Computational Intelligence Approaches to Ocean Color Inversion
The spectral qualities of light reflected from the ocean surface (termed "ocean color") can be used to deduce the quantities and qualities of the absorbing and backscattering constituents within the water. Parameters such as phytoplankton abundance, community structure, and the amounts and composition of o ther dissolved and particulate matter in the ocean can be used to gain knowledge about the biogeochemical cycling of carbon or to assess and monitor coastal water quality. Though existing bio-optical algorithms (such as to retrieve the amount of phytoplankton pigment) have been successful in satellite remote sensing of ocean color in open ocean water, there has been limited success in more optically complex (i.e. coastal) waters, due primarily to the complex nature of the light field and constituent dynamics in these environments. Computational intelligence-based techniques can be used to accurately and efficiently estimate bio-optical parameters in these environments from remote sensed data. This thesis presents two approaches to bio-optical inversion that employ computational intelligence methods. First, neural network-based models are presented for estimation of phytoplankton absorption and other bio-optical properties in optically diverse waters. Models were trained using multi-spectral remote sensing reflectance (Rrs) simulated for a wide range of water inherent optical properties. Inversion of the bio-optical model was carried out to retrieve not only phytoplankton absorption, but also other bio-optical parameters. Furthermore, the model was tested on a large database of in situ observations of Rrs matched with measured concentration of chlorophyll-a, and the retrieval results proved reasonable (2 = 0.76) given the strong variation of the relationship between chlorophyll and phytoplankton absorption. Given hyperspectral Rrs, a similar model could offer the capability for efficient processing of ocean color imagery in order to retrieve these biooptical parameters. Second, the inversion of ocean color Rrs measurements is cast as an spectral matching (optimization) problem, where parameters of a forward model are optimized in order to make the forward modeled spectral reflectance match a given measured spectrum. Here, a simulated ocean color dataset is used to test the capability of a recently introduced global optimization process, particle swarm optimization (PSO), in the retrieval of optical properties from ocean color. The performance of the PSO method is compared with the more common genetic algorithm (GA) in terms of model accuracy and computation time. The PSO method has been shown to outperform the GA in terms of model error. Of particular importance for application to ocean color remote sensing is the speed advantage which PSO affords over GA.