The impact of different data types on the Eta analysis

The EDAS uses 3-hour chunks of data to make a 3D-VAR analysis based on a previous first-guess forecast. The EDAS cycle starts 12 hours before the forecast initial time. The "analysis" is the last of this sequence of analyses, the one valid at the forecast initial time.

The basic experimental design is:

Run the assimilation and forecast system with all data
Run the assimilation and forecast system with all data except radiosonde winds
Run the assimilation and forecast system with all data except radiosonde temperatures
Run the assimilation and forecast system with all data except GOES sounder data
Run the assimilation and forecast system with all data except GOES IR cloud-drift winds
Etc. — Separate run for each data source
Compare each analysis and forecast against the one that used all data

Here are the resulting RMS differences between including all data and denying a data type. RMS differences in analysis temperatures are shown at three levels over the entire domain and over only the CONUS for each of the 34 data types used in the model for the one case studied. The results for the three levels are plotted in stacked form. (For example, RAOB1 [leftmost bar] has RMS differences of around 0.6 K, 0.3 K, and 0.3 K at 850 hPa, 500 hPa, and 300 hPa over the CONUS.) The bigger the difference, the more sensitive the analysis was to that data type at that level.