Friday, 30 September 2011: 11:45 AM
Monongahela Room (William Penn Hotel)
Data assimilation techniques try to optimally blend information from multiple sources of information. As a result, best success will be obtained if all of the following conditions are met:
- Data are unbiased;
- Observation operators that allow the model to simulate observations are accurate;
- Errors and their spatio-temporal correlation structure are well described.
Yet, I am not aware of a single data assimilation experiment where all three conditions were met.
In an attempt to improve radar data assimilation, we have embarked in a multi-pronged effort:
- Data cleaning for assimilation use: Using dual-polarization spectral information, we have improved the way we isolate ground targets. In parallel, we have a separate stream for data cleaning for data assimilation purposes (no tolerance for bias, even if it means data gaps) in addition to the more traditional one for forecaster use (limited tolerance to data gaps, more willing to accept imperfect data);
- Better observation operators: Properly simulating reflectivity and Doppler velocity observations from model data in an efficient manner is not as easy as it sounds. We have developed full observation operators for radial velocity and found them unusable, which led to the design of the best possible simplifications for real-time uses;
- A first attempt at establishing and coding the error correlation structure: Even with a perfect observation operator, the correlation structure of radar errors is far from simple and has considerable non-trivial radial and azimuthal structures. Examples of these will first be presented that will set constraints on how the correlation structure of radar errors should be expressed.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner