Nitrous oxide (N₂O) is a potent greenhouse gas and the dominant ozone-depleting pollutant of the 21^st century. The major sources of N₂O to the atmosphere are agriculture, fossil fuel combustion and industry, natural terrestrial soils, and the ocean; out of all of these fluxes, the marine source represents roughly 20%. Oceanic N₂O emissions, however, remain poorly constrained, due to their heterogeneity and the sparseness of marine N₂O observations. Here, we pair newly available, high-quality datasets with machine learning to resolve the shifting natural baseline of N₂O fluxes between ocean and atmosphere. Using shipboard observations from the Global Ocean Ship-Based Hydrography Program, we train supervised learning models to predict N₂O from temperature, salinity, oxygen, and nitrate. We then apply those models to Biogeochemical Argo-based gridded data products to reconstruct N₂O in the ocean at the global scale, from 0-2000 m depth, over the last 20 years. Based on this reconstruction, we present a monthly marine N₂O climatology, quantify the global marine N₂O source, and examine changing inventories over time. We then examine the effects of different scales of variability on N₂O emissions from oxygen minimum zones.