The VSRF module is an observation-based system that will provide the best possible forecast at very-short range. However, the skill of such a technique degrades rapidly in time. For that reason, a technique based on numerical weather prediction model output should prove to be superior for short-range forecasts beyond about 6-12 hours. It was decided to develop a perfect-prog system to forecast the different aviation weather parameters required to write a TAF.
The 40-year databases of hourly observations and predictors have been completed. NCEP reanalyses were used to derive site-specific predictors such as temperature, vorticity, moisture advection, stability indices, etc. Polynomial and spline interpolation were used to interpolate data hourly to specific airports from a 2.5 degree grid with a time resolution of six hours.
Equations for ceilings and visibilities have been produced using a multiple discriminant analysis (MDA) technique and various seasonal stratifications. Results improve with the number of seasons (from one to four seasons). As the number of seasons increases, the MDA software merges some categories together to compensate for the loss of cases. A four-season stratification of six months per season (i.e. two-month overlap) is being investigated. Preliminary results indicate this approach to be slightly less skillful than a straight four-season stratification.