|Title||Forecast sensitivity to observation (FSO) as a diagnostic tool|
|Year of Publication||2009|
|Secondary Title||Technical Memorandum|
This paper describes the use of forecast sensitivity to observations as a diagnostic tool to monitor the observation impact on the quality of the short range forecasts (typically 24 hour). The forecast error is provided by a control experiment (using all observations available) which has been run among two series of observing system experiments performed at ECMWF. The observation data impact obtained with the forecast sensitivity is then compared with the observation data impact as classically measured in the context of observing system experiments. Differences and similarities between the two approaches are highlighted. Overall, the assimilated observations decrease the forecast error. However, locally some poor performances are detected that are related either to the data quality, the sub-optimality of the data assimilation system or biases in the model. It is also found that some synoptic situation can deteriorate the quality of certain measurements or can induce some local weather variability over small areas that the assimilation system cannot correctly resolve. Finally, the performance of the current operational version (CY35R2) of the data assimilation system for the last four months of 2008 shows a consistent overall positive impact of the observations.