Satellite-Based Global Sea Surface Oxygen Mapping and Interpretation with Spatiotemporal Machine Learning.

Area of Science:

• Oceanography
• Environmental Science
• Machine Learning

Background:

• Accurate assessment of sea surface dissolved oxygen (DO) is crucial for understanding ocean processes and environmental challenges.
• Existing data and simulations for global sea surface DO are limited and imprecise, hindering research.

Purpose of the Study:

• To develop a novel machine-learning framework for high-resolution estimation of sea surface DO.
• To provide insights into the driving mechanisms of DO concentration and its environmental implications.

Main Methods:

• Integration of extensive in situ and high-resolution satellite data.
• Development of a spatiotemporal information embedding machine-learning framework.
• Generation of high-resolution (0.25° × 0.25°) DO concentration estimates.

Main Results:

• Achieved exceptional accuracy (R² = 0.95, RMSE = 11.95 μmol/kg) for near-global sea surface DO from 2010-2018.
• Estimated uncertainty at ±13.02 μmol/kg.
• Revealed precise spatial distribution and correlations with marine phenomena and environmental stressors.

Conclusions:

• The machine-learning framework provides an improved, high-resolution DO dataset.
• The model's interpretability identified key marine factors influencing surface DO.
• Facilitates exploration of oceanic DO variability, deoxygenation, and environmental consequences.