Retrieval of oceanic suspended sediment concentration with support vector regression

Proceedings of SPIE, the International Society for Optical Engineering/Proceedings of SPIE2008Vol. 7285, pp. 728507–728507

Ligang Cheng, Ying Zhang

Abstract

The aim of this study is to examine the feasibility of Support vector regression (SVR) in retrieval of suspended sediment concentration by comparing it with band ratio regression models. First, the remote sensing reflectance and the suspended sediment concentrations were measured in field and in laboratory. The in situ dataset and laboratory dataset were used in t developing retrieval models based on support vector regression and band ratio regression. Second, we select band ratio regression model with high R-square value and low Root Mean Squared Error as the best band ratio regression model. Finally, the best band ratio regression model was compared with SVR model in different datasets by leave-one-out cross validation. The experimental results demonstrate that the prediction accuracy of support vector regression outperforms the band ratio regression models based on the mean absolute error in general. SVR using all bands yielded slightly superior results than using TM1 and TM4 bands in terms of accuracy. The findings suggest that the SVR model is available using all bands data. The support vector regression can be applied in retrieval of suspended sediment concentration without selecting bands and constructing band ratio expression. SVR is a promising alternative to suspended sediment retrieval models.

Abstract

Related Papers