Analysis of Wind Roses Using Hierarchical Cluster and Multidimensional Scaling Analysis at La Plata, Argentina

Abstract

Knowledge of frequency wind patterns is very important for air pollution modelling, especially in a city like La Plata (approximately 850,000 inhabitants) with high vehicular and industrial activities and no air monitoring network. An hourly wind analysis was carried out on data from two local weather stations (points A and J). An initial result was that, in spite of differences in data quality, the local weather stations observations were consistent with local and regional National Meteorological Service (NMS) monthly based observations. Two non conventional multivariate statistical methods were employed to further analyse hourly data at points A and J. Hierarchical cluster resulted in a good summarising tool to visualise prevailing hourly winds. Resultant vectors emerging from the clustering process showed good similarity between sites and seasons; this allowed a further visualization of the average diurnal wind development. Multidimensional scaling (MDS) permitted a pairwise comparison of a large number of hourly wind roses. These wind roses were more similar to each other in colder seasons and at site A (the one that is closer to the river) than in warmer seasons and at site J. Most of the observed variations regarding seasons and sites revealed by cluster and MDS analysis are explained in terms of the sea-land breeze circulations. The methodology applied proved to be of utility for simplifying the analysis of high dimensional data with numerous observations.