From Short Histories to Long Futures: Horizon-Aware Graph Neural Networks for Long Horizon Forecasting

A new study introduces a multi-horizon graph neural network emulator designed to enhance long-range predictions of geophysical systems, addressing challenges such as nonlinear dynamics and computational costs. This model learns state-to-state transitions from a single current time to multiple future lead times, effectively predicting the evolution of key fields like ice thickness and velocities at various nodes.

This development is significant as it offers a more stable and efficient approach to long-term forecasting, which is crucial for understanding and managing the impacts of climate change, particularly in sensitive areas like the Pine Island Glacier.