4-D variational data assimilation and parameter estimation using the FSU global spectral model and its full-physics adjoint

Four-Dimensional variational (4-D VAR) data assimilation has become a very active research area during the past two decades. In this dissertation, the full-physics adjoint model of the FSU Global Spectral Model (FSU GSM) was completely derived by incorporating the adjoint of radiation and boundary layer parameterization packages into the data assimilation system. The radiation and boundary layer parameterization packages and the derivation of their corresponding tangent linear and adjoint counterparts in the FSU GSM are presented as part of this dissertation.; The full-physics adjoint of the FSU GSM of version T42L12 was applied to carry out 4-D VAR data assimilation and adjoint parameter estimation using initialized ECMWF analysis data sets. We first presented the formalism of 4-D VAR data assimilation and the methodology of adjoint parameter estimation, and closely examined the feasibility of performing 4-D VAR data assimilation using the FSU GSM and its full-physics adjoint model. Three key parameters (the bi-harmonic horizontal diffusion coefficient, the ratio of the transfer coefficient of moisture to the transfer coefficient of sensible heat, and the Asselin filter coefficient) along with the initial conditions were optimally recovered from the observations using an adjoint optimal parameter estimation approach. Then, we assessed the impacts of optimal initial conditions and key parameters estimation on the performance of the FSU GSM. The results show that the performance of the corresponding physical parameterization schemes was improved via tuning the physical parameter values. In the ensuing forecasts, although the impact of optimal initial conditions dominated that of the optimal parameter values at the early stages of the forecast, the model tended first to "forget" the impact of optimal initial conditions while the impact of the optimally identified parameter values persisted beyond 72 hours. The numerical weather prediction model yielded the best results when using both optimal initial conditions and optimal parameter values.; A preliminary experiment was also performed to calculate the sensitivity of the model 1-day forecast error to the initial conditions. The results were applied to identify regions with large analysis uncertainties.