A New Statistical Tool to Assess the Effectiveness of Observing System Configurations

Among a variety of tools available to assess the impact of changes in observing systems on the quality of analyses and forecasts, Observing System Experiments (OSEs) provide the most reliable results. However, their large computational cost seriously limits the range of observing system configurations that can be assessed with OSEs. To eliminate the need for the computationally intensive execution of many individual OSE experiments, we propose a novel statistical approach to analyze output from an operational forecast system over a period of time (e.g., NCEP’s GFS/GSI over a month or season) to assess the influence of different types of observations on the quality of analysis fields (FAst Statistical Tool for OSE - FAST-OSE).

     FAST-OSE is based on the estimated time mean gridpoint information content (IC) derived by a data assimilation system from all available observations (baseline configuration). IC is determined through an analysis of the behavior of data assimilation – forecast cycles. After a suitable assessment of the presence or proximity of each observing system in/to each gridpoint in the 3D analysis grid (Observing System Indicator fields - OSI), the OSI of each observing system is statistically related to the overall IC of all observations. The resulting measures (e.g., correlation coefficients between overall IC and individual OSIs) then are used to predict how analysis error variance will change due to any variations in the baseline configuration of the different observing systems.

     The new method is tested and evaluated in perfect model data assimilation – forecast experiments with a quasi-geostrophic model and simulated radiosonde, aircraft, and radar wind observations. Traditional OSEs will also be carried out and their results carefully compared to those from FAST-OSE. Recommendations for the use of FAST-OSE in the assessment of current and optimization of future observing systems will also be discussed.