No abstract

The reliability of the weather forecast models is a complex issue since it depends on numerous parameters and the technical infrastructure which supports them. In doing so, there is a need for advanced works oriented towards a better understanding of these models and the analysis of main associated parameters. Our approach is to study the applicability of the extracted association rules to provide a clearer understanding of atmospheric exchanges. In this work, the proposed methodology is based on the discovery of the interesting interpretable relationships between measured meteorological parameters at the Atmospheric Research Center of Lannemezan (South-West of France). In the preprocessing step, the proposed method is considered to be effectively flexible to account for data uncertainties, unlike the majority of classical evaluation methods mainly directed towards the reduction of variables and data redundancy. In postprocessing, the advantage of our approach is that the extracted rules are a metamodeling of interpretable useful knowledge for the clarity and conciseness of its representation. Moreover, in the processing, the interpretability in data sciences is recent and still in its infancy. The generated association rules with their statistical and semantic interpretations have globally highlighted the possibilities of explicit analysis of meteorological parameters. This study showed that among the generated relevant rules, three parameters (temperature, humidity, wind speed) have a high frequency in the antecedents of the rules and that the only consequence is rain. This is useful for the identification of potential improvements and gaps in the existing models of atmospheric observations, in particular, to understand the related parameterizations to the productivity of the rain phenomenon.