3.5 Forecast Sensitivity to Data Assimilation System Observation and Background Error Covariance Parameters

Tuesday, 25 January 2011: 12:00 PM
2B (Washington State Convention Center)
Dacian N. Daescu, Portland State University, Portland, OR; and R. H. Langland and R. Todling
Manuscript (155.0 kB)

The development of the adjoint of the forecast model and of the adjoint of the data assimilation system (adjoint-DAS) makes feasible the evaluation of the local sensitivity of a model forecast aspect with respect to a large number of DAS input components. To date, the adjoint-DAS approach has been mainly considered in numerical weather prediction (NWP) as an effective tool to assess the value of observations in reducing the forecast errors and the forecast impact as a result of changes in the observing system (Gelaro et al. 2010, Cardinali 2009, Baker and Langland 2009). This work brings forward additional adjoint-DAS capabilities and applications based on the forecast sensitivity to the specification of error covariance parameters in the DAS. New theoretical aspects of error covariance sensitivity and forecast impact assessment are presented together with illustrations of the adjoint-DAS ability to provide sensitivity information for DAS diagnostics and guidance to error-covariance parameter tuning procedures. Emphasis is placed on the intrinsic properties of the analyses derived from a minimization principle that are used to closely relate the error covariance sensitivity to the observation and background sensitivity. It is explained that the adjoint-DAS software tools developed at NWP centers for observation impact studies make feasible the evaluation of the forecast sensitivity to a large number of error covariance parameters. The adjoint-DAS observation impact assessment relies on information extracted from the innovations and observation sensitivity. In conjunction with the observation sensitivity, the observed-minus-analysis (o-m-a) information provides an assessment of the weighting between the information content of various DAS input components and allows the identification of the input components where improved estimates of the error statistics have a potentially large impact on the forecast error reduction. Preliminary results of forecast sensitivity to observation and background error covariance weight parameters are presented using the fifth-generation NASA Goddard Earth Observing System (GEOS-5) atmospheric DAS and its adjoint developed at the Global Modeling and Assimilation Office.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner