This paper describes calibration methods for models of agricultural production and water use in which economic variables can directly interact with hydrologic network models or other biophysical system models. We also describe and demonstrate the use of systematic calibration checks at different stages for efficient debugging of models. The central model is the California Statewide Agricultural Production Model (SWAP), a Positive Mathematical Programming (PMP) model of California irrigated agriculture. We outline the six step calibration procedure and demonstrate the model with an empirical policy analysis. Two new techniques are included compared with most previous PMP-based models: exponential PMP cost functions and Constant Elasticity of Substitution (CES) regional production functions. We then demonstrate the use of this type of disaggregated production model for policy analysis by evaluating potential water transfers under drought conditions. The analysis links regional production functions with a water supply network. The results show that a more flexible water market allocation can reduce revenue losses from drought up to 30%. These results highlight the potential of self-calibrated models in policy analysis. While the empirical application is for a California agricultural and environmental water system, the approach is general and applicable to many other situations and locations.