Modelling the performance of annual intercrops: an approach coupling functional ecology and data science

I studied at INSA in Rennes, with a major in Mathematical Engineering. I discovered agronomy during my Master 2 internship in 2018, and this field appealed to me greatly. I have long wanted to do a thesis and the opportunity arose during my M2 internship to continue the work I was doing at the time. I see this 3-year thesis as an opportunity to delve into this subject in greater detail, and to gain a better understanding of the workings of associated crops, which are one of the levers of sustainable agriculture.

Combined crops (two or more crops growing on the same agricultural plot over a significant period of their growth cycle) are a major asset for the development of sustainable agriculture. My thesis focuses on modeling the performance of this type of crop. The aim is to design statistical models to predict the performance (yield) of annual associated crops. To do this, I have a database in which functional trait values (height, biomass etc.) are measured on species (pea, wheat, lentil etc.) in different environments (sites, years). The idea is to calculate metrics that account for species complementarity and others for plasticity. Indeed, theories of community ecology indicate that the processes of complementarity and plasticity are strongly involved in the functioning of mixed cover crops. Based on these indicators and pedo-climatic variables (soil and climate), I will be able to develop these statistical models. The aim is to be able to determine which species and which varieties work well together, depending on the pedo-climate under consideration.

My subject is a natural follow-up to my Master 2 internship. It suits me because it's multidisciplinary (ecology, agronomy and statistics). It's part of the search for effective, sustainable solutions for agriculture, which is facing a number of challenges.

Since the end of my Master's internship, the database I had been working with has been greatly enriched by my supervisors. The first thing I had to do was to take it back in hand and carry out a complete descriptive analysis. At the end of my Master's 2 internship, I observed a quasi-linear relationship between wheat yield in associated cropping and the difference in biomass growth rates (wheat biomass growth rate - pea biomass growth rate). I'd like to see if this result can be found for other species.

• Starting date: November 2019
• University: University of Montpellier
• PhD school:  I2S –  Information Structures Systèmes
• Scientific field: Biostatistics
• Thesis management: Nadine Hilgert, UMR Mistea, Inrae
• Thesis supervisors: Pierre Casadebaig & Noémie Gaudio, UMR AGIR, Inrae
• Funding: #DigitAg – Inrae
• #DigitAg : Cofunded PhD – Axes 5, 6 – Challenge 1

Keywords: Combined cropping, statistics, agroecology

Abstract: In order to contextualise this research, this doctoral research project has focused on the potential of a time-series of Sentinel-2 satellite images for monitoring vineyards at the regional scale across the Occitanie region (France). This spatio-temporal Sentinel-2 dataset presents unique characteristics in terms of revisit time, spatial resolution, attribute information provided and cost. Moreover, the choice of spatial coverage is interesting in itself, as the Languedoc-Roussillon wine region represents a great diversity of agri-environmental conditions resulting in a large number of different grape varieties being cultivated as well as a large diversity in the management practices of the wine growers. Collectively these factors introduce additional levels of variability into the analysis of regional-scale viticulture data. This PhD work is based on the assumption that assessing the temporal variability in the satellite imagery, in addition to spectral variations, would allow a more complete analysis to derive new and relevant information about production variability of individual vineyards at the regional scale. With this in mind, the principal objective of this thesis is to integrate temporal analyses, as an additional descriptor of vineyard variability, in order to take into account, in a better and more holistic way, all the specific dimensions of remote sensing data (spectral, temporal and spatial). Different supervised and unsupervised multi-way analysis methods, derived from the field of chemometrics, were used, capable of generating information at the regional scale from time series of multispectral images. Unsupervised approaches demonstrated the possibility of extracting agronomic knowledge over time (e.g. different vegetative dynamics) without a priori knowledge. The supervised methods allowed, firstly, the spectral, temporal and spatial assessment of an extreme climatic event (e.g. a heat wave) and, secondly, the selection of multidimensional (spectral-temporal) variables to deepen the agronomic understanding of the impact of an extreme climatic event on grapevines at a regional scale.

Jury compound:

• Muriel Valentin-Morison (UMR Agronomie INRAE) - Rapportrice
• Nathalie Vialaneix (UMR MIAT INRAE) - Rapportrice
• Catherine Trottier (UMR IMAG CNRS) - Examinatrice
• Eric Garnier (UMR CEFE CNRS) - Examinateur
• Xavier Gendre (Institut Mathématiques de Toulouse CNRS) - Examinateur
• Pierre Casadebaig​ (UMR AGIR INRAE) - Encadrant de thèse
• Noémie Gaudio (UMR AGIR INRAE) - Encadrante de thèse
• Nadine Hilgert (UMR MISTEA INRAE) - Directrice de thèse