Two years of per minute data, measured by sensors located in five cities of Reunion Island, are investigated in this paper. Prior exploratory data analyses have enabled the statistical comparison of characteristics of cities with respect to the measured weather variables (diffuse and overall solar fluxes, atmospheric pressure, moisture, temperature, wind speed and direction). Data was preprocessed and univariate time-series and multivariate time-series aggregated over hours or days were analyzed in order to build simple and effective prediction models. A classical clustering approach was performed. Groups of days sharing weather parameters in common were found by two biclustering algorithms. The characterisation of found biclusters and their succession displayed in a calendar-based visualization tool have helped assess their interest.