Improving satellite ocean color observation processing by the use of neural network
Abstract
Spaceborne ocean color sensors which are now operational or under preparation, have a large number of spectral band. NAOC, which is an EC supported programme, proposes to use advanced neural methodology which are well suited to deal with this multi-spectral information for retrieving ocean constituents. NAOC is divided into for major tasks : First NAOC intends to perform classification of the satellite signal at the Top Of the Atmosphere according to specific criteria (energy, pattern of the spectrum) by using Topological Neural network Algorithm. This allows us to extract information on aerosol and on water type which will be used for atmospheric correction and ocean constituent retrieval. Second NAOC is determining improved algorithms for atmospheric correction which is crucial for obtaining accurate ocean product. As atmospheric correction is sensitive to ocean parameter for case 2 waters, NAOC intends to determine specific Multi-Layer-Perceptron (MLPs) for case 1 and case 2 water atmospheric correction. NAOC will then determine specific Neural Network algorithms for ocean constituent retrieval. At last NAOC is developing an advanced inversion algorithm based on the Spectral matching method of Gordon (1997). This algorithm inverts the radiative transfer equations both in the atmosphere and the ocean by using a combination of variationnal and Neural method. As it takes into account both atmospheric and oceanic parameters, it is well suited to deal with absorbing aerosols.
- Publication:
-
34th COSPAR Scientific Assembly
- Pub Date:
- 2002
- Bibcode:
- 2002cosp...34E1369C