Regional Sea Ice predictions in northern Norway, Svalbard and the Barents Sea using Deep Learning
Are Kvanum
The Norwegian Meteorological Institute develops short range sea ice information with two goals. Firstly, short range sea ice products provide valuable information for Arctic navigators. Secondly, sea ice is essential to force numerical weather prediction systems at high latitudes. This presentation will explore both sea ice goals of the Norwegian Meteorological Institute under the framing of recent achievements in machine learning developments for sea ice forecasting. We show that deep learning sea ice concentration prediction systems are able to outperform baseline and physical forecasts. Additionally, we highlight the importance of supplying numerical weather prediction systems with reliable, up to date, sea ice information. Therefore, we are evaluating the added value of using data-driven sea ice forecasting for tactical navigation in the Arctic, as well as a forcing in the numerical weather prediction system developed at the Norwegian Meteorological Institute for northern Norway, Svalbard and the Barents Sea. We expand upon previous work which have established that Arctic short-range weather forecasts are particularly sensitive to the sea ice cover by demonstrating the potential synergies and positive benefits combining physical weather forecasts with up to date deep learning sea ice forecasts.