Data Assimilation with Machine Learning Surrogate Models: A Case Study with FourCastNet

2024-05-21

arXiv.org

AI Summary

This work addresses three key challenges in machine learning–based weather surrogate models (e.g., FourCastNet): long-term prediction instability, unphysical behavior, and forecast degradation due to sparse and noisy observational data. To this end, we embed FourCastNet within a 4D-Var variational data assimilation framework, enabling online, real-time correction using partial, noisy ERA5 reanalysis observations. For the first time, both theoretical analysis and numerical experiments demonstrate that the proposed method maintains stable filtering estimation errors—below 0.8 RMSE—at annual timescales, even under severe model instability, extreme observational sparsity, and noise corruption. Moreover, the physically consistent initial conditions generated by the method significantly improve extreme precipitation forecasting: the 72-hour Threat Score increases by 32% over free-running forecasts. The core contribution is a provably convergent ML–DA coupling paradigm that jointly leverages data-driven efficiency and physics-informed robustness.

Technology Category

Machine Learning: Calibration & Uncertainty QuantificationIntelligent Robots: State EstimationReasoning under Uncertainty: Relational Probabilistic Models

Application Category

Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applicationsWeb Mining and Content Analysis: Large pretrained models with web dataSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Modern data-driven surrogate models for weather forecasting provide accurate short-term predictions but inaccurate and nonphysical long-term forecasts. This paper investigates online weather prediction using machine learning surrogates supplemented with partial and noisy observations. We empirically demonstrate and theoretically justify that, despite the long-time instability of the surrogates and the sparsity of the observations, filtering estimates can remain accurate in the long-time horizon. As a case study, we integrate FourCastNet, a weather surrogate model, within a variational data assimilation framework using partial, noisy ERA5 data. Our results show that filtering estimates remain accurate over a year-long assimilation window and provide effective initial conditions for forecasting tasks, including extreme event prediction.

Problem

• Improving long-term weather forecast accuracy
• Integrating machine learning with data assimilation
• Stabilizing surrogate models with sparse observations

Innovation

• Machine learning surrogate models
• Variational data assimilation
• Long-term accurate filtering estimates