Using machine learning techniques and remotely sensed reservoir data, we develop a workflow to derive reservoir storage bounds. We put these bounds in a global hydrologic model, PCR-GLOBWB 2, and evaluate the difference between generalized operations (the schemes typically in global models) and this data derived method. We find that modelled storage is more accurate in the data derived operations. We also find that generalized operations over estimate storage and can underestimate water gaps.