The latest dataset can be found at https://doi.org/10.12157/IOCAS.20260223.002 Description GEOXYGEN is a global, four-dimensional ocean dissolved oxygen dataset. It provides monthly DO fields on a 0.5° × 0.5° latitude–longitude grid and 187 standard depth levels from the upper ocean (~1 m) to the deep ocean (~5500 m) for the period 1960–2024. The dataset is intended for studies of ocean deoxygenation, marine biogeochemistry, and climate variability, and for evaluating ocean and Earth system model simulations of dissolved oxygen. Data content and format GEOXYGEN is distributed as CF- and ACDD-compliant NetCDF files. In version v1.0, the data are organised as monthly files named GEOXYGEN_DO_YYYYMM_0p5deg_v1.nc covering the period from January 1960 to June 2024. In total, v1.0 comprises 774 monthly NetCDF files. Each file contains a single monthly field of dissolved oxygen for all depth levels on the global grid. The main variables are: OXY — dissolved oxygen concentration (µmol kg⁻¹), dimensions: time × depth × latitude × longitude. time — monthly time coordinate, dimension: time. Units: days since 1950-01-01T00:00:00Z, Gregorian calendar. depth — standard depth levels (m, positive downward), dimension: depth. 187 levels between ~1 m and 5500 m. latitude — latitude of grid-cell centres (degrees north), dimension: latitude. longitude — longitude of grid-cell centres (degrees east), dimension: longitude. Additional global attributes follow CF-1.8 and ACDD-1.3 conventions and include, for example, dataset title, institution, keywords, spatial and temporal coverage, and processing history. Spatial and temporal coverage The dataset covers the global open ocean on a regular 0.5° × 0.5° grid in latitude and longitude. Land and shallow shelf regions are masked using a 200 m bathymetric/land mask, so that the fields represent offshore, open-ocean conditions. The depth axis extends from the upper ocean (~1 m) to about 5500 m on 187 standard depth levels. The temporal coverage of v1.0 is January 1960 to June 2024 at a monthly resolution. Method GEOXYGEN is generated by combining an extensive compilation of quality-controlled in situ DO profiles with a machine-learning framework. Regional and depth-resolved CatBoost models are trained using physical and biogeochemical predictor fields (e.g. temperature, salinity, oxygen solubility, circulation and productivity proxies, and spatiotemporal coordinates) and then applied to consistent gridded driver fields to reconstruct DO globally from 1960 onward. Contact Email zgwang24@m.fudan.edu.cn