The provided dataset includes near-bottom oxygen measurements (observations deeper than 10 m) in the months from August to October in the period from 1902-1969 from the western Baltic Sea (57.5 - 53.5 N° and 9.2- 15 E°). It serves as basis to predict historic oxygen deficiency areas in the period from 1949-1969 with a generalized additive model (GAM) and a machine learning model (GPBoost) as presented by Piehl et al. (2025). These data originate from ICES, including coastal data from national sources (Leibniz Institute for Baltic Sea Research Warnemünde, (IOW), Danish EPA (Miljøstyrelsen), and Swedish Meteorological and Hydrological Institute (SMHI)). The provided nc-files include the multi-annual averaged (1949-1969) modelled near-bottom oxygen concentrations averaged over the critical period from August to October by three different approaches, the generalized additive model (GAM), the biogeochemical model (ERGOM), and the machine learning model (GPBoost). The nc-files were produced using the software R (https://www.R-project.org/) for the GAM and GPBoost results and CDO (https://doi.org/10.5281/zenodo.10020800) for the ERGOM results.