(Received: February 28, 2001; Accepted for publication: October 10, 2001; Published on Web: December 7, 2001)
We attempted to extract chemical parameter characterizing the upper, middle and lower stream by the principal component and the analysis of differential coefficients of input parameter for perceptron type neural network with three layers. The analysis of differential coefficients of input parameter for perceptron type neural network was developed by Aoyama  and was newly equipped into a neural network simulator Neco. The data used are 12 chemical parameters at 17 points along the main stream of the Tamagawa river in Tokyo, Japan, for 1997-1999 .
The K-L plot of the first and second principal components (Figure 4) well divides 17 points into three groups corresponding to the three regions: upper, middle and lower streams, respectively. From results of the analysis of differential coefficients of input parameter for perceptron type neural network, Cl-, COND and NH4-N have relatively large differential coefficients and divide middle and lower streams. DO and pH are large in upper stream of Tamagawa river (Figure 5). The first principal component classifies well two groups: upper and middle-lower streams on the K-L plots. This result suggests that the water contamination is more drastic in the midstream of Tamagawa river than downstream. The water contamination in midstream should be decreased for keeping Tamagawa river clean.
Keywords: Water Contamination, Principal Component Analysis, Differential Coefficients Analysis of Input Parameter for Neural Network, Upper Stream, Middle Stream, Lower Stream
Text in Japanese