Abstract
The Southern Ocean represents less than 20% of the world ocean area, yet it plays a disproportionate role in regulating Earth’s climate through its control on heat transport, air-sea CO2 exchange and marine primary productivity. The latter is driven by phytoplankton, yet their long-term changes and their physical drivers remain poorly constrained in the Southern Ocean due to limited observations, strong regional heterogeneity, and large uncertainties in Earth System Models projections. This thesis addresses this challenge by integrating multi-decadal satellite observations, statistical analyses, deep learning reconstructions, and Earth System Model (ESM) projections to provide a comprehensive assessment of phytoplankton variability and change in the Southern Ocean.
Using 26 years (1998-2023) of satellite-derived Chlorophyll-a (Chl-a), this study first quantifies seasonal and interannual variability across the frontal systems that structure the Southern Ocean into distinct oceanographic regimes. Summer Chl-a concentrations are nearly twice larger than winter values, and a coherent 5-6-year oscillation emerges across much of the basin. Approximately 60% of the Southern Ocean shows increasing Chl-a over the satellite era, particularly in coastal and southern boundary regions. Among the large-scale environmental drivers examined, sea surface temperature emerges as the dominant control, while mixed layer depth, wind speed, and light availability exert regionally dependent influences.
During the analysis of long-term Chl-a trends, however, it became evident that a substantial portion of the observed variability could not be explained solely by these physical drivers. The temporal evolution of Chl-a displayed a striking correspondence with changes in sea ice extent. Although sea ice was not originally a primary objective of this thesis, this emerging pattern motivated a focused investigation into its role as a modulator of phytoplankton variability.
Explicit examination of sea ice dynamics reveals that Antarctic summer sea ice extent has declined at a rate several times faster than winter sea ice, marking a pronounced post-2015 regime shift. Positive sea ice extent anomalies are found to precondition the upper ocean for enhanced phytoplankton growth in the following year, particularly in coastal and southern boundary regions, likely through meltwater-induced stabilization. Conversely, recent sea ice losses are associated with reduced Chl-a and a transition toward a low-ice, low-productivity regime. These findings demonstrate that sea ice variability exerts a strong, spatially structured, and lagged influence on phytoplankton dynamics, complementing and, in some regions, amplifying the earlier-identified physical drivers.
Having established the relationships between Chl-a and its key environmental controls, the thesis then advances beyond correlation-based analyses by introducing a deep learning framework to better resolve the coupled physical-biological dynamics. A U-Net convolutional neural network is developed to generate Southern Ocean Chl-a fields using satellite and reanalysis datasets. The Chl fields produced by this model successfully match observed spatial patterns, seasonal cycles, and interannual variability, with particularly strong performance in the Pacific sector. Explainability analyses identify sea surface temperature and sea level anomaly as the most influential predictors, underscoring the roles of stratification and mesoscale circulation in regulating phytoplankton biomass. This approach not only improves spatial continuity in data-sparse regions but also provides a physically interpretable bridge between environmental forcing and biological response.
Building on this framework, the trained neural network is subsequently fed with future projections of physical parameters from four CMIP6 Earth System Models under low-, intermediate-, and high-emission scenarios. In contrast to the divergent and often contradictory responses of these ESMs in terms of chlorophyll, the observation-constrained Chl-a projections reveal consistent basin-wide declines in the Southern Ocean under all scenarios, with the magnitude of decline intensifying under higher emissions. Under the high-emission scenario, more than 90% of the basin exhibits negative trends by the end of the century. These projected declines are primarily driven by warming-induced stratification, shoaling mixed layers, reduced nutrient supply, and altered light conditions.
Taken together, these findings provide a unified, multi-scale perspective on Southern Ocean phytoplankton variability and change. By combining long-term observations, targeted analysis of sea ice dynamics, machine learning reconstruction, and climate projections, this thesis advances understanding of the physical controls governing phytoplankton ecosystems and reduces uncertainty in future projections. The results highlight the critical role of sea ice and frontal systems in shaping biological productivity and demonstrate that future Southern Ocean ecosystems are highly sensitive to emission pathways. Sustained monitoring and climate mitigation are therefore essential to preserve the productivity and carbon-sequestration capacity of this climatically pivotal region.