In traditional reservoir management, various types of predictive models have been applied over the years for either qualitative or quantitative optimization of various reservoir management decisions. Such models range from the very simple analytical models (type-curves, etc.) to the very complex reservoir simulation models. While analytical models are too simplistic for quantitative optimization, many issues such as the significant time and effort required to build and calibrate simulation models etc. generally prohibit their practical use for closed-loop quantitative optimization. Additionally, there have also been many attempts at the application of traditional machine learning approaches for predictive modeling of production performance. While such models can be built very efficiently and are very fast to evaluate, however, due to spatial sparsity of data, combined with poor measurement quality, and the absence of the underlying physics in such models, such purely data-driven approaches have only had limited success. This talk describes a unique modeling approach termed Data Physics. Data Physics combines state-of-theart in machine learning approaches and reservoir physics into unified models. These models can be created as efficiently as machine learning models, integrate all kinds of data, and can be evaluated orders of magnitude faster than full scale simulation models, and since they honor physics, they have good long term predictive capacity and can therefore be used for robust large scale optimization. We present applications of Data Physics models to real waterflood injection and infill drilling optimizations. A significant increase in actual incremental oil production and reduction in operational cost is demonstrated.