MET-AICE v1.0: An operational data-driven sea ice prediction system for the European Arctic

Abstract

There is an increasing need for reliable short-term sea ice forecasts that can support maritime operations in polar regions. While numerous studies have shown the potential of machine learning for sea ice forecasting, there are currently only a few operational data-driven sea ice prediction systems. Here, we introduce MET-AICE, a prediction system providing sea ice concentration forecasts for the next 10 d in the European Arctic. To our knowledge, it is the first operational data-driven prediction system designed for short-term sea ice forecasting. MET-AICE has been trained to predict sea ice concentration observations from the Advanced Microwave Scanning Radiometer 2 (AMSR2) at 5 km resolution. After one year of operation, we show that MET-AICE considerably outperforms persistence of AMSR2 observations (errors about 30 % lower on average), as well as forecasts from several dynamical models such as TOPAZ5, Barents-2.5 km and the European Centre for Medium-Range Weather Forecasts (ECMWF) Integrated Forecasting System.