Background
Manual ice charting from multi-sensor satellite data analysis has been for many years the primary method at the National Ice Services for producing sea ice information for marine safety. Ice analysts primarily use satellite synthetic aperture radar (SAR) imagery due to the high spatial resolution and the capability to image the surface through clouds and in polar darkness, but also optical imagery in clear sky and daylight conditions, thermal-infrared and microwave radiometer data from e.g. AMSR2.
The traditional manual ice charting method is time-consuming and limited in spatial and temporal coverage. Further, it is challenged by an increasing amount of available satellite imagery, along with a growing number of users accessing wider parts of the Arctic due to the thinning of the Arctic sea ice.
The automation of the time-consuming and labor-intensive sea ice charting process has the potential to provide users with near-real-time sea ice products of higher spatial resolution, larger spatial and temporal coverage, and increased consistency.
Convolutional Neural Network (CNN) has great potential within automated prediction of sea ice in satellite images. Automating the process on SAR data alone is challenging. SAR images show patterns related to ice formations, but backscatter intensities can be ambiguous, complicating the discrimination between ice and open water, e.g. at high wind speeds. To tackle the challenges, the training dataset made available in this challenge contains Sentinel-1 active microwave data and corresponding Microwave Radiometer (MWR) data from AMSR2, to enable challenge participants to exploit the advantages of both instruments. While SAR data has ambiguities, it has a high spatial resolution, whereas MWR data has good contrast between open water and ice. However, the coarse resolution of the AMSR2 MWR observations introduces a new set of obstacles, e.g. land spill-over, which can lead to erroneous sea ice predictions along the coastline adjacent to open water.
Data
The challenge training dataset contained Sentinel-1 active microwave (SAR) data and corresponding Microwave Radiometer (MWR) data from the AMSR2 satellite sensor. Label data in the challenge datasets were ice charts produced by the Greenland ice service at the Danish Meteorological Institute (DMI) and the Canadian Ice Service (CIS). The challenge datasets also contained other auxiliary data such as numerical weather prediction model data. The dataset consisted of 493 training and 20 test (without label data) data files.
Computing Resources
To help participants get started with the challenge, a Jupyter notebook (Starter Pack) was prepared to assist with data I/O, visualization, prediction using a baseline algorithm, and creating a valid submission.
Machine learning computing resources were available to challenge participants on Polar TEP. Polar TEP (polartep.polarview.org) provides a complete working environment where users can access algorithms and data remotely, providing computing resources and tools that they might not otherwise have, and avoiding the need to download and manage large volumes of data. Polar TEP has implemented the MLflow platform to support machine learning activities. MLflow is an open-source platform to manage all stages of the ML lifecycle, including experimentation, reproducibility, deployment, and a central model registry.
Evaluation
The AutoICE challenge took place online from November 21, 2022 to April 17, 2023.
Challenge participants were tasked to submit their model results for each of the three sea ice parameters: sea ice concentration (SIC), stage-of-development (SOD), and floe size. SIC performance was evaluated by calculating the R2 coefficient. SOD and floe size performances were evaluated using their F1 score. The three sea ice parameter scores were combined into one single final score using the following weighting scheme: SIC 40%, SOD 40%, Floe Size 20%.
Prizes
The top-five winners of the challenge were offered a prize from AI4EO sponsors and wereinvited to present their winning solution to the challenge expert team. The Winners event was held as a half-day virtual workshop following the challenge closure.