Abstract
Phytoplankton are microscopic, photosynthetic organisms that dwell in the sunlit layers of our planet’s fresh water and oceans. Marine phytoplankton communities are made up of phylogenetic classes, with each of these groups having specific influences on biogeochemical processes such as carbon export and food-web dynamics. Phytoplankton groups are commonly identified by pigment biomarkers, coloured compounds that aid in light harvesting, photosynthesis or photoprotection. This thesis focuses on pigment-based methods to determine the biomass and distribution of phytoplankton groups in the Southern Ocean, and how these tools can be used to explore their spatial and temporal trends.
From the late 20th century many field campaigns have collected pigment samples in the Southern Ocean, generating an extensive dataset (~15, 000 samples) over broad spatial scales but mainly confined to summer months. To use this data in characterising the spatial and temporal variation of different Southern Ocean phytoplankton groups requires a high throughput (i.e., processing many datasets) method to convert the pigment data into estimates of biomass for different groups.
The initial phase of this thesis focuses on the development of a novel inversion method “phytoclass” to convert pigment data into chlorophyll a biomass for the different phytoplankton groups. This conversion involves matrix operations between pigment data and the ratios of pigments to chlorophyll a for different phytoplankton groups. These matrix operations calculate an estimate of the true pigment concentrations, thus providing a metric for error between the estimated and true pigment samples. An optimisation algorithm “simulated annealing” was then applied to approximate the global minimum error between the estimated and true phytoplankton pigments. The phytoclass program was tested on many synthetic datasets and also compared to the similar (and widely used) CHEMTAX program which is currently the “scientific consensus” inversion technique. The phytoclass program showed higher accuracy than CHEMTAX, and also provided additional benefits of simultaneous analysis of multiple datasets.
Patterns in the abundances of phytoplankton groups in the Southern Ocean were determined by applying the phytoclass methodology to a large pigment dataset (n = ~15,000). From this analysis, it was clear that the Antarctic Polar Front (APF) represents a major oceanographic divide in phytoplankton community composition, with higher diatom abundance to the south and greater haptophyte abundance north of the APF. Diverse oceanographic zones (latitudinal bands) south of the APF showed very similar community structures. Using Principal Component Analysis, temperature, sea ice, and nutrient concentrations (macro and micro) were shown to primarily explain the variation in phytoplankton community composition on both sides of the APF. Phytoplankton communities south of the APF showed a strong correspondence to iron and sea ice concentrations consistent with previous observations, whereas communities to the north showed an inverse relationship with iron, and a strong relationship with sea surface temperature.
In recent years, machine learning algorithms have become more accessible. To assess how phytoplankton groups have changed over the course of the Ocean Colour satellite record (1997 – present), random-forest (RF) models were built for different phytoplankton groups based on estimates of their biomass from phytoclass. Environmental and bio-optical characteristics determined from satellites and hydrodynamic models were offered as explanatory data to each model. The RF method proved to be robust at estimating phytoplankton groups with modelled output closely fit to in-situ samples. Based on the RF models, linear trend analysis was carried out for seven key groups: Diatoms, Haptophytes, Cryptophytes, Green algae, Dinoflagellates, Pelagophytes and Synechococcus. The results showed that the seasonalities of phytoplankton groups have changed over the course of the ocean colour satellite record, with winter biomass for many groups showing an increase north of the APF. Furthermore, diatoms showed the largest increase of all the phytoplankton groups, both north and south of the APF. This analysis also characterised the seasonal succession of phytoplankton groups either side of the APF and indicated earlier development of haptophyte blooms in December, south of the APF, followed by diatoms in January and February. Cryptophyte biomass was elevated throughout the time series along the west Antarctic coast, and trends in their biomass were heterogeneous, in contrast to other observations of their biomass increasing at the expense of diatoms.
This thesis details the development of new methodology for analysis of large datasets, and their interpretation to increase understanding of the biomass and distribution of key phytoplankton groups in the Southern Ocean. Application of the methodology, datasets and interpretation will advance assessment of the multifaceted impacts of a warming climate on Southern Ocean phytoplankton and inform forecasting of regional carbon export and ecosystem dynamics.