Climate Informatics: Identifying relevant weather types in the Lesser Antilles
Résumé
In order to inform people of the risks, meteorologists or climatologists process big volumes of recorded data to predict the effects of the climate change (floods, droughts, rising sea levels, etc.). Weather types have ever been defined in temperate climate of Europe. Such a discrimination yields to identify or localize dry seasons for instance. However in tropical climates as in the Lesser Antilles, so far no result has been set about identifying and discriminating relevant weather types. Researchers in Climate Informatics, who are physicists or computer scientists, are used to apply statistical methods on climate model outputs as data. The objective of the present work is to show how Machine Learning methods allow to detect and identify relevant weather types in the Lesser Antilles. Several clustering algorithms were compared on data recorded in the frame of the european project ERA-Interim 1 : thirty-six years of meteorological parameters, each recorded four times a day, have been downloaded and stored, which represents about 5To of data. The best results were obtained with the Agglomerative Hierarchical Clustering algorithm that is presented in the poster. The method led to the identification of a specific cluster associated to cyclonic events. This weather type has never been identified by algorithm in the Lesser Antilles.