Littoral marine ecosystems are predicted to play a critical role in the ocean’s response to global climate change due to their generally high levels of primary production. However, the net trophic status of these systems has not been characterized adequately due to limited long-term time-series data of both autotrophic and heterotrophic activity. Although optical dissolved oxygen sensors are widely used and provide insight into many biological processes, dissolved oxygen measurements alone do not distinguish oxygen cycled via biological processes from oxygen cycled via physical processes. We leveraged gradient-boosted regression trees, long-term environmental data, and limited observations of dissolved oxygen and argon, measured via membrane inlet mass spectrometry, to decouple physical and biological oxygen cycling in a 6-year time series of optical dissolved oxygen in the littoral ecosystem of the Southern California Bight. Our results show that this littoral ecosystem is net heterotrophic, albeit with strong autotrophic and heterotrophic states observed. This study demonstrates that machine learning-based modeling can improve estimates of biologically cycled oxygen and net trophic status with some seasonal variability in model performance, providing a framework for estimating net trophic status in dynamic, productive ecosystems.
Hale et al. (Thu,) studied this question.