[Defended thesis] Cécile Laurent: Proposal of a grape yield elaboration and forecast model through learning process from heterogeneous data

Proposal of a grape yield elaboration and forecast model through learning process from heterogeneous data such as time series of observations and historical climatic data

• Starting date: September 2018
• University:  MUSE Montpellier Université d’Excellence – Institut Agro
• Scientific field: Agronomy/viticulture, data analysis, statistics
• Thesis management: Bruno Tisseyre, Institut Agro-Montpellier SupAgro, Itap & Aurélie Metay, Institut Agro-Montpellier SupAgro, AbSyS
• Thesis supervisors: Meïli Baragatti, Institut Agro, Mistea & James Taylor, Inrae, Itap
• Funding:  Cifre Fruition Sciences
• #DigitAg : Labeled PhD – Axes 6 & 5 – Challenges : sujet transverse, 1 & 5

Keywords: Vine, Wine, Estimation, Prediction, Time series of weather data, Yield elaboration, Bayesian

Abstract:  Understanding the key steps of grape yield elaboration and estimating at the intra-plot level are two important issues for the wine industry. The development of precision viticulture is leading to major changes in terms of data generation, thus making it possible to to explore new operational approaches. In this context, the PhD project proposes to analyse heterogeneous data such as time series of observations and historical data in order to : i) infer knowledge relating to yield elaboration mechanisms specific to the local context since they are determined based on the data, ii) identify the factors which can locally affect the yield throughout the vine cycle, iii) foresee the learning of empirical models which are locally adapted and allow yield estimation throughout the production cycle. The PhD project therefore focuses on two main issues : how to infer knowledge about grape yield elaboration from heterogeneous data ? How can this knowledge be used to develop an empirical yield forecasting model ? Although it does not exclude any approach at the moment, the thesis proposes to use new methods under development such as the Bliss method (Grollemund, 2017) to answer the first question. These are Bayesian functional statistical methods that allow the analysis of the global history of variables evolving over time. Regarding the second issue, the PhD will focus on fuzzy rules logic and inference (Grelier et al., 2007) to process heterogeneous data while taking their imprecision and complex interactions into account. This approach brings forward important potentialities for the agriculture sector. Indeed, the analysis of the relationships between dated quantitative variables and multi-variate time series is very often necessary to estimate the quality and quantity of production (yield, sugar content, acidity, grain protein content, etc.) as a function of time variables (temperature, water status, radiation, soil moisture, etc.). Grollemund P., 2017 Régression linéaire bayésienne sur données fonctionnelles. Université de Montpellier. Grelier M., Guillaume S., Tisseyre B., Scholasch T., 2007. Precision viticulture data analysis using fuzzy inference systems. Journal International des Sciences de la Vigne et du Vin 41 (1), 19-31

Contact :  
bruno.tisseyre [AT] supagro.fr
james.taylor [AT] inrae.fr

Social networks: ResearchGate – LinkedIn

Papers:

Download the thesis manuscript

Acts of conference

• C. Laurent, M. Baragatti, J. Taylor, B. Tisseyre, A. Metay, T. Scholasch, Data mining approaches for time series data analysis in viticulture. Potential of the BLiSS (Bayesian Functional Linear Regression with Sparse Step functions) method to identify temperature effects on yield potential, Accepted to the 21st Group of international Experts for Cooperation on Viti-vinicultural Systems INternational Meeting (GiESCO 2019) , Aristotle UNiversity of Thessaloniki, Thessaloniki, Greece

• C. Laurent, M. Baragatti, J. Taylor, T. Scholasch, A. Metay, B. Tisseyre, Evaluation of a functional Bayesian method to analyse time series data in precision viticulture, 12th European Conference on Precision Agriculture (ECPA 2019), Montpellier SupAgro, Montpellier, France

• C. Laurent, T. Scholasch, B. Tisseyre, A. Metay, Building New Temperature Indices for a local understanding of grapevine physiology, XIIIth International Terroir Congress, Virtual Congress, Adelaide, Australia