Recent developments of artificial intelligence methods for sea ice concentration monitoring using high-resolution imaging datasets

The negative effect of sea ice shrinkage on the world's climate has been widely studied. The shrinkage of sea ice affects the heat transfer balance between the ocean and atmosphere, especially in high-latitude regions. Given the influence of global warming, the number of sea ice areas has considerably decreased over the previous decades. Hence, precise, dependable measurement of sea ice concentration is necessary to ensure the security and effectiveness of marine operations in polar areas. Monitoring of sea ice concentration through remote sensing technology is critical in predicting sea ice-related hazards, facilitating safe ship navigation and enabling polar environmental research. Furthermore, a few imaging technologies, such as synthetic aperture radar (SAR), have become the predominant tools utilised for regional ice observation because of their capability to deliver high-resolution images all year round in nearly all locations so that ice charts can be mapped effectively by national ice organisations. This study examines the utilisation of artificial intelligence (AI), particularly deep learning, in monitoring sea ice concentration through high-resolution imaging datasets obtained from several resources. These datasets include comprehensive spatial information, which is essential for accurate sea ice mapping and categorisation. Some of them focus on several main obstacles, including speckle noise in SAR images, data sparsity and amalgamation of multi-modal datasets whilst suggesting AI-based ways to mitigate these problems. This study concentrates on high-resolution imaging datasets, but it also recognises the potential use of AI in other remote sensing modalities, including scatterometry, altimetry and passive microwave sensors. This review explicitly examines research articles published from 2019 to the present, particularly those that emphasise deep learning methods. It also investigates readily available sea ice datasets, such as SAR-, optical- and drone-based datasets. Subsequently, relevant sea ice mapping techniques, including conventional algorithms, machine learning-based approaches and deep learning-based methods, are assessed. This work focuses on three main applications, namely, weather prediction, marine security and environmental preservation, and their future opportunities. Comprehensive discussions of the aforementioned topics, including the main findings and patterns, are provided to stimulate future research. • A methodological review of sea ice mapping methods based on satellite imaging data utilising artificial intelligence methods. • Summarization of many categories of sea ice statistics as well as publicly available datasets for AI applications. • Present a comprehensive analysis of several eras of automated sea ice mapping technology. • Provide possible research directions for future improvement on machine learning-based sea ice mapping.