Globally there are over 24,000 dams and reservoirs that store over 7,000km3 of water, yet most of the data regarding reservoir operations is not openly accessible. As a result, many studies rely on generalized assumptions about reservoir storage dynamics and management to derive reservoir operations. To date, there has been no global assessment of modelled reservoir operations derived from openly accessible reservoir data (i.e. a data derived method). Our work focuses on the development and testing of a framework to derive reservoir operations using global satellite data of reservoir storage, a statistical model of operational bounds and a machine learning algorithm to extrapolate the operational bounds (i.e. flood and conservation levels) to data scarce reservoirs. From this, we observe that the data derived method accurately captures reservoir operational bounds and results in much lower storage levels. These lower storage levels align better with reservoir storage observations and suggest that the generic observations currently used in large scale models could overestimate regional water availability by overestimating reservoir storage.