Pythonissa: A Visual Language for Weather Forecast Preparation

This paper reports our current work on Pythonissa, a visual language tailored to graphically describe weather forecast bulletins. The internal representation of a bulletin is a sequence of graphs and visual sentences are therefore represented as graph transformations. Pythonissa is being implemented as part of a more generic framework, based on the well-known visualization software Vis5D, and using the scripting language Python. 1 Motivation Weather forecasting dramatically evolved in the recent years thanks to numerical prediction models and to a world-wide infrastructure of data acquisition and distribution. In spite of the quality of models, however, preparation of forecasts is still a job for experienced meteorologists (forecasters) which need to analyze model results and many other kind of information before writing a bulletin. Since the volume of data to explore is increasingly high, forecasters routinely resort to visualization tools. The normal result of that process is a textual bulletin, unadequate for modern image-oriented media like TV and WWW. Even for the press, a graphical representation of the forecast for sure is more attractive. To make forecast preparation easier and faster, forecasters should use a single, integrated software tool which allowed themselves to simultaneously observe the available data and draw forecasts over them. Currently, AFPS [1] provides such environment, but our aim is to produce a tool whith which forecasters can also generate the text bulletin from the drawn one. 2 A formally-specified visual language Drawings are normally a kind of nonor poorly-structured data, but to generate the forecast text we need to assure that meteorologists could not draw invalid forecasts. So, the bulletin must be internally represented in a parsable form. The adopted solution was modeling the bulletin as a sequence of graphs and use Graph Grammar techniques to parse them. Figure 1 shows an example. Composing the bulletin is then reduced to the task of laying out it’s elements over the visualized data using direct manipulation techniques. 3 A general-purpose framework Instead of writing a whole application to support the new language, we decided to include it as a feature in an existing visualization tool, Vis5D [2]. To make such task easMj Mj