Ad Hoc Modeling in Agronomy: What Have We Learned in the Last 15 Years?
Impact factor: 1.897 (Year: 2012)
MetadataShow full item record
Timeless limited access
Francois Affholder, Pablo Tittonell, Marc Corbeels, Sebastien Roux, Natacha Motisi, Philippe Tixier, Jacques Wery. (13/3/2012). Ad Hoc Modeling in Agronomy: What Have We Learned in the Last 15 Years. Agronomy Journal, 104 (3), pp. 735-748.
The “Use and Abuse of Crop Simulation Models” special issue of Agronomy Journal published in 1996 ended with the myth of the universal crop model. Sinclair and Seligman consequently recommended tailoring models to specific problems. This paper reviews the fate of the idea of such ad hoc approaches to crop simulation modeling during the past 15 yr. Most crop modelers have since adhered to the principles formulated by Sinclair and Seligman, but yet their practice faces two major issues: (i) how to define the structure of the model as depending on the question to be addressed (model conceptualization) and (ii) how to minimize efforts in software development (model computerization). Progress in model conceptualization as reported in the literature concerns (i) inferring a conceptual model from what is known of the problem to address, (ii) deriving summary models from comprehensive ones, and (iii) using multivariate methods to analyze the hierarchy of drivers of variability in the variable to be predicted. Considerable effort has been invested in the development of frameworks to facilitate model computerization, and the commercial modeling software is constantly improving. But there are limits in the flexibility permitted by these tools. Acquiring basic skills in coding a model using a scientific programming language is preferred by scientists wishing to keep the fullest understanding and control on their crop models. Connecting the model to commercial database software may facilitate this strategy. However, the computerization issue may still lead to tensions between modeling teams concerning the legitimacy to develop their own model.