Reparameterizing 4DVAR with neural fields

Jaemin Oh

Keywords: data assimilation, physics-informed neural networks, neural fields, spectral bias

TL;DR: We reparameterize 4DVAR with neural fields, exploiting spectral bias for stability and physics-informed losses for parallel-in-time optimization, all without requiring ground-truth data.

Abstract: Four-dimensional variational data assimilation (4DVAR) is a cornerstone of numerical weather prediction, but its cost function is difficult to optimize and computationally intensive. We propose a neural field-based reformulation in which the full spatiotemporal state is represented as a continuous function parameterized by a neural network. This reparameterization removes the time-sequential dependency of classical 4DVAR, enabling parallel-in-time optimization in parameter space. Physical constraints are incorporated directly through a physics-informed loss, simplifying implementation and reducing computational cost. We evaluate the method on the two-dimensional incompressible Navier--Stokes equations with Kolmogorov forcing. Compared to a baseline 4DVAR implementation, the neural reparameterized variant produces more stable initial condition estimates without spurious oscillations. Notably, unlike most machine learning-based approaches, our framework does not require access to ground-truth states or reanalysis data, broadening its applicability to settings with limited reference information.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

Submission Number: 1781