Turbulence modeling can shed light on the physics that occurs at the boundary between ocean waves and overhead wind, which is useful in the fields of oceanography, renewable energy, weather forecasting, and climate modeling. Previous research has leveraged deep learning techniques with classical turbulence models to enhance their granularity and decrease their computational cost. However, the physical credibility of these deep learning techniques is not guaranteed; they are merely used to identify prediction optimums in unknown conditions, and do not necessarily reinforce the known physical properties of the system in doing so. This project develops a physically credible mechanism for utilizing deep learning in the context of the wind-wave problem. In particular, we explored the feasibility of using a physics-constrained neural network (PCNN) to predict key parameters that can be exploited by turbulence models while also adhering to known physical truths. Our demonstrated PCNN can reliably predict the drag force that the waves exert on the overhead wind while capturing the notions of total work and total power of the system. In a broader sense, this project illustrates trade-offs between training a model to obey physical truths and training it to perform well on a particular objective.