![[2602.19406] LEVDA: Latent Ensemble Variational Data Assimilation via Differentiable Dynamics](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.19406] LEVDA: Latent Ensemble Variational Data Assimilation via Differentiable Dynamics
Summary
The paper presents LEVDA, a novel ensemble-space variational smoother for geophysical forecasting that improves data assimilation by operating in a low-dimensional latent space, enhancing accuracy and efficiency.
Why It Matters
This research addresses the challenges of chaotic dynamics and numerical errors in long-range geophysical forecasts. By introducing LEVDA, the authors provide a more efficient method for data assimilation, which is crucial for improving predictive models in various scientific fields, including climate science and meteorology.
Key Takeaways
- LEVDA operates in a low-dimensional latent space, enhancing computational efficiency.
- It assimilates states and unknown parameters without requiring adjoint code.
- The method accommodates irregular sampling at arbitrary spatiotemporal locations.
- LEVDA outperforms state-of-the-art methods under observational sparsity.
- Improved uncertainty quantification is achieved alongside assimilation accuracy.
Computer Science > Machine Learning arXiv:2602.19406 (cs) [Submitted on 23 Feb 2026] Title:LEVDA: Latent Ensemble Variational Data Assimilation via Differentiable Dynamics Authors:Phillip Si, Peng Chen View a PDF of the paper titled LEVDA: Latent Ensemble Variational Data Assimilation via Differentiable Dynamics, by Phillip Si and 1 other authors View PDF HTML (experimental) Abstract:Long-range geophysical forecasts are fundamentally limited by chaotic dynamics and numerical errors. While data assimilation can mitigate these issues, classical variational smoothers require computationally expensive tangent-linear and adjoint models. Conversely, recent efficient latent filtering methods often enforce weak trajectory-level constraints and assume fixed observation grids. To bridge this gap, we propose Latent Ensemble Variational Data Assimilation (LEVDA), an ensemble-space variational smoother that operates in the low-dimensional latent space of a pretrained differentiable neural dynamics surrogate. By performing four-dimensional ensemble-variational (4DEnVar) optimization within an ensemble subspace, LEVDA jointly assimilates states and unknown parameters without the need for adjoint code or auxiliary observation-to-latent encoders. Leveraging the fully differentiable, continuous-in-time-and-space nature of the surrogate, LEVDA naturally accommodates highly irregular sampling at arbitrary spatiotemporal locations. Across three challenging geophysical benchmarks, LEVDA matches...
