Use of the novelty detection technique to identify the range of applicability of empirical ocean color algorithms

Novelty detection is used to identify the range of applicability of empirical ocean color algorithms. This method is based on the assumption that the level of accuracy of the algorithm output depends on the representativeness of inputs in the training dataset. The effectiveness of the novelty detection method is assessed using two datasets: one representative of the northern Adriatic Sea coastal waters and the other representative of open sea waters. The two datasets are independently used to develop neural network algorithms for the retrieval of chlorophyll-a concentration (Chl-a). The range of applicability of the individual algorithms is presented using remote sensing data derived from the Sea-viewing Wide-Field-of-view Sensor (SeaWiFS) for three selected regions: the central Mediterranean Sea, the North Sea, and the Baltic Sea. An extension of the novelty detection technique is also proposed to blend the individual algorithms and to avoid discontinuities in the resulting Chl-a maps.