Kalman filter based data assimilation system to improve numerical sea ice predictions in the Arctic Ocean

Abstract

With the recent changes in the Arctic climate, increased transportation can be observed in the Arctic Ocean. For safe navigation along the Arctic Sea routes, it is important to accurately predict the ice conditions. In this study the iceocean coupled Ice-POM model is improved by a Kalman filter based data assimilation system. This system incorporates sea ice observation data such as sea ice concentration, sea ice thickness and sea ice velocity to improve the numerical predictions. Ocean part of the model is based on the Princeton Ocean Model while the Ice model considers the discrete characteristics of ice along the ice edge. In an ice-ocean coupled model, atmospheric forcing directly affects the accuracy of predictions. However, different atmospheric data sets produced by different weather agencies show large differences in the Arctic region. Model errors largely depend upon the inaccuracies in forcing data. This study uses an ensemble of multiple atmospheric data sets collected from different weather agencies and the spread of the ensemble is taken as an indicator of the model error covariance. The Observation errors were varied according to the location and the season. Assimilation has improved the predictions of sea ice variables. It has also indirectly improved the ocean conditions. This Atmospheric forcing based Kalman filter (AFKF) method outperforms other assimilation methods such as direct assimilation and nudging methods.