Author(s)

J. M. Angosto, B. Elvira-Rendueles, J. Bayo, J. Moreno, N. Vergara, J. Moreno-Clavel & S. Moreno-Grau

Abstract

Wind classification through cluster analysis for the development of predictive statistical models on atmospheric pollution J. M. Angosto, B. Elvira-Rendueles, J. Bayo, J. Moreno, N. Vergara, J. Moreno-Clavel & S. Moreno-Grau Department of Chemical and Environmental Engineering, Technical University of Cartagena, Cartagena, Spain Abstract This paper presents the results obtained after the use of a two-stage clustering analysis to get a classification scheme of surface winds. For this purpose, we have studied data from four meteorological towers within the city of Cartagena, a Mediterranean city located at Southeast of Spain, with a complex topography for wind circulation. Semi-hourly data for wind parameter have been used through different hours of the day, representing the daily wind behaviour. These values, projected on a X-Y axis have been processed thorough a factor analysis in order to avoid colineality. After the factor scores were calculated from the factorial matrix, a two-step cluster analysis procedure, using both hierarchical (average linkage) and non-hierarchical techniques (k-means) was performed. This classification scheme identified five distinct wind flows, which were represented for each meteorological station using wind roses. An homogeneous behaviour for each cluster within each meteorological station was determined. Once different wind flows were identified for this area, total daily pollen concentrations were studied, associated to each cluster situation. In this sense, statistically significant differences were determined. This analysis will be further used to develop refined regression models of pollen concentrations, in order to get better results than those obtained with the whole data. 1 Introduction Nowadays, one of the first objectives within the atmospheric pollution studies is to evaluate the environmental impact carried out for a single pollutant delivered

Keywords