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

2026-05-28

AI Summary

This work addresses the challenges of long-term forecasting in geophysical systems—namely, the complexity of nonlinear dynamics, the computational expense of full-physics simulations, and the error accumulation and instability inherent in autoregressive models. To overcome these issues, the authors propose a unified multi-step graph neural network surrogate model that represents spatial locations as graph nodes and local interactions as edges. Leveraging a shared backbone with multiple output branches, the model incorporates state-increment prediction and a coarse-to-fine rollback mechanism to directly and parallelly forecast multiple future time steps from the current state, thereby effectively mitigating long-term prediction drift. Evaluated on decadal-scale simulations of Pine Island Glacier, the approach significantly outperforms both initial-state baselines and single-step autoregressive methods in accuracy and stability, offering an efficient and reliable surrogate modeling framework for climate and sea-level change research.

Abstract

Accurate long-range prediction of geophysical systems is difficult due to strongly nonlinear dynamics, the high computational cost of full-physics simulations, and the error accumulation that arise when one-step autoregressive surrogates are rolled out over decades. Deep neural network can serve as efficient emulators, but most are trained only for next-step prediction and often drift or become unstable as the forecast horizon grows. We propose a multi-horizon graph neural network emulator that learns state-to-state transitions from a single current time to multiple future lead times within one unified model. The physical domain is represented as a graph, where nodes correspond to spatial locations with time-varying geophysical attributes and edges encode local spatial interactions. Given the current graph state, the model predicts the future evolution of key fields, ice thickness and ice velocities at all nodes, using a shared graph backbone with separate output branches for each target variable. To improve stability, the network predicts state increments relative to the current state, which are then added back to reconstruct future states. Training jointly optimizes all lead times with a unified regression objective, and inference uses a coarse-to-fine rollout that advances with larger jumps and selectively refines with shorter jumps to reduce drift and avoid redundant computation. Experiments on multi-decadal Pine Island Glacier simulations show that our approach achieves higher long-range accuracy and improved stability than both (i) an initial-state baseline that predicts each future time directly from the starting state and (ii) a standard single-step autoregressive rollout, producing a more reliable emulator for downstream climate and sea-level studies.

Problem

• long horizon forecasting
• geophysical systems
• error accumulation
• model drift
• stability

Innovation

• multi-horizon forecasting
• graph neural networks
• incremental prediction
• coarse-to-fine rollout
• geophysical emulation