The authors report on the implementation and evaluation of a 48-member ensemble adjustment Kalman filter (EAKF) for the ocean component of the Community Climate System Model, version 4 (CCSM4). The ocean assimilation system described was developed to support the eventual generation of historical oceanstate estimates and ocean-initialized climate predictions with the CCSM4 and its next generation, the Community Earth System Model (CESM). In this initial configuration of the system, daily subsurface temperature and salinity data from the 2009 World Ocean Database are assimilated into the ocean model from 1 January 1998 to 31 December 2005. Each ensemble member of the ocean is forced by a member of an independently generated CCSM4 atmospheric EAKF analysis, making this a loosely coupled framework. Over most of the globe, the time-mean temperature and salinity fields are improved relative to an identically forced ocean model simulation without assimilation. This improvement is especially notable in strong frontal regions such as the western and eastern boundary currents. The assimilation system is most effective in the upper 1000m of the ocean, where the vast majority of in situ observations are located. Because of the shortness of this experiment, ocean variability is not discussed. Challenges that arise from using an ocean model with strong regional biases, coarse resolution, and low internal variability to assimilate real observations are discussed, and areas of ongoing improvement for the assimilation system are outlined.
(ProQuest: ... denotes formulae omitted.)
In this paper, we document the development of an ensemble adjustment Kalman filter (EAKF) (Anderson et al. 2009) data assimilation system for the ocean component of the Community Climate System Model, version 4 (CCSM4). The ocean assimilation system described here was developed to support the eventual generation of historical ocean-state estimates and ocean-initialized climate predictions with the CCSM4 and its next generation, the Community Earth System Model (CESM).1 This development is part of a broader initiative at the National Center for Atmospheric Research (NCAR) to build assimilation capabilities for the atmosphere, land, sea ice, and ocean components of the community model.
There is currently an array of global ocean assimilation products available to the climate-science community that employ various ocean general circulation models and assimilation algorithms. The assimilation methods used to construct these products are all least squares methods that attempt to minimize the difference between an ocean model solution and a set of observations. Broadly speaking, they are distinguishable from one another by the constraints levied upon the solution, the way that prior knowledge of the state of the system is formulated, and whether observations can influence state estimates in the past. For example, adjoint [fourdimensional variational data assimilation (4DVAR)] methods, for example, the Estimating the Circulation and Climate of the Ocean (ECCO)2 products described by Wunsch and Heimbach (2007) and Keuroohl et al. (2006), strongly enforce the dynamical constraints of the model, allowing observations from the future to exert influence on the past ocean state. In contrast, methods such as 3DVAR, optimal interpolation, and Kalman filters only use past and current observations to estimate the ocean state. Within this latter category, the way in which background information is specified is one important distinguishing characteristic. The 3DVAR and optimal interpolation methods typically specify parameterized background covariance estimates, for example, the National Centers for Environmental Prediction (NCEP) Ocean Reanalysis system, version 4 (ORAS4) (Mogensen et al. 2012) and Global Ocean Data Assimilation System (GODAS) (Behringer and Xue 2004). Some implementations of these can include parametric covariance forms that are calendar month and/or model state dependent, for example, the Simple Ocean Data Assimilation (SODA) (Carton and Giese 2008). Notable exceptions to parametric methods are a newer class of methods that use a fixed set of ensembles to compute sample background statistics, for example, the Australian Bureau of Meteorology Bluelink Ocean Data Assimilation System (BODAS) (Oke et al. 2008) and Global Ocean Reanalysis and Simulations (GLORYS2V1)3 (Ferry et al. 2012). Instead of specified covariance forms, Kalman filters use time-varying background covariance estimates that are determined via model dynamics. This naturally enables the multivariate physical relationships to be captured by the evolving background covariance. EnsembleKalman filters, such as the one we use in the application presented here, share this general property of the Kalman filter. In the context of global ocean circulation models, variations on the ensemble Kalman filter are currently being employed with the Max Planck Institute Ocean Model (Leeuwenburgh 2007), the Poseidon ocean general circulation model (Keppenne and Rienecker 2001, 2002), and the Modular Ocean Model (MOM) (Zhang et al. 2007).