Divisive and hierarchical clustering techniques to analyse variability of rainfall distribution patterns in a Mediterranean region

One of the main characteristics of the Mediterranean climate is the irregul ar distribution of rainfalls and its high variability from year to year, wh ich makes it difficult to know if Mediterranean rainfalls are changing, in this work, a combination of divisive (K-means algorithm) and hierarchical ( average and Ward's methods) cluster techniques were used to analyse rainfal l distribution patterns over the year and their changes over time. The stud y was performed using daily data (1889-1999) from the Alt Penedes region (N E Spain), and annual, spring and autumn rainfall were analysed. The variabl es included in the cluster analysis were spring and autumn rainfalls, which represent on average 70% of the annual rainfall. The K-means and Ward's me thods led to similar classifications of the observations, whereas the avera ge method had a lesser discrimination power. However, the combination of al l of them was useful in the identification of the rainfall distribution pat terns over the years and their changes with time. Patterns including dry sp rings and wet autumns have increased during the last decade, although any c onsistent trend in annual rainfall was not found. (C) 2001 Elsevier Science B.V, All rights reserved.