Postprocessing Weather Prediction Model Output in the U.S. National Weather Service: Model Output Statistics from 1972 to 2012 (Invited Presentation)

While development of the Model Output Statistics (MOS) approach was initiated in the late 1960’s, the 1972 paper by Bob Glahn and Dale Lowry defined MOS and demonstrated that post-processing the output of dynamical weather prediction models would improve public and aviation weather forecasts. Developmental principles were subsequently refined and expanded in the National Weather Service’s (NWS’s) Techniques Development Laboratory (TDL).

By the mid-1970’s, MOS guidance products were available twice daily. These MOS products were based on the output of the NWS’s primitive equation (PE) and atmospheric trajectory (AT) models. Observations required in developing the PEATMOS equations were obtained from archived synoptic reports. Guidance was available for the probability of precipitation (PoP), maximum/minimum (max/min) temperature, surface wind, clouds, conditional probability of frozen precipitation, ceiling height, and visibility. Observations from radar summary charts were used to develop a MOS thunderstorm probability product. Recognizing the need for a more extensive archive of surface weather observations, TDL began in December 1976 to archive hourly observations available at the National Meteorological Center’s (NMC’s) central computer facility.

Because PEATMOS products were unavailable until shortly before deadlines for issuing the official weather forecasts, forecasters could not always use the objective guidance. In fact, NMC had implemented the Limited-area Fine Mesh (LFM) model in 1971 in recognition of timeliness concerns. After the LFM model was enhanced to make forecasts out to 48 hours, the LFM provided an “early” look at the weather for the next 48 hours, and became the model of choice in much of the United States east of the Rockies. TDL determined that PEATMOS equations could be applied to LFM model output with minimal deterioration. By 1976, “early” (LFM-based) and “final” (PE-based) MOS guidance packages were being issued, and TDL had shifted most new MOS development to the LFM model. In April 1979, PEATMOS guidance was eliminated except for max/min temperature and PoP. In August 1980, the PE model was replaced by the Global Spectral Model (GSM), and by March 1981, all PEATMOS guidance had been eliminated.

During the early 1980’s, LFM MOS guidance was enhanced in numerous ways. Manually digitized radar data or lightning strike reports were used to define the occurrence of thunderstorms. The max/min temperature guidance was modified to predict the daytime max and nighttime min, rather than calendar day values. Guidance became available for numerous military bases and for stations in Alaska. The LFM MOS guidance continued in operations until February 1996 when the LFM model was eliminated.

In March 1985, NMC implemented the Regional Analysis and Forecast System (RAFS) composed of the Regional Data Assimilation System and Nested Grid Model (NGM). TDL was asked to develop NGM-based guidance, but without a stable sample of NGM forecasts, a traditional MOS development was not feasible. Instead, a “modified perfect prog” system was created, and NGM-based guidance became operational in May 1987 for max/min temperature, PoP, surface wind, and clouds. By October 1987, most planned changes to the RAFS had been implemented, and accumulation of a suitable developmental sample of NGM forecasts became possible. Between August and December 1988, NMC and TDL developers ran the RAFS retrospectively for the October 1986-September 1987 period -- the first time the NWS had rerun a model system on such an extensive sample. In July 1989, the inaugural NGM MOS guidance package was implemented. Development of NGM MOS guidance for the contiguous United States and Alaska continued until 1995. During these same years, an extensive package of GSM-based guidance also became available for the 3- to 5-day projections. The NGM and NGM MOS products continued in operations until March 2009.

In 1994, TDL began work on MOS-2000, a new MOS software system designed to be modular and adjustable to the available computer platform. Moreover, the data used in both development and operations were to be packed in an efficient and platform-independent manner. In 1995, a major effort also began within the National Centers for Environmental Prediction (NCEP – a reorganization that included NMC) to convert operational software and operating system instructions to a UNIX environment. Observational and model data were to be stored in WMO-standard formats. By June 1997, TDL had restructured its archives and had rewritten the NGM MOS and GSM MOS operational programs to conform to new NCEP standards.

In May 2000, guidance from the MOS-2000 system was issued for the first time. MOS guidance packages were developed from the Aviation (AVN) and Medium-Range Forecast (MRF) runs of the GSM. Guidance became available for stations in Hawaii and in the Caribbean. During the next several years, the flexibility of the MOS-2000 system enabled significant upgrades to MOS guidance, including use of new observational datasets; guidance for mesonet, cooperative observer, and marine observing sites and for a complete set of weather elements; implementation of AVN guidance for the 0600 and 1800 UTC cycles; and development of a medium-range guidance package for the 1200 UTC cycle. In 2002, when separate AVN and MRF model runs were eliminated, the Global Forecast System (GFS) became the standard at the 0000, 0600, 1200, and 1800 UTC start times. Subsequently, more enhancements were made to the GFS-based MOS system, including the dissemination of virtually all MOS guidance in a station-oriented digital format. In August 2006, MOS guidance became available on a 5-km grid in support of the NWS National Digital Forecast Database (NDFD). By 2012, the Meteorological Development Laboratory (MDL- a reorganization of TDL in 2000) had implemented test versions of MOS guidance on a 2.5 km NDFD grid.

For four decades, changes in dynamical weather prediction models, computer resources, weather observations, and techniques affected the MOS guidance and enabled improvements that would not have been possible or even conceivable in 1972. This talk highlights some of the most significant events in the life of the MOS post-processing project.