Continuing dynamic assimilation of the inner region data in hydrodynamics modelling: Optimization approach

TL;DR

This paper introduces an optimization-based data assimilation method for hydrodynamics modeling that improves solution accuracy by jointly solving the model and data assimilation as a PDE-constrained optimization problem.

Contribution

It proposes a novel optimization approach that overcomes limitations of traditional nudging methods in dynamic data assimilation for regional hydrodynamics models.

Findings

Significantly improves numerical solution accuracy.

Effective even with boundary condition sensitivities.

Validated on simple model equations.

Abstract

In meteorological and oceanological studies the classical approach for finding the numerical solution of the regional model consists in formulating and solving the Cauchy-Dirichlet problem. The related boundary conditions are obtained by linear interpolation of data available on a coarse grid (global data), to the boundary of regional model. Errors, in boundary conditions, appearing owing to linear interpolation may lead to increasing errors in numerical solution during integration. The methods developed to reduce these errors deal with continuous dynamic assimilation of known global data available inside the regional domain. Essentially, this assimilation procedure performs a nudging of large-scale component of regional model solution to large-scale global data component by introducing the relaxation forcing terms into the regional model equations. As a result, the obtained solution is not a valid numerical solution of the original regional model. In this work we propose the optimization approach which is free from the above-mentioned shortcoming. The formulation of the joint problem of finding the regional model solution and data assimilation, as a PDE-constrained optimization problem, gives the possibility to obtain the exact numerical solution of the regional model. Three simple model examples (ODE Burgers equation, Rossby-Oboukhov equation, Korteweg-de Vries equation) were considered in this paper. The result of performed numerical experiments indicates that the optimization approach can significantly improve the precision of the sought numerical solution, even in the cases in which the solution of Cauchy-Dirichlet problem is very sensitive to the errors in the boundary condition.