Analysis of surface deoxygenation trends in the Benguela upwelling system: An interpretable machine learning approach

This is the authors' abstract. We don't add key points for this paper.

The Benguela upwelling system (BUS) exhibits strong seasonal variability in dissolved oxygen ([O 2 ]), driven by upwelling and remotely advected water masses, and primary productivity. In contrast, its interannual variability remains poorly understood due to limited long‐term observations. In this study, we applied an interpretable machine learning (ML) approach to investigate the interannual variability of surface [O 2 ] in the BUS, highlighting a deoxygenation trend. This trend is driven by widespread warming which reduces [O 2 ] solubility in the adjacent South Atlantic basin, along with interannual variability in upwelling water masses and declining coastal primary productivity. Conversely, an oxygenation trend is observed off Lüderitz, attributed to perennial upwelling that sustains primary productivity throughout the year. The deoxygenation trend in the BUS is significant and may intensify low‐[O 2 ] events, highlighting the need for improved monitoring and assessment of impacts on marine ecosystems.