The use of Targeted Observations in Operational numerical weather forecasting

Serious Numerical Weather Prediction (NWP) forecast failures are localized both in time and space. On the meso- and larger scales, these failures are largely associated with errors in the initial conditions (analysis) that amplify rapidly, rather than with model deficiencies. During the recent years, techniques have been developed to identify regions where the assimilation of extra observations taken at a future time can substantionally improve the ensuing forecast in a preselected verification region where the estimated forecast uncertainty is large. In this paper we give an overview of the use of one of these targeting techniques, the ensemble transform (ET, Bishop and Toth, 1999) technique, in the course of the FASTEX, NORPEX, CALJET, and the Winter Storm Reconnaissance 1999 and 2000 field programs.

The results are based on a total of 54 dropsonde flight missions and 12 data denial experiments, and are all statistically significant. The evaluation of the technique includes the use of "null" experiments where the impact of the dropsonde (or regular radiosonde) data from weakly sensitive or insensitive areas is tested and compared to that from highly sensitive areas. The overall results indicate that the ET technique is able to distinguish among highly sensitive, moderately sensitive, and insensitive areas. In the majority of the cases the adaptively collected data had a significant and positive impact on the quality of the targeted forecasts in the predefined verification region at the verification time. Based on the positive targeting experience accumulated at NCEP, the targeting technology is being transferred into the US National Weather Service operations.