Simulation Models: Science, Snake Oil, Education, or Engineering?

Abstract Crop simulation models can be divided into two groups: those that aspire to improve our understanding of the physiology and environmental interactions of crops ( science ), and those that aspire to provide sound management advice to farmers or sound predictions to policy makers ( engineering ). These quite different aspirations require quite different models. Scientific models are mechanistic. With a few exceptions, they have failed to meet their aspirations. They are typically flawed by being based on untestable guesses about the processes that control growth. They may, however, provide useful self‐education for their developers. The best engineering models are based on robust empirical relations between plant behavior and the main environmental variables. Because of their empirical nature, we should not expect them to apply outside the range of the environmental variables used in their calibration. Within their calibrated ranges, however, some have proved useful in providing sound management advice. It is hard to see a useful role, other than self‐education, for models that fall between the scientific and the engineering types.