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

TL;DR

This paper demonstrates that machine learning surrogate models like FourCastNet can be effectively used for long-term weather data assimilation, maintaining accuracy over extended periods despite model instability and observational sparsity.

Contribution

It introduces a novel integration of FourCastNet within a data assimilation framework, showing long-term accuracy and stability in weather forecasting.

Findings

Filtering estimates remain accurate over a year-long window.

Surrogate models can provide effective initial conditions for forecasting.

Long-term stability is achievable despite model and observational challenges.

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.