Constrained clustering of the precipitation regime in Greece Eftychia Rousi *1 , Christina Anagnostopoulou †1 , Angelos Mimis ‡2 and Marianthi Stamou §2 1 Department of Meteorology and Climatology, Aristotle University of Thessaloniki, Greece 2 Department of Economic and Regional Development, Panteion University of Athens, Greece November 4, 2014 Summary The aim of this paper is an objective clustering of the precipitation regime in Greece. The data consists of winter daily precipitation values obtained from a Regional Climate Model, the RACMO2/KNMI, for the period 1971-2000. The constrained clustering method is implemented by using three different linkages, single, complete and average, and for three different cluster numbers, 10, 20 and 30. Average and complete linkage both performed well, with the latter proving to be more detailed and its spatial resolution presents many similarities to the original data. The 20 and 30 clusters are clearly more representative than the 10 cluster results. KEYWORDS: Constrained Clustering, Precipitation, Greece, Regional Climate Model. 1. Introduction The problem of defining the various climate zones has a wide range of uses (Iyigun et al., 2013). These include the redefinition of climate zones and rainfall regimes as a result of ongoing climate changes while at the same time examining the reasons that lead to those changes. Also these have a direct effect to hydrology and flora. So the regional water management as well the farming strategies are affected. In this context, the famous classification system of Köppen – Geiger has emerged, which was originally published by Köppen in 1918. This system provides a set of rules applied to variables derived from long term values for temperature and precipitation. In these, with several rules at hand, various locations are classified into climate types (Cannon, 2012). This rule based approach has been adopted and extended by various researchers, as for example by Thornthwaite who by following manual classifications projected the various locations into climate regions which exhibit climate homogeneity. With the widespread use of personal computers a different approach has emerged. In this, climate classification is performed by clustering algorithms based on the assumption that areas with similar values of variables characterizing climate, such as temperature or precipitation, can be classified in the same climate type. In this way, the climate types are directly defined by the data. In this methodology, usually a two step approach is adopted. Firstly, a principal component analysis (PCA), followed by clustering analysis (CA) (Fovell and Fovell, 1993; Cannon, 2012). Those approaches treat the spatial problem of climate zones in an aspatial way, meaning that an area * erousi@geo.auth.gr † chanag@geo.auth.gr ‡ mimis@panteion.gr § marianthi.stamou@panteion.gr