebreaker C in the third beset instance.

This study investigates variability of ships’ operational limitations for ice conditions in view of national rules and international recommendations (the Rules for Navigation in the Water Area of the Northern Sea Route and the International Maritime Organization Recommendations). As an example, the Kara Sea region is considered for the period of 2017–2019. The ship data from two datasets were considered: (1) the Northern Sea Route Administration data and (2) the Automatic Identification System data. For each ship’s ice class, the ice information (concentration and type) was used to derive operational limitations based on the Polar Operational Limit Assessment Risk Indexing System (POLARIS), and the results were compared with the newly updated Northern Sea Route access criteria. Preliminary results indicate that national rules are generally stricter than international recommendations, however on some occasions, the national rules impose lesser restrictions on operations as ships with ice class PC7/Arc4 are allowed to operate when the international recommendations suggest an elevated probability of an accidental event.

This paper presents a three-stage approach for the automated analysis of close-range optical images containing ice objects. The proposed system is based on an ensemble of deep learning models and conditional random field postprocessing. The following surface ice formations were considered: Icebergs, Deformed ice, Level ice, Broken ice, Ice floes, Floebergs, Floebits, Pancake ice, and Brash ice. Additionally, five non-surface ice categories were considered: Sky, Open water, Shore, Underwater ice, and Melt ponds. To find input parameters for the approach, the performance of 12 different neural network architectures was explored and evaluated using a 5-fold cross-validation scheme. The best performance was achieved using an ensemble of models having pyramid pooling layers (PSPNet, PSPDenseNet, DeepLabV3+, and UPerNet) and convolutional conditional random field postprocessing with a mean intersection over union score of 0.799, and this outperformed the best single-model approach. The results of this study show that when per-class performance was considered, the Sky was the easiest class to predict, followed by Deformed ice and Open water. Melt pond was the most challenging class to predict. Furthermore, we have extensively explored the strengths and weaknesses of our approach and, in the process, discovered the types of scenes that pose a more significant challenge to the underlying neural networks. When coupled with optical sensors and AIS, the proposed approach can serve as a supplementary source of large-scale ‘ground truth’ data for validation of satellite-based sea-ice products. We have provided an implementation of the approach here.