Optimizing Monthly Grid-based CO2 Fluxes with 4D-Var Data Assimilation Technique
Abstract
We estimate a global carbon flux distribution on a transport model grid using 4D-Var data assimilation technique. An effective use of the satellite data is expected to reduce a flux estimation error. The GOSAT satellite gives us column CO2 measurements of approximate several thousand observations over land per month. Here, we show a method solving the large linear equations with high spatial resolution and long time window using the GOSAT data. In case of grid-based fluxes optimization, the derivative of a cost function for the optimization problem requires an adjoint of a transport model operator. A calculation of the adjoint code for four dimensions needs large computational time and memory. Furthermore, the high-resolution and long-term data assimilation using a satellite data is thought to require a huge coefficient matrix, inverse matrix and Hessian, or their approximations. Therefore, to avoid the huge computational cost and matrix by matrix operations, we use the optimization algorithm that is combination of the scaled memoryless BFGS method and the preconditioning technique in the framework of the conjugate gradient method, well known as a Krylov subspace method for solving large systems of linear equations. We developed a fast algorithm and find better conjugate parameters to optimize the monthly CO2 flux through numerical experiments.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2010
- Bibcode:
- 2010AGUFM.B31F0370S
- Keywords:
-
- 0400 BIOGEOSCIENCES