The expansion of tropical oxygen minimum zones (OMZs) is a growing concern as ocean deoxygenation intensifies due to climate change. However, accurately estimating dissolved oxygen (O₂) concentrations and tracking OMZ expansion is challenging due to sparse and irregular sampling. This study employs machine learning techniques to generate gridded four-dimensional (space and time) maps of O₂ distributions, focusing on the tropical Atlantic OMZ.  By combining historical shipboard and Argo observations, neural networks and random forest models are applied to reconstruct oxygen trends from 1980 onwards, using physical oceanographic reanalysis datasets as predictor variables. By incorporating reanalysis data at coarse (EN4, ORAS4, 1°x1°), medium (SODA3, 0.5°x0.5°), and fine (ORAS5, 0.25°x0.25°) resolutions, this study aims to assess how differences in spatial resolution affect the reconstruction of ocean deoxygenation and OMZ expansion trends. The study specifically investigates how finer spatial resolution data can improve the representation of OMZ boundaries, particularly in the regions where observational O₂ data is sparse. The models are trained on 80% of the data, with 20% withheld for validation, employing a range of hyperparameters to optimize performance. The application of finer resolution reanalyses better represents water mass boundaries and the tropical zonal jets, which may reduce the uncertainties in the reconstruction of O₂ variability despite sparse observations in tropics, especially when integrated with Argo profiles.  The potential of high-resolution reanalysis data is explored to constrain the range of deoxygenation trends and to interpret its underlying causes by utilizing the circulation fields from the reanalysis. This approach may offer new insights into how ocean reanalyses with different horizontal resolutions can be combined with biogeochemical observations through the use of machine learning algorithms. The resulting gridded O₂ dataset can be used to assess the deoxygenation trends and the expansion of the tropical OMZs and provide a reference dataset for the validation and calibration of the earth system models.

tropical oxygen minimum zones (OMZs), ocean deoxygenation, machine learning, climate change