HPC computation issues of the incremental 3D variational data assimilation scheme in OceanVar software

The most significant features of Data Assimilation (DA) are that both the models and the observations are very large and non-linear (of order at least O(10 8 )). Further, DA is an ill-posed inverse problem. Such properties make the numerical solution of DA very difficult so that, as stated in (19),"solving this problem in "real-time" it is not always pos- sible and many different approximations to the basic assimilation schemes are employed". Thus, the exploitation of advanced computing environments is mandatory, reducing the computational cost to a suitable turnaround time. This activity should be done according to a co-design methodology where software requirements drive hardware design decisions and hardware design constraints motivate changes in the software design to better fit within those constraints. In this paper, we address high performance computation issues of the three dimensional DA scheme underlying the oceanographic 3D-VAR assimilation scheme, named Ocean- VAR, developed at CMCC (Centro Euro Mediterraneo per i Cambiamenti Climatici), in Italy. The aim is to develop a parallel software architecture which is able to effectively take advantage of the available high performance computing resources.