Simulating dissolved oxygen in the oceans – perspectives for the Baltic Sea

An increasing number of dead zones characterised by toxically-low levels of dissolved oxygen has been reported in coastal oceans all over the globe. Efforts towards respective quantitative descriptions are ongoing but numerical simulations and predictions of hypoxia remain challenging. In this study, we present a suite of generic approaches towards more reliable simulations.

Along a test-case we showcase the coalescence of a suite of ultra-high (~ 100m horizontal) resolution general ocean circulation model of Eckernförde Bight (Baltic Sea) with machine learning approaches. The ocean model includes an elementary representation of the biogeochemical dynamics of dissolved oxygen. In addition, we integrate artificial “clocks” that measure the residence time of the water in Eckernförde Bight and the timescales of (surface) ventilation. Our approach starts with an ensemble of hindcast model simulations (covering the period from 2000 to 2018) designed to envelop a range of poorly known model parameters for vertical background mixing (diffusivity) and local oxygen consumption within Eckernförde Bight. In a subsequent step, feed-forward artificial neural networks trained with output from the model ensemble are put to work to identify predictors of hypoxia deep in Eckernförde Bight based on data at a monitoring site at the entrance of the bight. Our approach disentangles the relative importance of subduction and vertical mixing versus local oxygen consumption and the inflow of low-oxygenated waters from the Kiel Bight.