The search for exceptional water reflectance spectra

Helmut Schiller
GKSS, 21502 Geesthacht, Germany
schiller@gkss.de

ABSTRACT

Remote sensing is an indispensable tool for the monitoring of coastal water. Measurements of multispectral VIR water reflectances are used to derive parameters which allow to assess e.g. water quality. But using such procedures does not guarantee to meet an essential aim of monitoring: detection of exceptional events i.e. events which do not fit to the more or less elaborated models used in the derivation of the above mentioned parameters.

To meet this aim we developed a procedure which can be summarised as follows:

1. From the region of interest (space and time) sample and collect spectra.
2. Use PCA to filter the spectra and reduce dimensionality.
3. In this reduced dimensionality space generate points enwrapping the measured ones.
4. Train an artificial neural network to recognise the respective group membership.

Using examples from the North Sea, Skagerak and the Baltic Sea we demonstrate the high sensitivity of this procedure to detect exceptional spectra.

Last Update: 2005-04-7