Aviation terminal forecasts (TAFs) are site-specific forecasts that are currently prepared every 6h manually, using guidance from the operational NWP models and the most recent available local observations, including radar and satellite data. The Meteorological Service of Canada (MSC) believes that gains in forecast production efficiency can be realised by producing objective TAFs, leaving the final control of the forecast contents with the operational forecaster. Elements of interest to aviation are: ceiling, visibility, weather and obstructions to vision, and wind speed and direction. In the first phase of the project only the two most important elements for aviation, ceiling and visibility, will be considered. However, once the tools for those two elements are developed, it will be very easy to extend the development to the remaining elements.

It was decided to use statistical methods, which are relatively easy to develop and to run. The statistical software, the utilities to produce TAFs from the output of statistical techniques and the utilities to edit the resulting TAFs are collectively called TAFTools. Dallavalle and Dagostaro (1995) have shown that simple persistence is a very strong competitor for very short-range forecasts. Recent work by Vislocky and Fritsch (1997) supports the idea that a system based on observations only should be quite powerful for very-short range forecasting.

Those considerations have led us to attack the TAF forecast problem from two different angles: a component based on observations only for the very-short range forecasts (VSRF) and a Perfect-Prog component based on model output for the short-range forecast (SRF). We would expect that the accuracy of the observations-based forecast would deteriorate more rapidly in time than the accuracy of a model-based system. The third main component of TAFTools, the blending module, will not be discussed here.

Results indicate that ceiling and visibility probabilistic forecasts from the VSRF module are equivalent to those obtained from a conditional climatology technique. These forecasts are largely superior to both simple persistence and climatology. A procedure was designed to transform the probabilistic forecasts into categorical ones using a unit bias approach. This approach reproduces the observed category distribution. The VSRF now produces real-time ceiling and visibility forecasts, both probabilistic and categorical, for 8 Canadian sites.

The SRF module based on model outputs has been semi-operational for a few months. Different strategies were considered to transform probabilistic forecasts into categorical ones. Results and verifications for probabilistic and categorical forecasts will be shown.