Summary

In this case from the 12 UTC 13 August 2001 22-km operational Eta model:

• The analysis had a bizarre "hole" where it disagreed with radiosonde reports
  
• The analysis was strongly influenced by a large number of aircraft reports in this area
  
• The aircraft data corroborated each other
  
• The aircraft data appeared to be good and sampled a mesoscale feature
  
• Although the nearby radiosonde was just outside this narrow mesoscale feature, the aircraft and raob reports appear mutually consistent
  
• The narrow mesoscale feature was only sampled in the vicinity of the airport, though it may have extended a few hundred kilometers in length
  
• The analysis was unable to distinguish between the cluster of aircraft data representing a narrow mesoscale feature and data representing the larger-scale environment
  
• The analysis assumes that corrections to the model first guess are similar over a broad circular region
  
• The analysis smeared the influence of the aircraft data across a broad region outside the actual location of the narrow feature those data represented
  
• The analysis produced a saturated layer with a dry-adiabatic lapse rate because it does not include any link between temperature and moisture fields (they are linked in the first guess but not in the 3D-VAR analysis)
  

This case and its discussion demonstrate the following points, which hold in many other cases as well:

• The analysis may get overwhelmed by a large volume of upper-air data other than radiosonde reports
  
• The observations—whether radiosondes or other sources—may be sampling mesoscale features not representative of the larger-scale environment
  
• The observations may sample only part of the mesoscale feature, inadequately defining its extent or horizontal and vertical structure
  
• Distinguishing between data representing features the model should and should not include in its analysis is a difficult challenge with no consistently good solutions
  
• As models are made with finer resolution, more of the observations are representative of features the model analyses should include but they may be improperly included due to inadequate and uneven data coverage
  
• The operational Eta analysis in 2001 is not designed to represent mesoscale features
  
• Development is underway for the analysis to allow the spread of information from observations in more realistic, situation-dependent patterns. This may allow for decent—not perfect—shapes and structures of features only partially sampled by observations and is necessary for correctly describing well-sampled mesoscale features in the analysis.
  
• Analysis shortcomings and limitations leave room for the attentive and knowledgable human forecaster to improve on the model forecast