An under-appreciated aspect of ocean data assimilation is the assembly of the input data stream with appropriate quality control and with estimates of the observational errors. These issues have come to the forefront as preparations are made for GODAE, yet there are no plans in place for QC of the historical data stream for climate applications. ODASI consortium members currently use 3 different databases with 3 different QC implementations. Under this collaboration we will use a consensus QC’d in situ database that borrows from all 3 implementations. In addition, we will participate in GODAE-sponsored QC activities, especially testing real-time algorithms developed at FNMOC that include reliability statistics.
In most ocean data assimilation implementations today, observation errors are assumed to result primarily from instrument error and are specified to be white in space and time. The level of variance is tuned with the forecast error variance to give the best results. In fact, the level of variance and the covariances should depend on the context for the product, i.e., the state of the ocean consistent with the particular model or application. The consortium is focused on seasonal-to-interannual time scales, so the data representation error covariances are expected to be different from those for a high resolution mesoscale application. Investigation of data representation errors requires analysis of historical data in a modeling context. If there are any examples of these analyses, they are very few. We will start to explore this issue with the TAO and TOPEX observations and with experiments to explore the sensitivity of assimilation products to various assumptions made for the observation errors.
NCEP and GFDL will lead efforts to QC the in situ data streams and test QC algorithms proposed through GODAE. GMAO and NCEP will collaborate on data representation errors for GCMs, on assessing the feasibility of estimating representation errors and exploring the impact on the assimilation outcome. For baseline and benchmark estimates, LDEO will provide analyses of signal and error in historical climate data using a quasi-linear ocean model, intermediate coupled models, and statistical models.