Models For The Estimation Of Total Suspended Sediment Concentrations In Taihu Lake Using Field Spectral Data And Landsat TM/ETM

The concentration of total suspended sediment in inland waters is one of the important parameters for water quality and water environment evaluation. Total Suspended sediment (SS) is the non-dissolved matter in water which influences the total primary productivity because the quantity of suspended sediment affects the transmission of light in water and the use of light for phytoplankton's growing. However, the conventional method is to measure the SS from point to point, with great expense in time and money. Besides, the spatial distribution of suspended sediment is difficult to obtain in the conventional measurements, because SS is a non-homogeneous water-quality parameter in space. This could be solved through deriving a model for the estimation of SS using remote sensing images and in situ data.In this paper, the models for estimating of SS were built based on field spectral data and Landsat TM/ETM images. The field spectral reflectance of water in Taihu Lake was acquired by ASD FieldSpec Spectroradiometer. Then the correlations between SS and the measured spectral reflectance were analyzed to find the characteristic bands for estimating SS in Taihu Lake. The results reveal that band ratio is better than single band for estimating SS, and the optimal band ratio is varying due to the differences of water quality. R640/R545 is the best for SS estimation in 2003-11-13 while R624/R594 in 2004-7-27. However, the band ratio is around band 2 and band 3 in TM/ETM image, which is equal to TM3/TM2 (or ETM3/ETM2).According to the seasonal characters of SS in Taihu Lake, four TM/ETM images were used, which are in winter, spring, summer and autumn, respectively. The images were processed by 6S atmospheric correction and a 3*3 low-pass filtering before build model.The least squares regression and fuzzy regression were used to build model between remote sensing data and SS in situ. The results demonstrate that SS has the highest correlation with band combination (B2+B3)/(B2/B3). Moreover, this band combination shows good applicability in winter, spring and autumn (R~2>0.52) except summer because of the inhibitory effect between SS and Chla. The winter SS estimation model is the best one (R~2=0.81), which is lnSS=14.656*(B2+B3)/(B2/B3)+1.661, ln(SS) is the natural logarithm of SS, B2 and B3 are the reflectance of band 2 and band 3 in TM/ETM image.Fuzzy regression analysis breaks through the basic hypothesis of "the uncertainty of observation value is randomicity". The fuzzy regression equation was built based on data in 2001-1-15. Compared with the result from least squares regression, fuzzy regression provides more information and is more reliable.At last, the problems in the study and the further researches were discussed.