Submitted to: Soil Use and Management
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: May 23, 2012
Publication Date: March 12, 2013
Citation: Vadas, P.A., Bolster, C.H., Good, L.W. 2013. Critical evaluation of models used to study agricultural phosphorus and water quality. Soil Use and Management. 29:36-44. Interpretive Summary: This article uses recent research examples to evaluate potential problems when developing models to simulate agricultural phosphorus in the environment. Models are valuable research tools, but they need to be developed properly so they can make accurate predictions. First, a scientist must properly understand phosphorus cycling and movement to be able to develop an accurate model. However, a proper understanding does not guarantee that the equations used will give accurate predictions. Translating equations into computer code can also create model errors. Models must be structured to allow for validation with field data or there is little confidence they can predict phosphorus movement. It can be difficult to know why a model does not validate and when it should be changed or even rejected. Model development is often limited by data availability, but also by poor communication between field scientists and model developers. Currently, scientists may be concentrating too much on generating research data and not enough on using data to develop models.
Technical Abstract: This article uses examples from recent research to evaluate select issues related to developing models that simulate phosphorus in the environment. Models are valuable because they force scientists to formalize understanding of P systems and identify knowledge gaps, and allow quantification of P transfer when monitoring is not possible. Model development should follow a procedure from a perceptual model of qualitative understanding to a conceptual model of equations to a procedural model of code. There are few implications for a flawed perceptual model when publishing scientific articles, but an inaccurate perceptual model stands little chance of predicting reliably. Because many different model structures are possible, it is possible for a model to be perceptually correct but conceptually incorrect. Translation of conceptual equations into procedural model code can also create model errors. Many P models share the same equations, many of which may be 30 years old and should be updated to reflect current science. Furthermore, models may be applied in situations for which their algorithms are inadequate. Models must be structured to allow for validation with field data or there is little confidence they can quantify P transfer. It can be a challenge to understand why a model does not validate and know when a model should be rejected. Model development and validation are often limited by data availability, but also by poor communication between field scientists and model developers. There may currently be too great an emphasis on data generation and too little emphasis on model development.