S. Kameyama et al., Development of WTI and turbidity estimation model using SMA - application to Kushiro Mire, eastern Hokkaido, Japan, REMOT SEN E, 77(1), 2001, pp. 1-9
A new water-turbidity index (WTI) based on multispectral images was develop
ed and tested at Kushiro Mire, eastern Hokkaido, Japan. An algorithm for tu
rbidity estimation was developed and applied to Landsat TM images to monito
r the turbid water on the mire surface during the snow-melting season. We u
sed spectral mixture analysis (SMA) to produce a turbidity estimation model
. The SMA "unmixes" a mixed pixel determining the fractions due to each spe
ctral end member. In this study, we used four end members (1, alder; 3, ree
d; 3, high-concentration turbid water (485 ppm); 4, low-concentration turbi
d water(10 ppm) measured in the test site. The WTI was determined by the fo
llowing equation: WTI = a(max)/(a(max) + a(min)), where a(max) is abundance
of high-concentration turbid water and a(min) is abundance of low concentr
ation turbid water. The end-member spectra of alder and reed were measured
in the laboratory using specimens collected at the test site. The spectrum
of turbid water was measured at the test sites. The relative abundance of e
ach end member was estimated based on this spectral information using SMA.
The same formula was applied to Landsat TM images. Then we applied the WTI
equation to the end-member images to obtain a WTI map. In the mire wetland
region, turbid water spreads under alder trees and reed grasses. Tb verify
our turbidity estimation method based on WTI under these conditions, we con
structed a small experimental wetland consisting of mixed stands of alder a
cid reed. WTI was calculated from the mixed spectrum of this "artificial we
tland" and the regression curve for the relation between WTI and the actual
turbidity was determined (R-2=.91). Finally, this regression equation was
used to derive a turbidity map from the WTI image. (C) 2001 Elsevier Scienc
e Inc. All rights reserved.