Use of ESMF in the GEOS-5 Modeling and Assimilation System

The GEOS-5 system was developed during the last five years, alongside the development of ESMF. It was, therefore, designed and built as an ESMF-based system. It consists of some thirty ESMF gridded components connected in a hierarchical topology, using fine-grained ESMF componetization. In the atmospheric GCM, in particular, ESMF components are used down to the level of individual parameterizations. To maintain a uniform structure over some many ESMF components, we needed to develop standard methods and practices for using ESMF. These standards were codified in an ESMF-based software layer called MAPL. MAPL-ESMF gridded components are easily coupled and manipulated within the GEOS-5 structure, but appear as ordinary, fully-compliant ESMF gridded components if used in non-MAPL applications. The use of ESMF's component design has allowed us considerable flexibility in extending the GEOS-5 system. The current atmospheric model, for example, runs seamlessly with two dynamical cores: either the traditional latitude-longitude core or one recently developed at GFDL using a cubed-sphere grid. The development of this capability was greatly simplified by the MAPL-ESMF design. The coupled atmospheric-ocean model uses the MOM4 OGCM, also developed at GFDL, which has been wrapped as a MAPL-ESMF that allows us to keep up with the latest distributions coming from GFDL. Several atmospheric chemistry and aerosol components are also available, most notably the GMI chemistry and the GOCART aerosol model, which also run in the system with MAPL-ESMF wrappings. Finally, MAPL-ESMF versions of the radiation and fast physics used an NCEP and of the MIT ocean GCM have been developed and are now being tested as optional components in the system. In addition to the model components, GEOS-5 also includes analysis components for assimilation atmospheric, oceanic and land surface data. These components are also integrated in the system using the MAPL-ESMF methodology, which is now supporting both 3d_var and Ensemble Kalman filter approaches. A 4d-VAR atmospheric assimilation with forward and adjoint components is also in advanced stages of testing.