Super-Resolution-Aided Sea Ice Concentration Estimation From AMSR2 Images by Encoder–Decoder Networks With Atrous Convolution

Abstract

Passive microwave data is an important data source for the continuous monitoring of Arctic-wide sea ice concentration (SIC). However, its coarse spatial resolution leads to blurring effects at the ice-water divides, resulting in the great challenges of fine-scale and accurate SIC estimation, especially for regions with low SIC. Besides, the SIC derived by operational algorithms using high-frequency passive microwave observations has great uncertainties in open water or marginal ice zones due to atmospheric effects. In this article, a novel framework is proposed to achieve accurately SIC estimation with improved spatial details from original low-resolution Advanced Microwave Scanning Radiometer 2 (AMSR2) images, with joint the super-resolution (SR) and SIC estimation network. Based on the SR network, the spatial resolution of original AMSR2 images can be improved by four times, benefiting to construct AMSR2 SR features with more high-frequency information for SIC estimation. The SIC network with an encoder-decoder structure and atrous convolution, is employed to accurately perform the SIC retrieval by considering the characteristics of passive microwave images in the Arctic sea ice region. Experimental results show that the proposed SR-Aided SIC estimation approach can generate accurate SIC with more detailed sea ice textures and much sharper sea ice edges. With respect to MODIS SIC products distributed in Arctic scale, the proposed model achieves a root-mean-square error (RMSE) of 5.94% and mean absolute error (MAE) of 3.04%, whereas the Arctic Radiation and Turbulence Interaction Study (ARTIST) Sea Ice (ASI) SIC results have three and two times greater values of RMSE and MAE.

Sea Ice Extraction via Remote Sensing Imagery: Algorithms, Datasets, Applications and Challenges

“…Murashkin et al [121] applied UNET++ to the task of mapping Arctic sea ice in Sentinel-1 SAR scenes. Feng et al [122] proposed a joint super-resolution (SR) method to enhance the spatial resolution of original AMSR2 images. They used a DeepLabv3+ network to estimate SIC, which demonstrated good robustness in different regions of the Arctic at different times.…”

Advancing Arctic sea ice remote sensing with AI and deep learning: now and future

“…Additional input data encompass satellite products, including optical (Dawson et al, 2022) and SAR (Dawson et al, 2022;Shamshiri et al, 2022), capturing surface features of sea ice and indirectly inferring its thickness. Furthermore, thermodynamic parameters such as air temperature, sea surface temperature, wind speed, and snow depth, derived from reanalysis data such as ERA5 (Liang et al, 2023;Liu et al, 2023), can suggest sea ice thickness indirectly, reflecting the conditions conducive to ice formation and evolution. The strength of reanalysis data lies in their global coverage and continuity, but they are model-derived and not direct measurements, so they may contain biases or uncertainties.…”

“…Ground truth data can be obtained from in-situ measurements such as AWI Moored Upward Looking Sonar (ULS) data (Shamshiri et al, 2022) or calculated from CryoSat-2 data (Chi and Kim, 2021;Dawson et al, 2022;Liang et al, 2023), which are explicitly designed to monitor polar ice thickness fluctuations and variations in sea ice. Numerical simulation models like Pan-Arctic Ice Ocean Modeling and Assimilation System (PIOMAS; Zhang and Rothrock, 2003) provide another source of sea ice thickness data, despite potential limitations and uncertainties (Liu et al, 2023). In terms of output, deep learning models typically generate a continuous map of sea ice thickness, usually in the form of gridded datasets representing the estimated sea ice thickness for specific geographic locations.…”

“…Based on this result, Bayesian inversion was further applied to infer the ice thickness. SIT estimation can also be cast as a forecasting task, using temporal models such as PredRNN++ to predict future sea ice thickness (Liu et al, 2023). By introducing a spatiotemporal memory flow, PredRNN (Wang et al, 2022d) is capable of capturing the decoupled long-and short-term dynamics and the spatiotemporal variations in the sequence data.…”

Pan-Arctic Sea Ice Concentration from SAR and Passive Microwave

“…The current state-of-the-art in pan-Arctic sea ice concentration (SIC) retrieval primarily relies on passive microwave (PMW) sensors (Cavalieri et al, 1984;Andersen et al, 2007;Tonboe et al, 2016;Lavergne et al, 2019), which provide global coverage but suffer from relatively coarse spatial resolution (Feng et al, 2023). PMW-based products are crucial for monitoring long-term trends, and while some experimental products offer grid resolutions as high as 3.125 km (Meier and Stewart, 2020), they often struggle to capture fine-scale features and changes in the sea ice.…”