Imperfect knowledge and data-based approach to model a complex agronomic feature – Application to vine vigor

Vine vigor, a key agronomic parameter, depends on environmental factors but also on agricultural practices. The goal of this paper is to model vine vigor level according to the most influential variables. The approach was based upon a collected dataset in a French vineyard in the middle Loire valley and the available expert knowledge. The input features were related to soil, rootstock and inter-crop management, the output was an expert assessment of vine plot vigor. The approach included a data selection step, which was needed because of data imperfection and incompleteness. Usually implicit in the literature, data selection was carried out with explicit criteria. Then a fuzzy model was designed from the selected data. Owing to the fuzzy model interpretability, its structure and behavior were analyzed. Results showed that, despite the data imperfection, the approach was able to select data that yielded an informative model. Well-known relationships were identified, and some elements of new or controversial knowledge were discussed. This is an important step towards the design of a decision support tool aiming to adapt the agricultural practices to the environment in order to get a given vigor level.