Kathryn K. Hughes
Meteorological Development Laboratory Office of Science and Technology National Weather Service, NOAA
Since September 2000, the Meteorological Development Laboratory (MDL) of the National Weather Service (NWS) has provided forecasts of thunderstorms and severe thunderstorms from 6 to 72 hours in advance. The Model Ouput Statistics (MOS) technique was used to generate forecast equations to predict the probability of thunderstorms and severe thunderstorms based on output from the 0000 and 1200 UTC cycles of the Aviation (AVN) run of the Global Spectral Model. These thunderstorm probability forecasts are valid for 6-, 12-, and 24-h time periods for the contiguous U.S. (CONUS).
Cloud-to-ground lightning data from the National Lightning Detection Network (NLDN), provided by NASA's Global Hydrology Resource Center in Huntsville, Alabama, were used to define the presence of a thunderstorm in the MOS predictand data. Storm data, consisting of reports of large hail, tornadoes, and high winds, were collected by NWS forecast offices and provided to MDL by the NWS Office of Climate, Water, and Weather Services. These storm reports were used to define the occurrence of a severe thunderstorm for the MOS development. The severe thunderstorm predictand was made conditional on the occurrence of a thunderstorm. Because all of the data in these predictand data sets were random in place and time, special effort was required to place all of the observational data on the same grid so both types of data could be analyzed and compared. Because of the sparsity of the storm data, a Geographical Information System (GIS) was also used to study the relationship of the storm reports to the population density of the area covered by the grid. Data from areas with very low population density were not included in the development sample; however, operational forecasts are generated for the full grid.
Five years of the cloud-to-ground lightning and storm data (April 1994 - March 1999) were used to develop lightning and severe weather climatologies for the CONUS for 6-, 12-, and 24-h periods. Over 128 million cloud-to-ground lightning strikes, and 100,000 storm reports were processed to generate monthly relative frequencies for every grid point used in the development. These relative frequencies were then offered as potential predictors and were often selected by the linear regression program, along with model predictors of stability, wind, and temperature, as part of the forecast equations.
Details of the AVN MOS thunderstorm development and verifications of the skill of the new AVN MOS thunderstorm forecasts are presented in this paper. The AVN MOS thunderstorm forecasts are more skillful than the older NGM MOS system, with much of this skill due to the improvements in the NLDN data, as well as improvements in the numerical models. The planned development of new Eta-based and medium-range MOS thunderstorm forecast equations is also discussed.