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) for long-term physical forecasting on irregular geospatial meshes, presented by Zesheng Liu and Maryam Rahnemoonfar (arXiv:2606.11162). The model uses a graph-based history encoder to process a finite history of system states along with forcing fields and external forcings, producing node-wise context vectors that capture both local spatial interactions and temporal evolution. These context vectors initialize and condition a latent Neural ODE whose dynamics are driven by interpolated future forcings and explicit relative rollout time. Because the forecast trajectory is modeled as a continuous latent dynamical system, COGENT can generate predictions at arbitrary future times instead of being limited to a fixed temporal discretization. A residual decoder maps the latent trajectory back to physical states, enabling direct multi-step prediction. Evaluated on transient simulations of an ice-sheet–sea-level system model, COGENT shows improved long-range stability compared to autoregressive graph baselines.

Overview

COGENT (Continuous Graph Emulators with Neural Ordinary Differential Equations) targets long-term physical forecasting on irregular geospatial meshes.

Authors: Zesheng Liu, Maryam Rahnemoonfar
Field: Machine Learning
arXiv: 2606.11162
Posted: 2026-06-09

Key Ideas

Graph-based history encoder: encodes a finite history of system states along with forcing fields and external forcings, producing 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 ODE whose dynamics are driven by interpolated future forcings and an explicit relative rollout time.
Continuous-time 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 the latent trajectory back to physical states, enabling direct multi-step prediction instead of purely autoregressive rollout.

Evaluation

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

Original Abstract (excerpt)

> In this work, we present COGENT, a continuous graph emulator with Neural Ordinary Differential Equations for long-term physical forecasting on irregular geospatial meshes. COGENT encodes a finite history of system states and associated forcing fields and external forcings with a graph-based history encoder, producing node-wise context vectors that capture both local spatial interactions and temporal evolution. These context vectors initialize and condition a latent Neural Ordinary Differential Equation whose dynamics are driven by interpolated future forcings and explicit relative rollout time. 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...

---

*Auto-collected on 2026-06-11.*