The use of SFS as input pattern allows one to build a rapid diagnosis system for ecological monitoring. The proposed system is able to detect and to classify organic pollution in water environment. The performance of the system is insensitive to the DOM spectrum variations. Pollution classification is performed either on the basis of "generalized" classes of pollutants (phenols, light oil products, medium oil products, lubricants) or on the basis of end-user library of possible pollutants.

In the framework of the mathematical model a geometric interpretation of input data was developed. As a measure of similarity of SFSs of different pollutants an angle between planes containing corresponding SFSs is used. This model makes it possible a) to determine what pollutants should be joined into one class during design of classifier, and b) to determine minimal detectable concentrations of each pollutant. Within this approach, the distance between a presented pattern and each of the subspaces of different classes is calculated, and the class of the nearest subspace is considered to be the class of the pattern. The comparison of neural network and statistics based approaches shows that the latter one allows to detect lesser concentrations of pollutants. It also gives one an opportunity to effectively add new etalones to the systems' catalogue and to rapidly tune the system to the specified catalogue