Prediction of Phytoplankton Pigment Concentration from Absorption Spectra

Abstract

This thesis studies the relationship between light absorption spectra and pigment concentrations in oceanic waters. Neural networks including Multi-layer Perceptrons and Radial Basis Functions will be used in order to model this relationship. The data will first be investigated by a thorough visualisation before attempting to reconstruct the spectra using forward models. Bayesian learning techniques are then discussed and applied to the retrieval of pigment concentrations. A range of data driven models will be implemented and finally a generative model produced, using Hybrid Monte Carlo sampling techniques. Keywords: absorption spectra, chlorophyll, phytoplankton, pigment concentration retrieval, Principal Components Analysis, Multi-layer Perceptron, Radial Basis Function, Generalised Linear Model, Bayesian methods, Automatic Relevance Determination, Hybrid Monte Carlo.