CasEm: A Cascade Architecture for Long-Horizon Neural Emulation

2026-09-28

AI Summary

This study addresses the drift and divergence issues of autoregressive neural simulators in long-horizon predictions by proposing a feedback-free cascaded simulation architecture. The method leverages physically aggregated models, such as regional total water mass, to unidirectionally guide a full-state backbone network, such as a Fourier Neural Operator, in correcting prediction errors. Furthermore, finite-horizon error bounds are derived to inform subsystem selection. Experimental results demonstrate that this architecture significantly reduces long-term prediction errors across multiple benchmarks, achieving over a 46% reduction in decade-averaged error for climate simulations.

Abstract

Autoregressive neural emulators can drift or diverge over long rollouts despite accurate short-term predictions. We introduce Cascaded Emulation (CasEm), a one-way rollout architecture that augments an existing full-state backbone with an independently evolving model of physically specified aggregates. Its forecasts guide corrections to full-state predictions, without feedback from the backbone to the aggregate model. Effective guidance requires aggregates that cover substantial backbone error, remain accurately predictable, and support useful full-state corrections. We derive a finite-horizon error bound that clarifies these three factors and use empirical diagnostics to guide subsystem selection. Across four ODE/PDE benchmarks, CasEm reduces long-horizon rollout errors across diverse backbones and suppresses the trend toward error divergence in both diffusion tasks using Fourier neural operator backbones. In global climate emulation, CasEm with a regional total-water subsystem reduces 10-year full-state time-mean error by 66.6% and 46.3% for frozen ACE and Spherical DYffusion backbones, respectively, while adding less than 3% to inference time.

Problem

• neural emulation
• long-horizon rollout
• error drift
• autoregressive model
• error divergence

Innovation

• Cascade Architecture
• Neural Emulation
• Long-Horizon Rollout
• Error Bound
• Autoregressive Models