Multimodal Spatiotemporal Atmospheric Data Assimilation with Latent Flow-matching
2026-08-05 • Machine Learning
Machine Learning
AI summaryⓘ
The authors propose a new way to improve weather predictions by combining past data with current observations using a method called latent video flow-matching. They train their system on a large atmospheric dataset (ERA5) to create smooth weather scenarios over time. By integrating real-world measurements, their approach can update forecasts more accurately and handle missing data naturally. Their method can perform common data assimilation tasks like filtering and smoothing by simply choosing which times to observe, and it produces forecasts that compete with leading models.
Data assimilationBayesian inferenceLatent video flow-matchingERA5 reanalysisPosterior samplingNOAA Radiosonde ArchiveIntegrated Surface DatabaseFilteringSmoothingEnsemble forecasting
Authors
Dibyajyoti Chakraborty, Romit Maulik
Abstract
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.