Predicting historical oxygen deficiency areas in the western Baltic Sea: A multi-model approach

The oceans are under increasing pressure from human activities. In order to manage these, target values describing good environmental status are often determined based on a historical state without significant human impact. Deoxygenation is one of the most harmful indirect effects of human induced coastal eutrophication, and it is important to quantify oxygen deficiency in bottom waters. However, historical oxygen measurements are scarce and models are needed for integrating these point observations. We used a combined approach integrating the results of three models - a statistical, mechanistic, and machine learning model - to estimate the area affected by oxygen deficiency from 1949 to 1969 in the western Baltic Sea. Hypoxia (oxygen <2 mg/l) was rare with <1 % to 3 % of the near-bottom area being predicted as hypoxic with a high to low confidence, respectively. But all sub-basins exhibited oxygen deficiency (oxygen <6 mg/l) with a share of 11 % (high confidence) to 37 % (low confidence) of the whole area. The statistical model predicted the smallest areas of oxygen deficiency, while the mechanistic model predicted the largest. A major uncertainty identified relates to the estimation of oxygen concentrations in the bottom boundary layer, as it is considered differently by our three methods and observational data for model tuning in this layer are lacking. Our multi-model approach provides an estimate of the historical extent of oxygen deficiency with higher confidence than any of the three approaches alone. This estimate can help to define targets for ecological indicators of oxygen deficiency.