Robustness and Behavior of Adjoint Calculations of Observation Impacts in Numerical Weather Prediction

Adjoint models are powerful tools that can be used to estimate the impact of observations on a chosen norm for numerical weather prediction forecasts. In this study, the Global Modeling and Assimilation Office (NASA/GMAO) Observing System Simulation Experiment framework is employed to investigate the behavior of the adjoint tool in an environment where the 'true' state of the atmosphere is fully known. This allows for the calculation of adjoint estimates of observation impact for very short forecast times including the zero-hour analysis state. The adjoint calculations using self-analysis verification can also be compared to adjoint calculations using the 'truth' as verification in order to characterize the robustness of adjoint estimations in the operational setting. Results from a experiments exploring various aspects of performance of the adjoint tool will be presented.