Explaining Weather Bulletins via Inductive Logic Programming

This paper presents an interpretable pipeline for explaining weather bulletins with Inductive Logic Programming (ILP). Using the FastLAS2 framework, it converts simulated meteorological raw data and OSMER FVG bulletins from Italy’s Friuli-Venezia Giulia region into Answer Set Programming facts and ILP examples. FastLAS2 then infers a hypothesis that is translated into natural language, explaining why human forecasters selected particular symbols in the bulletin’s pictogram map. The authors describe the approach as region-independent and applicable to bulletins from other sources, while the reported pipeline is based on simulated raw data.