IceGAT: Arctic Sea Ice Prediction based on Multi-scale Graph Modeling with Conservation Laws

Authors: Lan Wei and Nikolaos M. Freris

Description

This is the online forecast from IceGAT, a neural network for predicting the sea ice concentration in the Arctic. Using the daily updated data from the ERA5 dataset (with a lag of 5 days), we plot the ground-truth for four days, followed by the prediction for one week. When you click the button below the figure, you can see the visualization of the ice for each day along with the numbers for sea ice extent (SIE), sea ice area (SIA), and the corresponding changes (Δ) between consecutive days.

Definitions: 1. SIC is the sea ice concentration ([0,1]; see the color bar). 2. SIE is the integral sum of the areas with at least 15% sea ice concentration. 3. SIA is the total sea ice area (excluding regions below 15% sea ice concentration). 4. Δ is the change from the previous day to the current.

Visualization

(Forecasts updated on Apr 28, 2025, 08:00:44 AM (GMT+8))

The latest data provided by ECMWF ERA5 Dataset are until Date, 11:00 PM (GMT+0).

Supplementary Video

Code

https://github.com/LannWei/IceGAT

Publication

https://agupubs.onlinelibrary.wiley.com/doi/10.1029/2024JD042136

Contact

If you have any questions, feel free to ask us via: nfr@ustc.edu.cn or weilan@mail.ustc.edu.cn.