Estimating marine carbon sink in an Arctic shelf Sea using machine learning and remotely sensed data

Accurately quantifying carbon sinks and sources in coastal regions is essential for global carbon budgeting, but it can be a challenging task. This is especially true for rapidly changing sub-Arctic and Arctic seas where baseline observations of seawater CO₂ partial pressure (pCO₂) are limited. Hudson Bay, Canada, is a prime example of an area with sparse data geographically and temporally. To bridge this gap, we utilized a novel approach by integrating predictor variables from satellite imagery and reanalysis data with advanced machine learning algorithms to provide more precise regional estimates of pCO₂. In addition, we examined the ocean's carbon uptake and spatiotemporal fluctuations over different periods by incorporating wind speed and atmospheric CO₂ data. Our study not only reveals insights into the dynamics of Hudson Bay CO₂ sources and sinks but also demonstrates the potential of machine learning in extrapolating ship observations over space and time.