Estimating the export of biomass across the shelf break via AI applied to satellite images

Ocean carbon uptake has an important role in the global carbon cycle, absorbing one third of anthropogenic atmospheric carbon. The biological carbon pump, powered by phytoplankton primary production, transports organic carbon from surface waters to deeper ocean zones, effectively sequestering it. Shelf regions, characterized by shallow waters and nutrient abundance, exhibit elevated productivity compared to the open ocean. Submesoscale filaments, or streamers, form through the instability of coastal currents, concentrating phytoplankton and chlorophyll into elongated structures that enhance long-distance transport. Streamers can transport cold, chlorophyll-rich shelf waters into the open ocean, potentially enhancing carbon fluxes to deeper layers. This thesis presents an automated image analysis tool utilizing K-means clustering on an 18-year dataset of chlorophyll-\textit{a} and sea surface temperature anomalies from Level 3 satellite data. Three coastal regions (California, Mauritania, and South China Sea) are studied. The algorithm effectively identifies streamers within a single cluster, enabling estimation of chlorophyll content, lateral export, and carbon fluxes. The analysis of the chlorophyll content time series reveals distinct patterns. While the biomass export intensity in upwelling regions follows the primary production seasonality, the South China Sea shows high-frequency variabilities due to purely turbulent processes. The annual lateral biomass export through streamers totals 3.5 ± 0.7 Tg / yr. Carbon fluxes vary between regions, with upwelling areas having larger fluxes compared to the South China Sea, indicating that streamers have different areas and carbon transport capacities. This study offers valuable insights into lateral carbon export from coastal to open ocean regions. Extending this research to a global estimate is essential to understand the role of this transport in the processes of carbon burial to depth.