Prediction of vine vigor and precocity using data and knowledge-based fuzzy inference systems

Aims: The evolution of the economic and environmental context (low-input management practices, increase of energy cost and climate change) requires adaptation and/or optimization of winegrower’s practices in order to elaborate competitive and yet still qualitative wines. To adapt and sustain their practices at the plot scale (e.g., rootstock selection or plantation density), winegrowers and viticultural consultants need indicators to predict vine development based on permanent environmental factors (soil, parent rock and landscape). As of today, such indicators are either nonexistent or too basic. The aim of this work is to develop operational and useful indicators based on strong scientific evidence.Methods and results: This paper proposes a new approach based on a computer model composed of a cascade of fuzzy expert systems to estimate the two variables that best characterize vine development: vigor (VIG) and precocity (PRE). This model combines pedological expertise and data analysis. Based on scientific literature, and in particular on a previous expert system using analytical equations (Morlat et al., 2001), the new approach allows a continuous estimation of VIG and PRE imparted by soil, parent rock and landscape. Further, it avoids the drawbacks of the previous expert system, due to the use of traditional crisp partitions for continuous input variables. Another novel aspect is the parameter setting, which efficiently combines expert knowledge and data mining. Finally, the method is tuned and validated against two different databases.Conclusion: VIG and PRE imparted by environmental factors can now be evaluated more efficiently than with the former methods. The new method eliminates the need for post-evaluation correction by experts, which saves time. It also allows a continuous estimation of these variables. Each step can be controlled and analyzed during the design. Finally, the method is generic in the sense that the reasoning used to represent the relations between variables is not restricted to a given area. It can easily be customized and adapted to new areas by adjusting the parameters using local pedological knowledge and data.Significance and impact of the study: This work answers the significant problem of VIG and PRE assessment according to environmental factors, which is a prerequisite in order to best adapt long-term cultural practices.