A new range of multi-plankton biogeochemical models have recently been developed, designed to advance our understanding of the ocean carbon cycle to improve predictions of its future influence on climate. Synoptic measurements of the different phytoplankton communities are required to validate and ultimately improve such models. Measuring ocean colour from satellite is the only method currently available for synoptically monitoring wide-area properties of ocean ecosystems, such as phytoplankton chlorophyll biomass. Recently, a variety of bio-optical methods have been established that use satellite data to identify and differentiate between either phytoplankton functional types (PFTs) or phytoplankton size classes (PSCs). In this thesis, several of these techniques were evaluated against in situ observations (6504 samples) to determine their ability to detect dominant phytoplankton size classes (micro-, nano- and picoplankton). Results show that spectral-response, ecological and abundance-based approaches can all perform with similar accuracy. However, abundance-based approaches provide better spatial retrieval of PSCs. Based on insights into the abundance-based models, and by utilising a large pigment database, a new three-component model was developed which calculates the fractional contributions of three phytoplankton size classes (micro-, nano- and picoplankton) to the overall chlorophyll-a concentration. Using a globally representative, independent, coupled pigment and satellite dataset the model estimates fractional contributions with a mean accuracy of 9.2 % for microplankton, 17.1 % for nanoplankton and 16.1 % for picoplankton. The effect of optical depth on the model parameters was also investigated and explicitly incorporated into the model. Using the three-component model, the two-component absorption model of Sathyendranath et al. (2001) and Devred et al. (2006) was extended to three-component populations of phytoplankton, namely, pico-, nano- and microplankton. The new model infers total and size-dependent phytoplankton absorption as a function of the total chlorophyll-a concentration. A main characteristic of the model is that all the parameters that describe it have biological or optical interpretation. The three-component model performs better than the two-component model, at retrieving total phytoplankton absorption. Accounting for the contribution of pico- and nanoplankton, rather than the combination of both used in the two-component model, improved significantly the retrieval of phytoplankton absorption at low chlorophyll-a concentrations. The three-component model was applied to a decade of ocean colour observations. In the equatorial region of the Pacific and Indian Oceans, phytoplankton size class anomalies (% total chlorophyll-a) were highly correlated with indices of both the El Niño (La Niña) Southern Oscillation and the Indian Ocean Dipole. Furthermore, in these regions, micro- and nanoplankton size class anomalies were negatively correlated with anomalies of the sea surface temperature, sea surface height and stratification. Whereas, the picoplankton size class anomalies were positively correlated with these physical variables. Results from this thesis indicate that phytoplankton size class can be retrieved from Earth Observation with reasonable accuracy. It is recommended that such information can now be assimilated into multi-plankton biogeochemical models, or alternatively, verify them.

A major limitation for remote sensing analyses of oceanographic variables is loss of spatial data. The Data INterpolating Empirical Orthogonal Functions (DINEOF) method has demonstrated effectiveness for filling spatial gaps in remote sensing datasets, making them more easily implemented in further applications. However, dataset reconstructions with this method are sensitive to the characteristics of the input data used. The spatial and temporal coverage of the input imagery can heavily impact the reconstruction outcome, and thus, further metrics derived from these datasets, such as phytoplankton bloom phenology. In this study, the DINEOF method was applied to a three-year time series of MODIS-Aqua chlorophyll-a of the Salish Sea, Canada. Spatial reconstructions were performed on an annual and multi-year basis at daily and week- composite time resolutions, and assessed relative to the original, clouded chla datasets and a set of extracted in situ chla measurements. A sensitivity test was performed to assess stability of the results with variation of cross-validation data and simulated scenarios of lower temporal data coverage. Daily input time series showed greater accuracy reconstructing chla (95.08-97.08% explained variance, RMSExval 1.49 - 1.65 mg m-3) than week-composite counterparts (68.99-76.88% explained variance, RMSExval 1.87 – 2.07 mg m-3), with longer time series of both types producing a better relationship to original chla pixel concentrations (R 0.95 over 0.94, RMSE 1.29 over 1.35 mg m-3, slope 0.88 over 0.84). Original daily chla achieved a better relationship to in situ matchups than DINEOF gap-filled chla, with annual DINEOF-processed data performing better than the multi-year. The results of this study are of interest to those who require spatially continuous satellite-derived products, particularly from short time series, and encourage processing consistency in future DINEOF studies to allow unification for global purposes such as climate change studies (Mélin et al., 2017). / Graduate