| Bathymetric maps derived from passive remotely sensed images have the attractive features of being both readily available for most areas and relatively affordable. Unfortunately, because of their lack of accuracy such maps have not been widely used to date. But with the emergence of the newer hyperspectral data, the difficulty in accurately classifying coastal waters' bottom types and depth should be greatly reduced.; A previously developed derivative based bathymetric map procedure showed promise when applied to a synthetic hyperspectral data set. This data set was relatively constrained, and it was also error free. Thus to fully evaluate the effectiveness of the procedure, this method should be applied to real imagery. While this would seem relatively straight forward, there are many issues that were ignored in the original study that would have to be investigated.; This investigation first dealt with the gathering of the in situ data needed to train the algorithm. A commercial instrument that measured the data directly was not available. So instrumentation and the corresponding procedures were developed. Also a method for handling areas in which only partial data sets could be measured was investigated.; Next, this study looked in to the hyperspectral data produced by the Naval Research Laboratory's PHILLS hyperspectral scanner. This instrument is still in its developmental stages, and thus there were still some error in their data sets. Causes for these errors were examined as well as how to handle them.; Finally, the bathymetric derivative method was applied to the corrected PHILLS data. A variety of study sites and conditions were selected to determine its robustness. A promising procedure to develop a bottom change map was constructed. This procedure still needs to be properly validated. |
