In this work we aim at generating association rules starting from meteorological measurements from a set of heterogeneous sensors displaced in a region. To create rules starting from the statistical distribution of the data we adaptively extract dictionaries of values. We use these dictionaries to reduce the data dimensionality and represent the values in a symbolic form. This representation is driven by the set of values in the training set and is suitable for the extraction of rules with traditional methods. Furthermore we adopt the boosting technique to build strong classifiers out of simpler association rules: their use shows promising results with respect to their accuracy a sensible increase in performance.