Ensemble Forecast Systems and MOS (Invited Presentation)

In the early 1990s, Model Output Statistics (MOS) was well accepted throughout the weather enterprise of the United States (US). That is when the National Centers for Environmental Prediction (NCEP; then named the National Meteorological Center [NMC]) implemented their first Ensemble Forecast System (EFS). Even then, MOS forecasts had been part of the operational mainstream for almost 30 years. As MOS and EFS both evolved, there were many calls for an “ensemble MOS” product. The earliest and simplest technique applied MOS equations that were developed with a single deterministic model to the various members of the EFS.

Developing a reliable and accurate ensemble MOS product, however, proved to be theoretically challenging, computationally intensive, and data intensive. In 2006, Bob Glahn formed a team within MDL to develop the product we would eventually name Ensemble Kernel Density MOS (EKDMOS). The team quickly learned that they faced twin challenges--the creation of accurate and reliable probabilistic forecasts and the development of appropriate forecast verification techniques.

This talk will review the workings of EKDMOS, its predecessor, enhancements that have been developed over the years, and the evolution of the EKDMOS products themselves. This talk will also show how the EKDMOS techniques are contributing to the NWS's Model Blender project.