PnP-Corrector: A Universal Correction Framework for Coupled Spatiotemporal Forecasting

2026-05-09

AI Summary

This work addresses the challenge of long-term prediction collapse in coupled spatiotemporal forecasting, which arises from the mutual amplification of simulator errors across interacting subsystems—a phenomenon termed “feedback error amplification.” To mitigate this issue, the authors propose the PnP-Corrector framework, which decouples physical simulation from error correction for the first time. Building upon a frozen pre-trained physics engine, the framework introduces a trainable, plug-and-play correction agent that actively compensates for systematic biases in the coupled system. Leveraging the DSLCAST architecture as its core predictive model, the method effectively suppresses feedback error amplification. Evaluated on a 300-day global ocean–atmosphere coupled forecasting task, it reduces prediction error by 29% relative to the baseline and outperforms existing state-of-the-art approaches across multiple key metrics.

Abstract

Coupled spatiotemporal forecasting is important for predicting the future evolution of multiple interacting dynamical systems, such as in climate models. However, existing methods are severely constrained by the persistent bottleneck of compounding errors. In coupled systems, errors from each subsystem simulator propagate and amplify one another, a phenomenon we term Reciprocal Error Amplification, leading to a rapid collapse of long-range predictions. To address this challenge, we propose a universal framework called PnP-Corrector (Plug-and-Play Corrector). The core idea of our framework is to decouple the physical simulation from the error correction process: it freezes pre-trained physics simulation engines and exclusively trains a correction agent to proactively counteract the systematic biases emerging from the coupled system. Furthermore, we design an efficient predictive model architecture, DSLCast, to serve as the backbone of this framework. Extensive experiments demonstrate that our method significantly enhances the long-term stability and accuracy of coupled forecasting systems. For instance, in the challenging task of a 300-day global ocean-atmosphere coupled forecast, our PnP-Corrector framework reduces the prediction error of the baseline model by 29% and surpasses state-of-the-art models on several key metrics.

Problem

• coupled spatiotemporal forecasting
• compounding errors
• Reciprocal Error Amplification
• long-range prediction
• error propagation

Innovation

• PnP-Corrector
• Coupled Spatiotemporal Forecasting
• Reciprocal Error Amplification
• Error Correction Framework
• DSLCast