Title:Observation Impact on the Short Range Forecast Under Simulated Analysis/Forecast System Improvement

Author: Brett Hoover (CIMSS, UW-Madison)
Rolf Langland (NRL-MRY)

The adjoints of a data assimilation (DA) system and numerical weather prediction (NWP) model can be used to derive an estimate of the impact of an assimilated (individual) observation on some aspect of a short-range forecast - typically defined as the energy- based error-norm of the 24-hr forecast. Such an estimate is used routinely in the NAVGEM (Naval Research Laboratory), GEOS-5 (NASA GMAO) and GDPS (ECMWF) for monitoring the impact of any observation or group of observations on the forecast. Often intercomparisons of observation impact are made between different groups of observations within a DA system, or a particular observation-type between various centers; it is found, for example, that the NAVGEM derives higher observation impact from satellite derived wind observations than other centers. These intercomparisons raise several questions. What does it really mean that an observation, or a type of observations, has a high impact? Is the NAVGEM deriving benefit from these satellite winds that is lost to other centers? Ultimately, what is the relationship between observation impact and the quality of the analysis/forecast system?

We investigate some of these questions in the NAVGEM, which utilizes a 4-dimensional variational data assimilation system, the NAVDAS-AR. The system undergoes "simulated" improvement through the assimilation of analysis data from the ECMWF analysis, treating profiles from ECMWF analyses as radiosondes on a global 1°x1° grid. When assimilating these profiles, the NAVGEM analysis displays characteristics more similar to the ECMWF analysis, and the forecast system is consistently improved out to 120 hours, based on several standard forecast metrics. The adjoint-derived observation impact estimate is used analytically to observe the impact of this data on the assimilation of the rest of the observing system and the relationship between the improved analysis, the improved forecast - especially the 6-h forecast defining the model background, and the observation impact. It is shown that substantial improvement of an analysis/forecast system is expected to yield a decrease in total observation impact, making the relationship between observation impact and forecast improvement less intuitive than has often been previously implied.

GMAO Head: Steven Pawson Global Modeling and Assimilation Office NASA Goddard Space Flight Center

Curator: Nikki Privé Last Updated: Feb 9 2015