Leif Toudal Pedersen

Mapping sea ice in the Arctic is essential for maritime navigation, and growing vessel traffic highlights the necessity of the timeliness and accuracy of sea ice charts. In addition, with the increased availability of satellite imagery, automation is becoming more important. The AutoICE Challenge investigates the possibility of creating deep learning models capable of mapping multiple sea ice parameters automatically from spaceborne synthetic aperture radar (SAR) imagery and assesses the current state of the automatic-sea-ice-mapping scientific field. This was achieved by providing the tools and encouraging participants to adopt the paradigm of retrieving multiple sea ice parameters rather than the current focus on single sea ice parameters, such as concentration. The paper documents the efforts and analyses, compares, and discusses the performance of the top-five participants’ submissions. Participants were tasked with the development of machine learning algorithms mapping the total sea ice concentration, stage of development, and floe size using a state-of-the-art sea ice dataset with dual-polarised Sentinel-1 SAR images and 22 other relevant variables while using professionally labelled sea ice charts from multiple national ice services as reference data. The challenge had 129 teams representing a total of 179 participants, with 34 teams delivering 494 submissions, resulting in a participation rate of 26.4 %, and it was won by a team from the University of Waterloo. Participants were successful in training models capable of retrieving multiple sea ice parameters with convolutional neural networks and vision transformer models. The top participants scored best on the total sea ice concentration and stage of development, while the floe size was more difficult. Furthermore, participants offered intriguing approaches and ideas that could help propel future research within automatic sea ice mapping, such as applying high downsampling of SAR data to improve model efficiency and produce better results.

Mapping sea ice in the Arctic is essential for maritime navigation, and growing vessel traffic highlights the necessity of timeliness and accuracy of sea ice charts. In addition, with the increased availability of satellite imagery, automation is becoming more important. The aim of the AutoICE Challenge was to encourage the creation of models capable of mapping sea ice automatically from spaceborne Synthetic Aperture Radar (SAR) imagery using deep learning while inspiring participants to move towards multiple sea ice parameter model retrieval instead of the current focus on a single sea ice parameter, such as concentration. Participants were tasked with the development of machine learning algorithms mapping the total sea ice concentration, stage of development and floe size using a state-of-the-art sea ice dataset with dual-polarised Sentinel-1 SAR images and 22 other relevant variables while using professionally labelled sea ice charts from multiple national ice services as reference data. The challenge had 129 teams representing a total of 179 participants, with 34 teams delivering 494 submissions, resulting in a participation rate of 26.4 %, and was won by a team from the University of Waterloo. Participants were successful in training models capable of retrieving multiple ice parameters with convolutional neural network and vision transformer models. The top participants scored best on the total sea ice concentration and stage of development, while the floe size was more difficult. Furthermore, participants offered intriguing approaches and ideas that could help propel future research within automatic sea ice mapping, such as applying high downsampling of SAR data to improve model efficiency and produce better results.

Mapping sea ice in the Arctic is essential for maritime navigation, and growing vessel traffic highlights the necessity of timeliness and accuracy of sea ice charts. In addition, with the increased availability of satellite imagery, automation is becoming more important. The aim of the AutoICE Challenge was to encourage the creation of models capable of mapping sea ice automatically from spaceborne Synthetic Aperture Radar (SAR) imagery using deep learning while inspiring participants to move towards multiple sea ice parameter model retrieval instead of the current focus on a single sea ice parameter, such as concentration. Participants were tasked with the development of machine learning algorithms mapping the total sea ice concentration, stage of development and floe size using a state-of-the-art sea ice dataset with dual-polarised Sentinel-1 SAR images and 22 other relevant variables while using professionally labelled sea ice charts from multiple national ice services as reference data. The challenge had 129 teams representing a total of 179 participants, with 34 teams delivering 494 submissions, resulting in a participation rate of 26.4 %, and was won by a team from the University of Waterloo. Participants were successful in training models capable of retrieving multiple ice parameters with convolutional neural network and vision transformer models. The top participants scored best on the total sea ice concentration and stage of development, while the floe size was more difficult. Furthermore, participants offered intriguing approaches and ideas that could help propel future research within automatic sea ice mapping, such as applying high downsampling of SAR data to improve model efficiency and produce better results.

Mapping sea ice in the Arctic is essential for maritime navigation, and growing vessel traffic highlights the necessity of timeliness and accuracy of sea ice charts. In addition, with the increased availability of satellite imagery, automation is becoming more important. The aim of the AutoICE Challenge was to encourage the creation of models capable of mapping sea ice automatically from spaceborne Synthetic Aperture Radar (SAR) imagery using deep learning while inspiring participants to move towards multiple sea ice parameter model retrieval instead of the current focus on a single sea ice parameter, such as concentration. Participants were tasked with the development of machine learning algorithms mapping the total sea ice concentration, stage of development and floe size using a state-of-the-art sea ice dataset with dual-polarised Sentinel-1 SAR images and 22 other relevant variables while using professionally labelled sea ice charts from multiple national ice services as reference data. The challenge had 129 teams representing a total of 179 participants, with 34 teams delivering 494 submissions, resulting in a participation rate of 26.4 %, and was won by a team from the University of Waterloo. Participants were successful in training models capable of retrieving multiple ice parameters with convolutional neural network and vision transformer models. The top participants scored best on the total sea ice concentration and stage of development, while the floe size was more difficult. Furthermore, participants offered intriguing approaches and ideas that could help propel future research within automatic sea ice mapping, such as applying high downsampling of SAR data to improve model efficiency and produce better results.

Mapping sea ice in the Arctic is essential for maritime navigation, and growing vessel traffic highlights the necessity of timeliness and accuracy of sea ice charts. In addition, with the increased availability of satellite imagery, automation is becoming more important. The aim of the AutoICE Challenge was to encourage the creation of models capable of mapping sea ice automatically from spaceborne Synthetic Aperture Radar (SAR) imagery using deep learning while inspiring participants to move towards multiple sea ice parameter model retrieval instead of the current focus on a single sea ice parameter, such as concentration. Participants were tasked with the development of machine learning algorithms mapping the total sea ice concentration, stage of development and floe size using a state-of-the-art sea ice dataset with dual-polarised Sentinel-1 SAR images and 22 other relevant variables while using professionally labelled sea ice charts from multiple national ice services as reference data. The challenge had 129 teams representing a total of 179 participants, with 34 teams delivering 494 submissions, resulting in a participation rate of 26.4 %, and was won by a team from the University of Waterloo. Participants were successful in training models capable of retrieving multiple ice parameters with convolutional neural network and vision transformer models. The top participants scored best on the total sea ice concentration and stage of development, while the floe size was more difficult. Furthermore, participants offered intriguing approaches and ideas that could help propel future research within automatic sea ice mapping, such as applying high downsampling of SAR data to improve model efficiency and produce better results.

We report on results of an intercomparison of 10 global sea-ice concentration (SIC) data products at 12.5 to 50.0 km grid resolution from satellite passive microwave (PMW) observations. For this we use SIC estimated from >350 images acquired in the visible–near-infrared frequency range by the joint National Aeronautics and Space Administration (NASA) and United States Geological Survey (USGS) Landsat sensor during the years 2003–2011 and 2013–2015. Conditions covered are late winter/early spring in the Northern Hemisphere and from late winter through fall freeze-up in the Southern Hemisphere. Among the products investigated are the four products of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Ocean and Sea Ice Satellite Application Facility (OSI SAF) and European Space Agency (ESA) Climate Change Initiative (CCI) algorithms SICCI-2 and OSI-450. We stress the importance to consider intercomparison results across the entire SIC range instead of focusing on overall mean differences and to take into account known biases in PMW SIC products, e.g., for thin ice. We find superior linear agreement between PMW SIC and Landsat SIC for the 25 and the 50 km SICCI-2 products in both hemispheres. We discuss quantitatively various uncertainty sources of the evaluation carried out. First, depending on the number of mixed ocean–ice Landsat pixels classified erroneously as ice only, our Landsat SIC is found to be biased high. This applies to some of our Southern Hemisphere data, promotes an overly large fraction of Landsat SIC underestimation by PMW SIC products, and renders PMW SIC products overestimating Landsat SIC particularly problematic. Secondly, our main results are based on SIC data truncated to the range 0 % to 100 %. We demonstrate using non-truncated SIC values, where possible, can considerably improve linear agreement between PMW and Landsat SIC. Thirdly, we investigate the impact of filters often used to clean up the final products from spurious SIC over open water due to weather effects and along coastlines due to land spillover. Benefiting from the possibility to switch on or off certain filters in the SICCI-2 and OSI-450 products, we quantify the impact land spillover filtering can have on evaluation results as shown in this paper.

Convolutional neural networks (CNNs) are being increasingly investigated as a means to extract sea ice concentration from synthetic aperture radar (SAR) in an automated manner. This is often done using ice charts as training data. However, in these charts, an ice concentration label is given to a large region, which may not have a spatially uniform sea ice concentration distribution at the prediction scale of the CNN. This leads to representativity errors, which can be more pronounced at intermediate sea ice concentrations. In this study, we first investigate ways to perturb the ice chart labels to obtain improved predictions to account for the label uncertainty for intermediate ice concentrations. We then propose a method to augment the ice chart data by rescaling the information in the SAR imagery. The method is found to lead to improved accuracy in comparison to using the ice chart labels alone, with accuracy improving from 0.919 to 0.966. The sea ice concentration maps with the augmented labels also have much finer detail than the other approaches evaluated. These details are visually in agreement with expected sea ice concentration from the SAR data.

Sea ice concentration algorithms using brightness temperatures ( $T_{B}$ ) from satellite microwave radiometers are used to compute sea ice concentration ( $c_{\text{ice}}$ ), sea ice extent, and generate sea ice climate data records. Therefore, it is important to minimize the sensitivity of $c_{\text{ice}}$ estimates to geophysical noise caused by snow/sea ice thermal microwave emission signature variations, and presence of WV and clouds in the atmosphere and/or near-surface winds. In this study, we investigate the effect of geophysical noise leading to systematic $c_{\text{ice}}$ biases and affecting $c_{\text{ice}}$ standard deviations (STD) using simulated top of the atmosphere $T_{B}$ s over open water and 100% sea ice. We consider three case studies for the Arctic and the Antarctic and eight different $c_{\text{ice}}$ algorithms, representing different families of algorithms based on the selection of channels and methodologies. Our simulations show that, over open water and low $c_{\text{ice}}$ , algorithms using gradients between V-polarized 19-GHz and 37-GHz $T_{B}$ s show the lowest sensitivity to the geophysical noise, while the algorithms exclusively using near-90-GHz channels have by far the highest sensitivity. Over sea ice, the atmosphere plays a much smaller role than over open water, and the $c_{\text{ice}}$ STD for all algorithms is smaller than over open water. The hybrid and low-frequency (6 GHz) algorithms have the lowest sensitivity to noise over sea ice, while the polarization type of algorithms has the highest noise levels.

We report on results of an inter-comparison of 10 global sea-ice concentration (SIC) data products at 12.5 to 50.0 km grid resolution from satellite passive microwave (PMW) observations. For this we use SIC estimated from > 300 images acquired in the visible / near-infrared frequency range by joint the National Aeronautics and Space Administration (NASA)/United States Geological Survey (USGS) Landsat sensor during years 2003–2011 and 2013–2015. Conditions covered are late winter / early spring in the Northern Hemisphere and from late winter through fall freeze-up in the Southern Hemisphere. Among the products investigated are the four products of the European Organisation for the Exploitation of Meteorological Satellites (EUMETSAT) Ocean and Sea Ice Satellite Application Facility (OSI SAF) and European Space Agency (ESA) Climate Change Initiative (CCI) algorithms: SICCI-2 and OSI-450. We stress the importance to consider inter-comparison results across the entire SIC range instead of focusing on overall mean differences, and to take in account known biases in PMW SIC products, e.g. for thin ice. We find superior linear agreement between PMW SIC and Landsat SIC for the 25 km and the 50 km SICCI-2 products in both hemispheres. We discuss quantitatively various uncertainty sources of the evaluation carried out. First, depending on the number of mixed ocean-ice Landsat pixels classified erroneously as ice only, our Landsat SIC is found to be biased high. This applies to some of our Southern Hemisphere data, promotes an overly large fraction of Landsat SIC under-estimation by PMW SIC products, and renders PMW SIC products overestimating Landsat SIC particularly problematic. Secondly, our main results are based on SIC data truncated to the range 0 % to 100 %. We demonstrate using non-truncated SIC values, where possible, can considerably improve linear agreement between PMW and Landsat SIC. Thirdly, we investigate the impact of filters often used to clean up the final products from spurious SIC over open water due to weather effects and along coastlines due to land spillover. Benefiting from the possibility to switch on or off certain filters in the SICCI-2 and OSI-450 products we quantify the impact land spillover filtering can have on evaluation results as shown in this paper.