Abstract
The objective of this research was to develop an accurate bathymetric map for an area around Roatan Island, Honduras using high-resolution multispectral IKONOS data based on a variation of a linear regression model. Linear regression models estimate water depths by regressing brightness values over known benthos (albeit non-homogeneous) and known depths. However, we contend that if mixed bottom types are used, the regression coefficients deteriorate because the variability in brightness values from a heterogeneous bottom has a deleterious effect on the correlation coefficient. By selecting uniform bottom types, this variability can be reduced and a strong correlation between depth and brightness value can be established, thus improving the accuracy of estimated depths. Three uniform bottom types (seagrass, coral, and sand) were selected, and the transformed brightness values derived from principal components analysis for each bottom type were regressed against known depths. The most statistically significant coefficient (r 2 = 0.909 for seagrass benthos) was then used in the depth estimation algorithm and a bathymetric map was derived. A comparative evaluation between estimated and actual depths was performed and the bathymetric map was found to be within a standard error of 0.648 m. Consequently, our results suggest that accurate depth estimates can be derived by using transformed input brightness values over homogeneous bottom types from IKONOS multispectral imagery.