390
C. Barcelo· V. Pawlowsky . G. Bohling
with the assumed model and thus in principle more reliable than those obtained
applying traditional clustering procedures to untransformed compositional data.
2
Case Study
We applied the clustering procedure based on a mixture model using both types
of transformations to granulometric data from the Darss Sill area of the Baltic
Sea (Fig.I). As stated in Davis et al. (1995) and in Bohling et al. (1998), the samples of Darss Sill data set represent the weight percentage of each sample within
each of eight grain size classes: gravel (>2000 !!mm), very coarse sand (20001000 is !!mm), coarse sand (1000-630 !!mm), medium sand (630-400 !!mm),
medium fine sand (400-200 !!mm), fine sand (200-100 !!mm), very find sand
(100-63 !!mm) and silt «63 !!mm). The Darss Sill area represents the primary
bottleneck in the exchange of water between the North Sea and the Baltic Sea. In
the study area the predominant current flow and the sediment transport direction is from the southwest across the Darss Sill and northeast and east into the
Baltic. There is evidence that current patterns in the eastern portion of the study
area follow a counterclockwise circulating pattern. The Darss Sill acts as a barrier between the Mecklenburgian Bight in the southwest and the Arkona Basin
in the east. It consists of glacial tills and is covered in areas with lag sediments.
The Kadet Channel was incised in the glacial tills by postglacial drainage.
Sections 2.1,2.2 and 2.3 show the results obtained applying different clustering procedures to the Darss Sill data set. Results obtained applying the clustering
procedure based on mixture models to data transformed by the additive logratio
transformation are shown in Section 2.2, while the case for data transformed by
a multivariate Box-Cox transformation is presented in Section 2.3. These results
•• :J" ..
e~..,
~'''''''''',;........J':=::::'''--- - - -.. , , ss·
Fig.l Darss Sill and surrounding area of the eastern
part of the Baltic Sea. (After
Davis et al. 1995)
C. Barcelo· V. Pawlowsky . G. Bohling
with the assumed model and thus in principle more reliable than those obtained
applying traditional clustering procedures to untransformed compositional data.
2
Case Study
We applied the clustering procedure based on a mixture model using both types
of transformations to granulometric data from the Darss Sill area of the Baltic
Sea (Fig.I). As stated in Davis et al. (1995) and in Bohling et al. (1998), the samples of Darss Sill data set represent the weight percentage of each sample within
each of eight grain size classes: gravel (>2000 !!mm), very coarse sand (20001000 is !!mm), coarse sand (1000-630 !!mm), medium sand (630-400 !!mm),
medium fine sand (400-200 !!mm), fine sand (200-100 !!mm), very find sand
(100-63 !!mm) and silt «63 !!mm). The Darss Sill area represents the primary
bottleneck in the exchange of water between the North Sea and the Baltic Sea. In
the study area the predominant current flow and the sediment transport direction is from the southwest across the Darss Sill and northeast and east into the
Baltic. There is evidence that current patterns in the eastern portion of the study
area follow a counterclockwise circulating pattern. The Darss Sill acts as a barrier between the Mecklenburgian Bight in the southwest and the Arkona Basin
in the east. It consists of glacial tills and is covered in areas with lag sediments.
The Kadet Channel was incised in the glacial tills by postglacial drainage.
Sections 2.1,2.2 and 2.3 show the results obtained applying different clustering procedures to the Darss Sill data set. Results obtained applying the clustering
procedure based on mixture models to data transformed by the additive logratio
transformation are shown in Section 2.2, while the case for data transformed by
a multivariate Box-Cox transformation is presented in Section 2.3. These results
•• :J" ..
e~..,
~'''''''''',;........J':=::::'''--- - - -.. , , ss·
Fig.l Darss Sill and surrounding area of the eastern
part of the Baltic Sea. (After
Davis et al. 1995)