This present work is carried out during my ERASMUS exchange for the year 2019-2020 between Clermont Auvergne University - Institute of Computer Science, Modeling and their Applications and Universitat Politècnica de Catalunya - Facultat d'Informatica de Barcelona, which involved the Italian company "TriM" whose domain of expertise is Meteorology. The latter focuses on weather patterns usually defined as repeated values of weather variables, such as cloud coverage, pressure, wind speed and direction... etc. Identification of the evolution of a certain weather variable depending on other ones, is also important in case of a decision making process depending on weather conditions. The last case, is what presented in this thesis dissertation, we are focusing on finding weather patterns helping decisions for the Olympic sailing - Tokyo 2020, based on only two main variables : wind speed and wind direction, since they are the most important components for sailors. The previous work on the same subject have proposed a flexible framework that is able to perform a manual clustering analysis of wind data, using unsupervised learning approach, trying to recognise and show details of wind patterns for specific areas, and making comparisons between those patterns to deeply understand their similarities and differences. The aim of this work firstly, is to improve what was done before, since the environment was correctly implemented but not properly tested, and to execute the previous implementation on the new data of Summer 2019. Then, we focus on automating the clustering process, making the decision about number of clusters to be independent and automatic, and building an area-based clustering process, (i.e: a clustering which is not performed on the whole data, but on data splitted by areas), and a comparison software, which can compare between different results of clustering, visualize different meteorological parameters of the matched clusters and generate results as PDF files.
