Abstract:
|
The aim of this paper is to introduce a methodology for defining groups from regionalized compositional
data, through a hierarchical clustering algorithm aware of both the spatial dependence
and the compositional character of the data set. This method is used to define a regionalization of
Catalunya (NE Spain) with respect to its precipitation patterns in the Winter season. This region
is characterized by a highly contrasted topography, which plays a dominant role in the spatial
distribution of precipitation. Each rain gauge station is characterized by the relative frequencies
of occurrence of six intervals of daily precipitation amount (classes ranging from “no rain” for
precipitation below 3 mm, to “heavy storm” above 50 mm). Recognizing that frequencies are compositional
data, the spatial dependence of this data set has been characterized by variograms of the
set of all pair-wise log-ratios, in the fashion of the variation matrix. Then, a Mahalanobis distance
between stations has been defined using these variograms to ensure that gauges with high spatial
correlation get smaller distances. This spatially-dependent distance criterion has been used in a
Ward hierarhical cluster method to define the regions. Results reveal 5 quite homogeneous groups
of stations, which can be mostly ascribed a physical meaning. Finally, possible links to regional
circulation patterns are discussed |