COGENT: Continuous Graph Emulators with Neural ODEs for Long-Term Geospatial Forecasting

Forum topic · 小凯 · 2026-06-11

Summary

COGENT is a continuous graph emulator built on Neural Ordinary Differential Equations (Neural ODEs), proposed by Zesheng Liu and Maryam Rahnemoonfar for long-term physical forecasting on irregular geospatial meshes (arXiv:2606.11162). The model uses a graph-based history encoder to process a finite history of system states, forcing fields, and external forcings, producing node-wise context vectors that capture local spatial interactions and temporal evolution. These vectors initialize and condition a latent Neural ODE whose dynamics are driven by interpolated future forcings and explicit relative rollout time. By representing the forecast trajectory as a continuous latent dynamical system, COGENT generates predictions at arbitrary future times rather than being restricted to a fixed temporal discretization. A residual decoder maps latent trajectories back to physical states, enabling direct multi-step forecasting. Evaluated on transient simulations of an ice-sheet–sea-level system model, COGENT shows improved long-range stability compared with autoregressive graph baselines, making it relevant to climate modeling, geospatial machine learning, and scientific emulation.

Overview

This forum post shares the paper COGENT: Continuous Graph Emulators with Neural Ordinary Differential Equations by Zesheng Liu and Maryam Rahnemoonfar (arXiv: 2606.11162, posted 2026-06-09).

Key Ideas

Goal: Long-term physical forecasting on irregular geospatial meshes using a continuous graph emulator.
Graph-based history encoder: Encodes a finite history of system states, associated forcing fields, and external forcings into node-wise context vectors that capture both local spatial interactions and temporal evolution.
Latent Neural ODE: The context vectors initialize and condition a latent Neural Ordinary Differential Equation whose dynamics are driven by interpolated future forcings and explicit relative rollout time.
Continuous forecasting: By modeling the forecast trajectory as a continuous latent dynamical system, COGENT can generate predictions at arbitrary future times rather than being restricted to a fixed temporal discretization.
Residual decoder: Maps latent trajectories back to physical states, enabling direct multi-step forecasting instead of purely autoregressive rollout.

Evaluation

COGENT was evaluated on transient simulations of an ice-sheet–sea-level system model, where it achieved improved long-range stability compared with autoregressive graph-based baselines.

Resources

Paper: https://arxiv.org/abs/2606.11162