3rd Conference on Artificial Intelligence Applications to the Environmental Science


Classifying TOA Ocean Color signals by using a Kohonen map methodology

Sylvie Thiria, University of Paris, Paris, France; and A. Niang, F. Badran, C. Moulin, and M. Crepon

The Kohonen map with PRSOM was used to analyze a time sequence of SeaWifs images These images were observed on the Mediterranean Sea from the 6th of August to the 12th. We processed the normalized reflectance of 4 near infra-Red channels (510, 670, 765 and 865 nm) by first using a 10X10 Kohonen map and then we aggregated these 10X10 classes into 4 classes by using the PRSOM algorithm. The classifier was trained on a one year (1999) Mediterranean SeaWifs images using a temporal homogeneity. (2 images/month). The 4 classes show the presence of an important event of Saharian dust coming from the Sahara, crossing the Mediterranean Sea and invading North of the Mediterranean. A class (the Red one) can be interpreted as devoted to the optical thickness. A meteorological map taken the 10th of August shows a strong South West wind supporting the above interpretation. This study clearly shows the possibility to use the above algorithm for automatically classify the aerosols at the Top of the Atmosphere. A problem is to have expertise to label the different classes.

Poster Session 1, All AI applications
Tuesday, 11 February 2003, 9:45 AM-11:00 AM

Previous paper  Next paper

Browse or search entire meeting

AMS Home Page