This study investigates how the number of data collected for algorithm development in typical case-1 waters and optically complex cases can affect the remote sensing of ocean color (OC) products. Marine conditions dominated by the colored dissolved organic matter (CDOM) and nonalgal particles are considered. The applied OC inversion schemes are based on multilayer perceptron (MLP) neural nets for data classification and regression. Simulated data for MLP training are generated with a forward OC model. Results show that a disproportion of samples representing different marine optical cases influences the MLP learning and hence also the data product retrieval, mostly in mixed case-1 and CDOM-dominated environments. A postclassification correction is then employed for performance improvements. Methodological developments are presented, acknowledging the coastal water monitoring prioritized by the Copernicus Earth Observation program.
|Number of pages||13|
|Journal||IEEE Transactions on Geoscience and Remote Sensing|
|Publication status||Published - Nov 2016|
- Inverse problems
- remote sensing