I am quite new to ROMS-DA, and trying to follow WC13 test case as illustrated in ROMS-wiki. I well understand the advantage of 4dvar algorithm and fully appreciated with the efforts that has been done by 4dvar developers.
Maybe it is odd to ask this question but I am very curious about it : can i4dvar be run in 3dvar-FGAT mode? I mean how to make changes with the calls to tl_*operators and ad_*operators, to get the effect of something like Identity-Matrix-operation (which is the 3dvar-form of tangent and adjoint matrix appeared in 4dvar formula). Is this straightforward or I totally go astray?
Appreciated for any advice and discussion. Thanks.
Kelvin
We never coded 3D-Var in ROMS native variational data assimilation algorithms. But, if you check the literature, 4D-Var performs better than any 3D-Var algorithm. In 3D-Var, we assume that the linearized model (tangent linear kernel) is the identity matrix! It doesn't use the adjoint model. Anyway, in the future, any of the 3D-Var family of algorithms, including FGAT and hybrid 3D-Var/EnKF, will be possible in the ROMS-JEDI framework. We will announce when such a framework becomes public.
The I4VAR algorithm uses the primal formulation, which is deprecated. Instead, use the dual formulation RBL4D-Var algorithm. It allows additional algorithms that are also available in ROMS. I suggest you read our publications.
Hernan, Thanks for your kind reply.
I notice that you gave a nice talk about ROMS-JEDI last year @Ocean Predict DA-TT Meeting, and had shown some excellent work already done within this framework. Looking forwards to have a try with ROMS-JEDI, especially considering the more flexible DA options provided. Would you give a rough estimate of its release time, or is there an opportunity to get a beta-version prior to the public release?
Sincerely,
Kelvin
to ask this question but I am very curious about it : can i4dvar be run in 3dvar-FGAT mode? I mean how to make changes with the calls to tl_*operators and ad_*operators, to get the effect of something like Identity-Matrix-operation (which is the 3dvar-form of tangent and adjoint matrix appeared in 4dvar formula). Is this straightforward or I totally go astray?