USING REMOTE SENSING TO DEFINE WATER DEPTH IN THE PEARL RIVER ESTUARY

We established a variety of statistical correlation models between water depth and remote sensing image in this research. A best model was selected for feasibility study in two experimental zones with different concentrations of suspended sediment in the Pearl River Estuary. In the testing area 1 with high suspension, the correlation between the observed water depth and the band DN values of remote sensing image was less than 0.5, and the capacity to define the actual water depth was poor. The correlation of the statistical model was much improved after taking the sediment factor into account, but the precision of remote sensing water depth remain low and beyond the required precision in application. In the testing area 2, where sediment concentration is low, the correlation between the observed water depth and the band DN values of remote sensing image was higher than 0.7, but there is no improvement after taking the sediment factor into consideration. The precision of exponential model was the highest using B2 as variable and the best results were gained in the area with water depth at 5-10m. The mean-relative-error was 22.5% and the mean-absolute-error was 1.56m, with an overall mean-relative-error of 31.9% and a mean-absolute-error of 1.92m. It means that the water depth derived from remote sensing are more satisfied in the testing area 2. As the conclusion, the nonlinear model was better than the linear model, and the multiple factor model better than single factor model.