DigitalSeaIce: Multi-scale integration and digitalization of Arctic sea ice observations and prediction models

Digital Sea Ice – a Chinese-Norwegian research collaboration on precise forecasting of ice conditions in the Arctic

A strong research collaboration between the universities NTNU, Jiangsu, and Dalian and the Ice Service at MET will give better insight in polar Earth systems that can be useful both in predicting ice conditions and improving maritime safety in the Arctic Ocean.

The Norwegian Research Council has granted 9,99 million NOK to a project on multi-scale integration and digitalization of sea ice observations and prediction models in the Arctic Ocean. The grant is awarded to the Norwegian partners in an academic research collaboration between Norway and China.

The project is in cooperation with the Norwegian Meteorological Institute and two Chinese partners, namely Jiangsu University of Science and Technology (JUST) and Dalian University of Technology (DLUT).

The primary objective is to build a multiscale digital infrastructure that connects sea ice forecasting models on a regional scale with local and more detailed ice-ice and ice-structure discrete element models. These models are then updated by in-situ and shipboard measurements locally and regionally by satellite measurements. This will enable improved spatial and temporal resolution in our models, to achieve more precise forecasting of ice conditions in the Arctic – including better understanding of long-term variations in the polar ice cover. Novel methods for use of artificial intelligence (AI)-based analytics of synthetic aperture radar (SAR) and optical imagery from satellites, marine radars, visual and infrared cameras, and other enabling technologies will be developed in the project.

The secondary objectives are to accurately map the sea ice flow in high resolution and generate quality-controlled sea ice drift forecasting. Novel methods for monitoring and analysis of sea ice dynamics and fracturing processes based on data from heterogenous sources will be developed. This will be used to update the multiscale model from the real observations.

The expected impact is novel methods and a digital infrastructure for improved spatial and temporal forecasting and decision support in an increasingly dynamic Arctic environment due to climate changes. Such infrastructure will enable more accurate data and information to be produced, thus resulting in better insight on polar Earth systems. A biproduct is improved decision support for maritime safety.

Background: The Arctic region plays a key role in regulating the world's climate and is the region most affected by the ongoing climate change. Multiscale approach to modelling natural phenomena is a powerful technique to analyze, visualize, and forecast what is happening in the Arctic. There is a potential of AI-infused multiscale modelling of Arctic sea ice to supplement traditional remote sensing and climate models in the polar regions.

Goal: To build a multiscale digital method and system that integrates regional sea ice forecasting models and local ice-ice/ice-structure numerical models with in-situ, shipboard, and regional Arctic sea ice and environmental observations. The aim is to enable improved spatial and temporal resolution to achieve more precise forecasting of ice conditions in the Arctic – including better understanding of long-term variations in polar ice cover.

A common methodology to achieve this is the use of AI-based analytics of SAR and optical imagery from satellites, marine radars, and visual and infrared cameras.

The most advanced method for detecting all individual ice floes identifies individual ice floes, one by one, in the image. This method of estimating FSD may take from minutes to hours. Nowadays, pixel-based deep learning (DL) methods have proven to deliver superior accuracy in many image segmentation tasks. The encoder-decoder network architectures, e.g., U-Net and its developments, can achieve good results in semantic image segmentation, even with limited training dataset, and have the potential to map complex features at the pixel level from satellite imagery in an automated process, e.g., detecting individual ice floes (which is instance image segmentation) by identifying their boundaries. A processing chain for S2 tiles will be developed to implement cloud-masking and ice/water classification to yield high resolution mapping of ice floes and water features. From this, individual ice floe outlines will be extracted, and the data volume is further reduced by vectorization. The vectorized dataset will be used to derive a FSD for each S2 tile and will be used as initialization and validation for the regional modelling (WP4). The routine processing will allow analysis of temporal changes in the FSD due to regional ice drift, and weather events that result in wave-induced fracturing of the floes.

A number of short-range forecasts are currently evaluated routinely for sea ice edge accuracy at MET. This include MET's Barents-2.5km, the Copernicus neXtSIM, and US NRL GOFS3.1. The evaluation will be expanded to investigate the accuracy and skill of ice drift. This will utilize buoy tracks from the International Arctic Buoy Programme (IABP), plus field data acquired during the course of the project. This will be expanded to evaluate past conditions contained in ECMWF ERA5 and NASA MERRA-2 reanalysis products.