The major goal for this session was to discuss the role of observations for AOMIP, and the need of taking optimal advantage of them through rigorous estimation (data assimilation) methods. It was recognized that depending on the application, very different requirements are placed on the estimation/assimilation system which have to be recognized and respectively evaluated. Another problem was to identify the relevant data (both observational specifically organized for AOMIP model validation), and where and how to archive the data for better distribution among AOMIP collaborators and throughout the Arctic observational and modeling communities. The major conditions and recommendations are presented below.
The session discussed the role of observations for AOMIP, and the need of taking optimal advantage of them through rigorous estimation (data assimilation) methods.
It was recognized that depending on the application, very different requirements are placed on the estimation/assimilation system:
(1) extrapolation: data assimilation in the often-used context of forecasting;
(2) interpolation: state estimation, directed at understanding the system through provision of a model trajectory which is consistent with model equations as well as available observations within prior error bars, and from which closed budgets can be calculated.
Our session was mainly concerned with (2), in recognition that AOMIP's primary goal is understanding the system. Moreover, the development of the adjoint or Lagrange Multiplier Method by several groups (NAOSIM, ECCO) was recognized as a novel tool within AOMIP. Discussion focused on ways in which this tool could be best deployed in support of AOMIP.
(A) Use of the adjoint method for sensitivity calculations:
(B) Model-data synthesis or state estimation:
Estimation system should use ALL available observations to the extent practicable (the purpose being that the coupled ocean/sea-ice model serves as dynamical interpolator).
To this end, the observation community ought to be made aware of the importance of distributing existing data sets. Collection in unified data format and at centralized data archives would be highly desirable (but will probably not be achievable). Categories of observations may be (and may provide a unifying structure for data servers): (a) remotely-sensed ocean, (b) in-situ ocean, (c)remotely-sensed sea-ice, (d) in-situ sea-ice, (e) atmospheric state and (f) air-sea fluxes in the Arctic or/and the Southern Ocean.
(D) Uncertainty/error estimates:
A key ingredient for formulating a least-squares model vs. data estimation problem, besides models and observations, are error or uncertainty estimates which should be attributed to each (!) observational element. These include mainly instrument errors, representation errors, and correlations among uncertainties in individual data streams. Without such quantitative estimates no useful estimation system can be put in place.
(E) Distribution & archiving of observations and AOMIP model results
Various data servers already exist (e.g. NSIDC for sea ice, Damocles, etc), and questions are, how to best harness existing servers, facilitate data gathering for modelers, harmonize data formats, and encourage (or enforce?) provision of error estimates and their correlations for each data set.
(1) AOMIP is an ARCSS funded project within the NSF Office of Polar Programs, and archiving of metadata for the AOMIP project results would be through the NCAR ARCSS Data Archive, see: http://www.eol.ucar.edu/projects/arcss/
(2) The archiving of AOMIP model files – for setting up the model runs and also the output when desired – should be centralized, in order to facilitate data exchange during the experiments and for data stewardship when the project is completed. Metadata records in other archives should point to the data at the centralized archive.
(3) NCAR/EOL will work with AOMIP to investigate and pursue the higher level of support for its data management needs. see: http://www.eol.ucar.edu/