Continuous Data Assimilation with Stochastically Noisy Data

We analyze the performance of a data-assimilation algorithm based on a linear feedback control when used with observational data that contains measurement errors. Our model problem consists of dynamics governed by the two-dimension incompressible Navier-Stokes equations, observational measurements given by finite volume elements or nodal points of the velocity field and measurement errors which are represented by stochastic noise. Under these assumptions, the data-assimilation algorithm consists of a system of stochastically forced Navier-Stokes equations. The main result of this paper provides explicit conditions on the observation density (resolution) which guarantee explicit asymptotic bounds, as the time tends to infinity, on the error between the approximate solution and the actual solutions which is corresponding to these measurements, in terms of the variance of the noise in the measurements. Specifically, such bounds are given for the the limit supremum, as the time tends to infinity, of the expected value of the $L^2$-norm and of the $H^1$ Sobolev norm of the difference between the approximating solution and the actual solution. Moreover, results on the average time error in mean are stated.

💡 Research Summary

The paper investigates the performance of a continuous data‑assimilation algorithm when the observational data are corrupted by stochastic measurement errors. The authors focus on the two‑dimensional incompressible Navier–Stokes equations as a model problem, and they consider two types of linear interpolant observables: (i) volume‑average type operators satisfying an (H^{1}) approximation property (inequality (6)), and (ii) higher‑order interpolants satisfying a combined (H^{1})–(H^{2}) approximation property (inequality (7)). In the noise‑free setting, the algorithm consists of adding a linear feedback (nudging) term (-\mu(R_{h}(u)-R_{h}(U))) to the Navier–Stokes dynamics, where (U(t)) is the true solution and (u(t)) the approximating solution.

To model measurement errors, the authors assume that the error vector (E(t)) is a collection of independent Gaussian white noises, which can be represented by a finite‑dimensional Wiener process (W(t)). Consequently, the data‑assimilation system becomes a stochastic partial differential equation (SPDE): \