Search papers, labs, and topics across Lattice.
This study introduces a novel approach to atmospheric data assimilation by leveraging latent video flow-matching to create temporally consistent trajectories from a trained prior using ERA5 reanalysis data. By employing posterior sampling with real observational data, the method enables seamless information propagation between observed and unobserved frames, facilitating various data assimilation tasks such as filtering and smoothing. The results show that this approach can produce full-state ensemble forecasts from sparse observations, achieving performance on par with leading observation-to-forecast models.
Latent video flow-matching enables atmospheric data assimilation that not only propagates information seamlessly but also competes with state-of-the-art forecasting models using sparse data.
Data assimilation (DA) uses Bayesian inference to update the state of a numerical forecast model with observed data. In this study, we propose a fundamentally different, unified approach to atmospheric data assimilation. We use latent video flow-matching to sample temporally consistent trajectories from a prior trained using ERA5 reanalysis (69 variables over an 8-day window). We also use posterior sampling to assimilate real observation sources, such as those from the NOAA Integrated Global Radiosonde Archive and the Integrated Surface Database. Because the prior generates a continuous trajectory, it naturally propagates information between observed and unobserved frames. Therefore, we can perform various DA tasks, such as filtering and smoothing, simply by changing the observed frames. Moreover, we generate full-state ensemble forecasts directly from sparse observations, achieving performance competitive with state-of-the-art observation-to-forecast models.
