Geoscience Reference
In-Depth Information
COMBINATION OF CLUSTER ANALYSIS
AND DISCRIMINATION ANALYSIS USING
SELF-ORGANIZING MAP
GWO-FONG LIN
∗
, CHUN-MING WANG
Department of Civil Engineering, National Taiwan University
Taipei 10617, Taiwan
∗
gflin@ntu.edu.tw
Regionalization is an important technique that uses existing information to
extrapolate where the information is required but cannot be obtained. The
cluster analysis and discrimination analysis are the two important procedures
of regionalization. In this paper, a simple method based on the self-organizing
map is proposed to combine the cluster analysis and the discrimination anal-
ysis. The advantages of the proposed method are that it can determine the
proper number of clusters, reveal the relative relationship of the input pat-
terns and allocate the unknown patterns into known clusters. The design
hyetographs of northern Taiwan are analyzed using the proposed method. The
clustering results of the design hyetographs of northern Taiwan using the pro-
posed method exhibit homogeneities within clusters and heterogeneities among
clusters. Regarding the capability of determining the proper number of clus-
ters, the proposed method is superior to conventional clustering method. The
discrimination results also show that the assignments of unknown patterns to
known clusters are reasonable using the proposed method. Thus the proposed
method can be applied to regionalization to reduce the complexity and the
di
culty.
1. Introduction
There is a problem that some specific hydrological informations are required
at a certain location where the necessary information cannot be easily
obtained often encountered by hydrological engineers. These problems can
be solved by using the regionalization. Regionalization is used to extrap-
olate some hydrological informations from sites at which the hydrological
information can be derived to others at which the hydrological informa-
tions are required but unavailable.
1
The processes of regionalization may
often combine several procedures, including the cluster analysis and the
discrimination analysis.
2
,
3
The cluster analysis is to explore the relative
relationships and the grouping of the hydrological factors. However, dif-
ferent methods of cluster analysis applied to the same set of data often
25
