On the ability to develop MOS guidance with short dependent samples from an evolving numerical model

A properly-constructed statistical post-processing system can effectively remove systematic biases in numerical weather prediction (NWP) model output. It also provides forecasters with a valuable first guess regarding the sensible weather to be expected given a specific set of model-forecast conditions. Ideally, such robust and skillful statistical guidance would require a long, stable sample of dependent data. However, numerical models do change in time as new technology is incorporated and computational ability is expanded. Despite such changes, statistical post-processing techniques can still be applied effectively to offer forecasters reliable guidance based upon the NWP models.

This paper discusses the development strategies tested as the Meteorological Development Laboratory (MDL) sought to update its operational Model Output Statistics (MOS) suite after the North American Mesoscale model (NAM), run by the National Centers for Environmental Prediction (NCEP), underwent changes in June 2006. Specifically, the NCEP eta-coordinate (Eta) model was replaced by the newer Nonhydrostatic Mesoscale Model (NMM). While equations derived from the earlier (Eta) version could be applied to some MOS elements without any noticeable change, other elements did show a significant enough decrease in forecast skill. The impact varied by weather element, region of the country, season, and equation type (single-station vs. regionalized operator). As a result, an updated NAM MOS package was needed.

We tested several developmental configurations in an effort to obtain a stable and robust NAM MOS system. Results were encouraging, with longer samples generally yielding better NAM MOS performance. This seemed to hold true even when lengthening the dependent sample meant including output from the final configuration of the Eta model in addition to output from the NMM. With these positive results, MDL was able to proceed with developing and operationally implementing NAM MOS based upon NMM data.