Physics-Informed Neural Networks as Differentiable Surrogates for 4D-Variational Data Assimilation
Accepted for publication in Journal of Advances in Modeling Earth Systems (JAMES), 2026
This work presents a machine-learning approach to 4D-Variational Data Assimilation in which a differentiable neural surrogate provides model sensitivities through automatic differentiation, eliminating the need for separately developed tangent-linear and adjoint models.
The manuscript has been accepted for publication in the Journal of Advances in Modeling Earth Systems (JAMES). A preprint is available on ESS Open Archive.
Recommended citation: Egan, K. & Powell, B. (2026). "Physics-Informed Neural Networks as Differentiable Surrogates for 4D-Variational Data Assimilation." Journal of Advances in Modeling Earth Systems. Accepted for publication.
Download Paper