The Graphical Forecast Editor (GFE) is the grid editing component of the Interactive Forecast Preparation System (IFPS) which allows meteorologists to define weather forecasts in a digital database. Since forecasters are required to adjust and monitor literally millions of forecast points per forecast shift over the seven-day forecast period, the capability to automatically generate high-quality text from the digital data is an essential part of the system. This paper describes the challenge of generating text for a set of highly interdependent weather elements, and offers a solution that relies on multipass processing of a tree-like data structure representing the forecast product. This approach gives each piece access to the whole so that dependencies can be enforced, and then builds the text iteratively. As a simple example, the probability of precipitation wording "Probability of snow 20 percent" is dependent on the type of weather expected, i.e., "snow." So the weather phrase may be built in the first pass, and the probability of precipitation phrase can be built in a second pass having access to the results of the weather phrase. The same principle is applied to generate more complex phrasing such as local effects.

Finally, we look at the parallels to forecast methodology and the possibility of applying these concepts to the grid-editing portion of the forecast process.