AbstractLakes and reservoirs provide multiple essential ecosystem services to humans, and several of these require the presence of a cool, oxygen‐rich hypolimnion. These ecosystem services include supporting habitat for recreational fish species and the maintenance of a non‐eutrophic state by limiting internal nutrient loading. However, changes in land use and global warming are modifying the thermal structure of lakes worldwide, creating episodic or prolonged periods of hypoxia, changing a lakes ability to deliver certain ecosystem services. Here, we used National Lake Assessment data and machine learning approaches to identify which lake or watershed features determined the presence of a hypolimnion and which contributed to the development of hypoxia. A random forest of 1000 trees predicted the presence of a hypolimnion with 85% accuracy using commonly measured morphometric and physicochemical variables as well as land use. Mean and maximum depths had a disproportionate influence, entirely obscuring the influence of variables commonly associated with lake mixing, such as light penetration and fetch. In contrast, depth had a more restrained role to predict the presence of hypoxia; the latter was predicted with an overall accuracy of 85%, from epilimnetic nitrogen and carbon concentrations. Although the nutrient‐color groups had a minor influence in the random forests, there was a clear trend of increased hypoxia with higher turbidity. Our approach allows for the broad‐scale assessment of lakes with hypolimnia and suggests that increasing eutrophication and browning will promote profundal hypoxia, altering the delivery of certain ecosystems services.