Time series prediction of gridded sea ice concentration using feature-engineered machine learning method

Other conference contribution, 2025

The rapid decline of Arctic Sea ice has significant implications for ship activities in the Arctic area. Accurately forecasting sea ice information is crucial for the safe and efficient operation of ships in the Arctic. Traditional physical sea ice-ocean models struggle to handle some applications, such as ship route planning, which need ice forecast for large-scale grids at a high spatial resolution, due to their computational complexity and reliance on reliability of initial state and boundary conditions. To address these limitations, we propose a feature-engineered machine learning method for time series forecasting of Arctic Sea ice concentration (SIC) for large amounts of geographic grids. The proposed method utilizes a combination of statistical and machine learning techniques to extract and analyze relevant features from historical SIC data and hydrological variables. These features are then used to train a LightGBM (Light Gradient Boosting Machine) model to predict future SIC values. The average root mean square error (RMSE) of different forecast horizons and geographic grids is analyzed. The performance of different prediction time windows is obtained through simulations. These results suggest that the proposed method can provide valuable insights into Arctic SIC forecast and support effective ship route planning in the Arctic, particularly for handling large amounts of geographic grids.