Ensemble MOS forecasts from multiple models

- Indicates paper has been withdrawn from meeting
- Indicates an Award Winner
Thursday, 21 January 2010: 4:30 PM
B305 (GWCC)
John Wagner, NOAA/NWS, Silver Spring, MD; and B. Glahn

Presentation PDF (317.9 kB)

The Meteorological Development Laboratory has developed a technique for creating probability density functions (PDF) and cumulative distribution functions (CDF) from an ensemble of forecasts from members of a single model or from multiple models. The statistical regression model is used in the MOS concept to create error estimates for a particular weather variable and forecast projection associated with the individual ensemble members. Those errors and the single value forecasts from the members can be used to create an ensemble of distribution functions. These individual functions can be combined with kernel density fitting to arrive at the PDF and CDF for that weather variable.

This technique, called Ensemble Kernel Density Model Output Statistics (EKDMOS), has been applied to the American and Canadian components of the North American Ensemble Forecast System (NAEFS). A sample was obtained that consisted of the control member and 20 perturbations from each component. This paper will show the similarities and differences of these two ensemble components and provide a comparison of their reliability and accuracy. In addition, both components of NAEFS will be combined, and the gain in using both instead of one independently will be shown.