Principal component analysis a tool for interpretation of multispectral remote sensing data

• Main Authors: Krawczyk, H., Neumann, A., Hetscher, M., Zimmermann, G., Walzel, T.
• Other Authors: Office of Naval Reseach, Ocean, Atmosphere, and Space S&T Department
• Format: Conference Object
• Language: unknown
• Published: 1998
• Subjects: Institut für Weltraumsensorik; North Atlantic
• Online Access: http://elib.dlr.de/10702/

The investigation of the ocean and the coastal zones plays an important role in climatological and ecological research. The application of remote sensing devices is an effective method to regularly obtain data over large areas. For the investigation of the bio-optical state of oceanic waters the use of the visible and near-infrared parts of the electromagnetic spectrum had to be rendered as very effective ones. Since CZCS has stopped it's work, a new generation of multispectral devices with higher spectral resolution and more channels has been developed, e.g. SeaWifs, OCTS, MOS and are planned for future missions, like MERIS and MODIS. These devices need new interpretation techniques to benefit from the higher information contained in the data. Especially in coastal waters together with chlorophyll pigments there are additional components influencing the spectral behavior of the signal, e.g. Gelbstoff and Sediments. The knowledge of the distribution of these substances is important for the investigation of human influence of the ecosystem especially near the coast, where river discharges can bring a significant amount of substances into open waters. The Principal Component Analysis (PCA) is an effective mathematical tool for the investigation of the information content of multispectral data. It allows assertions of the intrinsic dimensionality, characterizing the maximal number of independent parameters, which can be estimated from the data. The principal components itself are these "most informative" parameters. Because of the pure mathematical character of the PCA, it is very difficult to give them a physical sense. In the presentation will be shown how with the help of the PCA an effective inversion algorithm can be constructed for multispectral remote sensing data. It will be used the fact that the PCA is an reversible transformation, similar to the Fourier transformation. With the application of a specific bio-optical and atmospheric model can be constructed an effective linear estimator for water constituents and atmospheric turbidity applicable to real satellite data. A special property of the algorithm is that no separate atmospheric correction is needed. This is contained within in the algorithm, the atmospheric state is estimated equal ranked with the water constituents. This algorithm was developed for the Modular Optical Scanner (MOS), regularly sending data since April 1996. Examples from Baltic Sea, Mediterranean Sea and North Atlantic will demonstrate the potential of the algorithm.