Background Fit to Satellite Observations

Millions of observations that are already monitored and assimilated are also available for verification and diagnostics of 6-hour forecasts. These observations span the globe and cover the depth of the atmosphere far better than the much more limited set of radiosondes currently used for verification in observation space. We have developed a tool to compute a quantitative measure of the fit of the 6-hour forecast (background) to these global observations, along with uncertainty estimates. This method has several considerable advantages over traditional forecast statistics and scorecards. It requires an NWP simulation of only one to two weeks to accurately evaluate whether the experiment has improved or degraded analyses and forecasts (Alan Geer, ECMWF, personal communication), a much shorter time frame than the three to six months of simulation needed with traditional forecast statistics. The savings in software development time for NWP and DA developers, and in actual compute time for expensive NWP models, is considerable We also produce a scorecard with a decision matrix to determine whether an experiment is a win, loss, or neutral with respect to a control run.

The tool has diagnostic applications as well. For example, we might find that an experiment improves the background fit to radiances sensitive to moisture in the boundary layer, while degrading the background fit to radiances sensitive to stratospheric temperature. This very specific information can inform the experimenter’s understanding of what went right and wrong, and provide guidance on how to resolve issues. The software automatically groups the plots and scorecard by atmospheric variable type and vertical location, so it does not require expertise in radiance assimilation to interpret and use the results.