Chapter 8· Analysis of Stream Macroinvertebrate Communities
171
network, with eaeh neuron being represented as j (See Fig. 8.3). The input veetor
x is eonsidered to be an input layer to the Kohonen network, and the set of exergy
measured from benthie maeroinvertebrate eommunities for a eertain period, e.g., 3
months, 4 months, ete., are provided as input to the network in this ease. The
training proeess was the same as for patternizing the eommunity data in the
seetion of .... A. Statie Grouping".
For patterning the ehanges in exergy, monthly measurements of exergy eould be
segmented in intervals in different periods, e.g., 3, 4, 6, ete. In eaeh interval, the
data sets for exergy of different taxa from the same sampie sites were used as input
to the Kohonen network for training. The exergy was normalized to the maximum
of 3 kJ r 1 with the values ranging from 0.01 to 0.99. Detailed proeedure eould be
referred to Park et al. (2001b). Learning rate was assigned at the values of 0.1 -
0.4.
a)
0 .8
• Field
OT+O
>OT+1
!C" 0 .4
q)
><
UJ
0
0
N
D
J
F
M
A
M
J
J
A
S
'97
'98
b)
0.8
.Field
OT+O
>OT+1
0>
:D 0.4
x
UJ
o
o N D J F M A M J
J
A
S
'97
'98
Fig. 8.24. Comparison with field and predieted data of exergy by the
baekpropagation algorithm on benthie maeroinvertebrate eommunities eolleeted in
the Suyong River from Oetober 1997 to September 1998 "without time delay" and
"with one-time delay". a) Soktae Stream, b) Suyong Stream. (From Park et al.
2001b).