MIT researchers propose a "physics-enhanced deep-surrogate" (PEDS) method for developing data-driven surrogate models for complex physical systems in such fields as mechanics, optics, thermal transport, fluid dynamics, physical chemistry, and climate modeling.