Operational NWP Postprocessing—The Early Years (Invited Presentation)

Operational NWP Postprocessing—The Early Years

Bob Glahn

Meteorological Development Laboratory

Office of Science and Technology Integration

National Weather Service, NOAA

Silver Spring, Maryland

Operational postprocessing of NWP output and dissemination of postprocessed products started in the United States, and likely in the world, in the Techniques Development Laboratory (TDL) [now the Meteorological Development Laboratory (MDL)] of the Weather Bureau (now the National Weather Service) in 1968. Both the first operational perfect prog (PP) forecasts of maximum and minimum temperature and of model output statistics (MOS) forecasts of 10m winds were initiated in that year. Within a few years, TDL was producing MOS forecasts of most weather elements observed at the earth’s surface. Other organizations were also postprocessing NWP output for their individual needs, such as the National Hurricane Center.

The Joint Numerical Weather Prediction Unit (JNWPU) of the United States was formed and began operating on July 1, 1954, and by the summer of 1955 began issuing numerical predictions twice daily. However, according to Fred Shuman, it was discovered that the model used was unable to predict reliably and accurately, and it also provided little or no useful information to the forecaster. After solving some basic problems and changing from a 3-level model to a 1-layer barotropic model, by 1958 the skill began to improve, although it was still quite low compared to today’s standards. Also, the skill was in forecasting upper air variables, and what the public was interested in was weather at the surface of the earth. This naturally led to statistically relating surface weather to model forecasts of upper air variables.

By 1959, Bill Klein, who had been involved in statistical forecasting since the 1940’s, applied relationships he had derived between observed upper air observations and surface temperature to the upper air forecasts produced by the NWP model; this later came to be called the “perfect prog” method. Roger Allen and Joe Sassman, following some work by the U.S. Navy, related precipitation occurrence to “observed” vertical motion at initial time from the thermotropic NWP model and dew point depression. They applied these relationships to forecasts from the model.

During the late 1950’s and early 1960’s, considerable statistical forecasting work was being done in the United States Weather Bureau and at the Travelers Research Center, but not involving NWP—the so called “classical method.” This approach had seemed to run its course, and NWP was recognized as the way of the future.

In 1964, in the context of forecasting for aviation interests, Charlie Roberts directed that a short-range, subsynoptic model be developed, and I was intent on developing statistics from its output. At this time, the model in operations at NMC was being run on a grid spacing of 381 km, output was at 6-h intervals, there was no stable archive of NWP output, nor any way of producing and disseminating statistical forecasts.

Dale Lowry and I tailored two existing NWP models to run at ¼ the NMC scale and produced output of sea level pressure and moisture at hourly intervals out to 17 hours over the eastern United States. We called this combined model the subsynoptic advection model (SAM). The first MOS statistical product from it was sent to Weather Bureau field forecasters on or about June 10, 1968.

The NMC numerical models progressed from the simple barotropic to better models, and changes became less frequent. Archives of these models were established, and TDL began basing its postprocessing on NMC’s primitive equation (PE) model. By 1972, MOS forecasts of maximum temperature, PoP, wind direction and speed, cloud amount, and conditional probability of frozen precipitation, derived from both SAM and the PE model predictors, were produced, and several were transmitted to WB field offices.

Bill Klein, Frank Lewis, and others had been working on forecasting maximum and minimum temperature by the perfect prog method, and such forecasts were implemented on or about September 19, 1968. By 1972, Art Pore and others in the Marine Branch were producing wind wave charts based on NMC’s PE model. Also, Ron Reap had programmed a trajectory model based on NMC’s PE model and was furnishing forecasts of temperature and dew point primarily for use in severe storm forecasting.

Breakthroughs occurred when the PP temperature forecasts replaced the NMC forecaster prepared product in 1970 and when the MOS probability of precipitation (PoP) forecasts replaced the NMC forecaster prepared product in 1972.

This paper will review the early postprocessing work in TDL and some of the earlier statistical work that enabled postprocessing.