Poster Session P1.42 On the performance, impact, and liabilities of automated precipitation gage screening algorithms

Monday, 1 August 2005
Regency Ballroom (Omni Shoreham Hotel Washington D.C.)
Edward Tollerud, NOAA Research-FSL, Boulder, CO; and R. S. Collander, Y. Lin, and A. Loughe

Handout (859.4 kB)

Automated quality control (QC) procedures applied to real time data streams have the distinct advantages of reducing tedious and repetitious work, generalizing procedures across regions and personnel, and reducing many types of human error (transcription, for instance). When the data itself is heavily automated and subject to inaccuracy, however, and the observations are difficult and wildly variable (as are precipitation measurements), automation of QC presents difficult problems. The bottom line is that however thorough and calibrated the automated procedures are, counter examples clear to a knowledgeable eye are almost certain to arise. A good example are the hourly gage precipitation observations from the Hydrometeorological Automated Data System (HADS). To facilitate timely use as initialization data in numerical models, we have developed a system of algorithms written in Perl script specifically designed for the HADS and for the operational stations that are part of ASOS. These data and the results of their QC provide a good opportunity to assess the impacts of such screening. We describe factors that affect the performance of the QC procedures using several months of CONUS rainfall data and diagnostic output designed into the system itself. Among these factors are the handling of missing hours; determining manageable constraints for neighbor checks; setting time windows for off-hour daily reports (used as neighbors in proximity checks); and estimating a proper level of influence from a station's past performance. To compare these impacts quantitatively, we use a performance algorithm based on common verification scoring applied between stations and sets of their neighbors (more usually these scores are applied between verification data and model results). From a user's perspective, interest in QC is primarily dependent on its impact on a particular application. We thus present diagnostic descriptions of the impacts on two fundamental gage applications: analysis to grids (eg.,model initial fields) and real time verification of model predictions.
- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner