PhD positions 2019

Every year, #DigitAg launches a PhD campaign and supports new theses, from blue skies to applied research. Discover all our Autumn 2019 PhD opportunities!

PhD positions – Campaign 2019

Applied Mathematics – Life & Environmental Sciences

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

Keywords: Agroecology, intercrops, data science, modelling, data mining, machine learning, phenotyping

  • Contact: Noémie Gaudio – noemie.gaudio [AT] inra.fr
  • PhD director(s): Nadine Hilgert, Inra MISTEA,
  • Supervisor(s): Nadine Hilgert, Inra MISTEA / Pierre Casadebaig, Inra AGIR / Noémie Gaudio, Inra AGIR
  • Research Units: AGIR (Inra Toulouse), MISTEA (Inra Montpellier)
  • Co-Funding:  Inra
  • #DigitAg: Axis 6 : Multiscale modelling and simulation, Challenge 1 : Agreécology, Axis 5:Data Mining, Data Analysis and Knowledge Discovery

Increasing plant diversity in agriculture is suggested as one of the main mechanisms to move towards more sustainable production systems. Plant diversity enhances and stabilizes primary production through complementarity between plants. Currently, the challenge is to determine which species mixtures improve the overall performance of the agroecosystem through better use of available environmental resources in the crop system considered. Although the agronomic relevance of the mixed canopies (i.e. intercrops) has been experimentally highlighted in low input context, the results are highly variable depending on the environmental conditions.

Our goal is to develop predictive approaches based on concepts from community ecology to design intercrops. These concepts suggest that two ecological processes are particularly involved in the performance of intercrops: niche complementarity can be quantified by the distance between key functional traits (plant morphological and physiological characteristics) of the mixed plant species; phenotypic plasticity can be quantified by the variance of these traits between different cropping conditions. We seek to apply these concepts to mixed annual crop species by calculating complementarity and plasticity metrics to predict the performance of intercrops in a wide range of environments.

For this, we have a database of traits measurements on a dozen crop species, each represented by several varieties and in different environments (sites, years). From these measurements and pedo-climatic variables, we will mobilize data science approaches in order to reduce the size of observed data (data mining) and to predict the performance (regression, machine learning) of intercrops in a large range of cropping conditions.

Applied Mathematics – Life & Environmental Sciences

Hyperspectral imaging and omic methods to characterize variability of plant responses to combined stresses. Application to wheat crop

  • Keywords: multi sensors ; biotic stress, abiotic stress ; image analysis ; data processing ; data fusion  ; machine learning
  • Contact: Nathalie Gorretta – nathalie.gorretta [AT] irstea.fr
  • PhD director(s): Ryad Bendoula, Irstea ITAP / Pierre Roumet, Inra AGAP
  • Supervisor(s): Nathalie Gorretta, Irstea ITAP / Elsa Ballini, Montpellier SupAgro BGPI / Martin Ecarnot, Inra AGAP
  • Research Units: ITAP, AGAP
  • Co-Funding:  Irstea
  • #DigitAg : Axis 3 : Sensors and data acquisition/processing, Challenge 3: ICT and crop protection, Axis 2: Innovation in digital agriculture Axis 5: Data Mining, Data Analysis and Knowledge Discovery, Challenge 1 : Agreécology,Challenge 2: Digital solutions to optimize the genotype in changing production systems and markets

The development of new agricultural practices based on principles of agroecology needs to analyze complex interactions between plants and their biotic and abiotic environments. Our ability to analyze and to decipher these multiple stresses requires, on the one hand, adopting a new conceptual framework such as phytobiome or extended phenotype, and on the other hand, coupling methodologies that make it possible to analyze the whole plant phenotype – or targeted organs – at different scales (metabolome, cell, composition and tissue structure). In this work, we propose to combine innovative approaches based on light-matter interactions with more classical ‘omics’ approaches to assess the variability of durum wheat response to multiple stresses: biotic (Septoria, Brown rust) and abiotic (water stress).

Applied Mathematics – Life & Environmental Sciences

Methods and metrics to assess the spatialisation of crop models for Precision Agriculture: application to a vine water status model.crop modelling, spatialisation, vineyards

  • Keywords: crop modeling, spatialiazation, vineyard
  • Contact: Taylor, James – james.taylor@irstea.fr
  • PhD director(s): Taylor, James, ITAP, Irstea / Tisseyre, Bruno, ITAP, Montpellier SupAgro
  • Supervisors : Taylor, James, ITAP, Irstea / Roux, Sebastien, MISTEA, INRA / Tisseyre, Bruno, ITAP, Montpellier SupAgro
  • Research Unit(s): ITAP, MISTEA
  • Co-funding :  Irstea
  • #DigitAg : Axes 6 & 5 ; Challenges 5 & 3

This thesis will investigate how high-resolution spatial and temporal agricultural data can be incorporated into existing crop models to enhance the short-term predictive capabilities of the crop model. The intent is to make crop models more useful for short-term, tactical management at a sub-field scale. Unlike most previous work at spatialising a crop model this project downscales the crop model to provide predictions at finer spatial and temporal resolution. This increases the potential for stochastic effects and spatial and temporal autocorrelation in the input data to negatively impact the model simulation. It also generates a question as to the effect of changing the scale of the input data and the scale of the output on the quality of the simulation. To address these concerns, a vine water status (VWS) model will be used as a first test case. The VWS model will be spatialized with high-resolution vine canopy and soil information.

Within the project, methods for assessing the spatialized crop model performance (under changing scales of inputs and outputs), that account of potential spatial autocorrelation in the data and outputs, will be investigated to ensure the assessment of model performance when targeting precision agriculture applications.

The originality of the thesis will be in the development of methods to assess spatial crop models for precision agriculture (site-specific crop management) applications. Interfacing the increasing amounts of observed agriculture data with the inherent physiological knowledge captured in crop models is a clear area where digital agriculture can have a rapid, short-term effect. Spatialised crop models have the potential for use at national and regional scales for food security, agriculture policy and supply chain management. They also have the potential to enhance site-specific tactical management, and this thesis addresses this latter area. It will make use of multi-temporal spatial crop production data (evolving vine canopy data), digital soil mapping products and the latest in vineyard sensing to generate information to support the use of VWS models.

Humanities & Social Sciences

Digitalisation and transformation of agriculltural R&D. New service models for new agricultural models?

Keywords: Digitalisation, advisory services, R&D, public-private partnership, institutional economics, collective organisations

  • Contact: Pierre Labarthe – pierre.labarthe [AT] inra.fr
  • PhD director(s): Pierre Labarthe, Inra AGIR / Jean-Marc Touzard, Inra Innovation
  • Supervisor(s): Toillier, Aurélié, Innovation, CIRAD / Labatut, Julie, Agir, INRA / Vermeulen, Ben, , Université de Hohenheim
  • Research Units: AGIR (Inra Toulouse), INNOVATION (Inra Montpellier)
  • Co-Funding:  Inra
  • #DigitAg : Axis 1: TIC and rural societies, Challenge 5: ICT and new farm advisory services

The aim of the PhD is to analyse the effects of the digitalisation of agriculture on the functioning of agricultural R&D.  There is a need to better understand how digitalisation transforms the collective organisations that farmers have set to produce knowledge on the technologies they use. Digitalisation of agriculture comes with the emergence of a market for Decision Support Tools, with alliances and competition between actors to create value and knowledge with and for the farmers. New actors are active in this market: start-ups, multinational firms…  These actors can potentially impact the economic models of services delivered to farmers (advisory services, knowledge brokering, experimental platforms…). They can also contribute to changes in the networks, rules and institutions of agricultural R&D. These trends also question the nature and content of agronomic knowledge on which farmers’ Decision Support Tools are built. What are the sources for this knowledge? Who invest in the validation of the relevance and robustness of this knowledge? Is this knowledge compatible with the development of agro-ecological conceptions of agriculture? The PhD thesis will benefit from the research project H2020 AgriLink, coordinated by Pierre Labarthe. This will allow the PhD candidate to implement a comparison between three countries: France, the Netherlands and United Kingdom. The methodology will be based on qualitative field work, with interviews of public and private actors of agricultural R&D and of the supply of services for farmers. An in-depth analysis will be implemented on the development of a given technology in the three countries. The technology is to be identified by the PhD candidate. This PhD will reinforce the relations between the laboratories AGIR and Innovations for the analysis of digitalisation of agriculture in a pluridisciplinary perspective, building on economics and management sciences.

Humanities & Social Sciences

Digital transition in agriculture: description of its deployment and impact on the positioning of farmers in the value chain

Keywords: Digital Transition, Technological Adoption, Appropriation, Technological Use, Economic performance, Value Chain, Organizational Capabilities, Competitive advantage

  • Contact: Isabelle Piot-Lepetit – isabelle.piot-lepetit [AT] inra.fr
  • PhD director(s): Isabelle Piot-Lepetit, Inra MOISA / Isabelle Bourdon, Université de Montpellier MRM
  • Supervisor(s): Karine Gauche, Montpellier SupAgro MOISA/ Magali Aubert, Inra MOISA
  • Research Units: MOISA, MRM
  • Co-Funding:  Inra
  • #DigitAg : Axis 1: TIC and rural societies, Challenge 7: ICT for a better acknowledgement of agriculture in the global value chain, Axis 2: Innovation in digital agriculture, Challenge 5: ICT and new farm advisory services

The project focuses on describing the current state of the digital transition in agriculture and future directions. The objective is to understand and explain the determinants and obstacles to the use and appropriation of digital technology in agriculture, with a perspective that goes beyond the mere diffusion and adoption of innovations. Using concepts from the domains of economics, management and sociology, the selected approach makes it possible to characterize the heterogeneity of behaviors, the implied evolutions for current jobs, and the possibilities of changes in the positioning of farmers in the value chain. From a methodological point of view, the project will develop an analytical framework with a temporal dimension and crossing points between the different stages of the digital transition. The economic impact will be captured through the organizational capabilities approach that allows the identification of practices that generate a competitive advantage and therefore a better positioning in the value chain. Empirically, surveys will be conducted with farmers, agricultural advisers, experimental farms, and digital solution publishers to describe and understand the digital transition in agriculture. Then, interviews will be held to verify the results coming up from surveys and characterize the future directions of this digital transition.

Humanities & Social Sciences

Application phase completed

Digital agriculture: the question of labor in agricultural holdings. Territorial trajectories and coexistence

Keywords: Digital agriculture, labor, agricultural holding, territory, farm management, automation, ICT, innovation

  • Contact: Pierre Gasselin – pierre.gasselin [AT] inra.fr
  • PhD director(s): Lucette Laurens, Université de Montpellier III Paul Valéry, UMR Innovation
  • Supervisor(s): Pierre Gasselin Pierre, Inra UMR Innovation / Lucette Laurens, Université de Montpellier III Paul Valéry, UMR Innovation / François Purseigle, Inra AGIR
  • Research Units: Innovation (Inra Montpellier, AGIR (Inra Toulouse=)
  • Co-Funding:  Inra
  • #DigitAg : Axis 1: TIC and rural societies, Challenge : cross-cutting subject, Axis 2: Innovation in digital agriculture, Axis 6: Multiscale modelling and simulation

Digital agriculture (DA), including information and communication technologies (ICT) and automated agricultural machinery, is the subject of many hopes to transform agricultural models. Some even talk of a revolution and consider it as a source of innovation sometimes for a more sustainable agriculture sometimes for more competitive farms. The scientific literature focuses on the operation of these digital tools and their economic impacts. A broad field of research seeks to understand the contribution of this DA to the transformations of agriculture. Agricultural holding  is a relevant entry to understand both the consequences of the DA on the ongoing changes in labor in agriculture but also to identify changes in the management strategies. Mobilizing analytical frameworks of economic and social geography, this thesis project explores the changes induced by the DA in the socio-economic and territorial logics of farms. These results will allow us, in a second time, to study how the DA changes the modes of coexistence of agricultural holdings and agricultural models in the territories studied.

Engineering Science – Humanities & Social Sciences

Building of a know-how of experience knowledge base using natural language processing guided by a domain ontology. Application to AOP cheese making chains

Keywords: Decision support system, Knowledge engineering, Natural language processing, Social LCA, Ontology, Recommendation

  • Contact: Patrice Buche – patrice.buche [AT] inra.fr
  • PhD director(s): Patrice Buche, Inra IATE / Catherine Macombe, Irstea ITAP
  • Supervisor(s): Nathalie Hernandez, Université de Toulouse Jean-Jaurès IRIT,/ Pierre Bisquer, Inra IATE
  • Research Units: IATE, ITAP
  • Co-Funding:  Inra
  • #DigitAg : Axis 5: Data Mining, Data Analysis and Knowledge Discovery, Challenge 7: ICT for a better acknowledgement of agriculture in the global value chain, Axis 4: Information systems, data storage, transfers and sharing, Challenge 5: ICT and new farm advisory services

Cheese making chains enjoying an “Indication Géographique” (AOP/IGP) base their product differentiation strategy on the valorization of local resources in connection with their terroir and on the expression of the know-how of experience as well at the level of the production as the transformation. Internal evolutions to appellations, especially in terms of turn-over and training of operators, strongly weaken the preservation and transmission of these know-how.

The development of artificial intelligence methods allowing the exploitation of knowledge bases opens new perspectives in terms of preservation and reuse in systems for recommending know-how from operational experience. As part of the CASDAR Docamex project (2017-2020), the knowledge engineering team (ICO team) of UMR IATE is developing a method and a tool for decision support (DSS) for the farmer who helps to control a defect or quality of manufacture by recommending the most relevant technological actions to undertake. The DSS also allows, for a given action, to determine all the defects and qualities impacted. These recommendations are based on the formal representation of the causal relationships linking default / quality to actions by explanatory mechanisms. In Docamex, the ICO team works with several AOP chains of different characteristics (Comté, Reblochon, Emmental of Savoy, Salers, Cantal) in order to develop a generic and adaptable DSS. The knowledge manipulated by the DSS is formalized with the Semantic Web languages ​​well adapted to integrate knowledge from heterogeneous sources in the chains.

The generic methodology developed in the PhD thesis will propose, in order to enrich the DSS knowledge base, to exploit and adapt Natural Language Processing (NLP) approaches developed by the IRIT MELODI team based on the semantic support provided by a core ontology and domain ontologies defined by the ICO team of UMR IATE as part of the CASDAR DOCAMEX project. The recommendation tool using the knowledge base will be evaluated by a social LCA method proposed by UMR ITAP’s Environmental and Social Performance Assessment team.

Humanities & Social Sciences- Engineering Science 

Risk Management, Governance and Performance of Agricultural Cooperatives: Development of an exploratory and theoretical analysis and design of a risk management tool Agricultural cooperatives, farmers members, risk management, governance, performance, data mining, modelling, literature review, decision support tool

  • Keywords : Agricultural cooperatives, member farms, risk management, gouvernance, performance, data mining, modeling, literrature review, decision support tool
  • Contact : PIOT-LEPETIT Isabelle – isabelle.piot-lepetit@inra.fr
  • Supervisors : PIOT-LEPETIT Isabelle, UMR MOISA, Inra / TEISSEIRE Maguelonne, UMR TETIS, Irstea
  • SAISSET Louis-Antoine, UMR MOISA, Montpellier SupAgro / SAUTOT Lucie, UMR TETIS, AgroParisTech / VALETTE Justine, MRM, Université de Montpellier
  • Unit receiving the PhD student: MOISA, TETIS
  • Co-funding :  Inra
  • #DigitAg : Axes 1 & 4 ;  Challenges 7 & 5

Business and financial risk management is at the heart of the new challenges faced by agricultural cooperatives and a main element supporting their performance and sustainability. Cooperatives finance their investment, but also cover risks by adjusting their proportion of equity and debt, ensuring the sustainability of their activities by absorbing unexpected losses. Since the risk is affected by various factors, the topic of this project is, on the one hand, to characterize the existing relationships between risk, governance and performance in French agricultural cooperatives and, on the other hand, to identify stable structural relationships between the different elements considered, so as to design a decision support tool allowing cooperatives to first characterize their risks and the impact on their performance according to their governance type and second, to follow their variations in order to set up early warnings and consider actions to fix them from the very beginning. This risk prevention tool will also help to advise cooperatives’ farmers members. To do this, two methodological approaches will be considered: data mining to provide knowledge on existing structural relationships among data and a theoretical analysis based on a meta-analysis of the literature in Economics and Management Science with an empirical application to the available dataset. A synthesis of the results from both approaches will make it possible to position their complementarity and identify elements that lead to the similar or divergent conclusions, so as characterizing stable structural relationships. It may also allow the characterization of potential new structural relationships, not captured by the theoretical analysis, thus complementing and extending existing knowledge on the relationship between business and financial risk, governance and performance in French agricultural cooperatives.

Engineering Science – Life & Environmental Sciences

Knowledge representation and reasoning for agro-ecological systems

Keywords: Agroecology – Data and knowledge management – Reasoning – Semantic Web – Artificial Intelligence – Semantic Web

  • Contact: Marie-Laure Mugnier – mugnier [AT] lirmm.fr
  • PhD director(s): Marie-Laure Mugnier, Univ. Montpellier Lirmm & Christian Gary, Inra SYSTEM
  • Supervisor(s): Pascal Neveu, Inra MISTEA  & Raphaël Metral, Montpellier SupAgro SYSTEM
  • Research Units: LIRMM, SYSTEM & MISTEA
  • Co-Funding:  Inria
  • #DigitAg : Axis 4: Information systems, data storage, transfers and sharing, Challenge 1: Agroecology

The scientific question addressed by this thesis is the following: how to formally represent complex systems such as agro-ecological systems, to allow an automatic exploitation of this representation based on its semantics?

It is located in computer science in the field of knowledge representation and reasoning, a branch of artificial intelligence, which provides the theoretical and algorithmic foundations for the research to be carried out. This work has an interdisciplinary dimension and will be developed in close collaboration with agricultural researchers studying agro-ecological systems. Indeed, understanding agro-ecological systems poses challenges common to both disciplines: addressing the complexity of agro-ecological processes and their interactions, articulating several types and forms of knowledge, representing and managing dynamic and multi-scale processes.

The results of the thesis will contribute to the creation of a tool (i) for eliciting, formalizing, integrating and sharing data and knowledge on the functioning and management of agroecosystems for the agro-ecological transition of agriculture, and (ii) offering various services based on the semantics of this data and knowledge, including: exploration and query based on domain ontologies, verification of the consistency of the modelling, analysis of the behaviour of the system and explanation of the inferences made, highlighting the consequences of system disturbances (in particular with a view to helping to formulate scientific hypotheses, for example, is the association of a certain type of plant with a crop likely to reduce the risk of pests and diseases?), the evaluation of technical change scenarios that meet specific objectives (e. g. reducing pest attacks).

This thesis will be accompanied by projects in partnership with agricultural development organizations that will assess its products with a view to supporting farmers in an agro-ecological transition process.

Engineering Science – Life & Environmental Sciences

Interpretability of distribution models of plant species communities learned through deep learning – application to crop weeds in the context of agro-ecology

Keywords: Agro-ecology, Deep Neural Networks, Transfer learning, Interpretability, Interactions, Landscape, Agricultural Practice, Biodiversity, Crop Weeds

  • Contact: Alexis Joly – alexis.joly [AT] inria.fr
  • PhD director(s): Alexis Joly, Inria Lirmm, Inria / François Munoz, Univ Grenoble Alpes, LECA
  • Supervisor(s): Pierre Bonnet, Cirad, AMAP, Maximilien Servajean, Univ. Montpellier 3-Lirmm
  • Research Units: LIRMM, AMAP
  • Co-Funding: Inria
  • #DigitAg : Axis 5: Data Mining, Data Analysis and Knowledge Discovery, Challenge 1: Agroecology

The modelling of interactions between biodiversity, landscape and agricultural practice is one of the major challenges of agro-ecology. Very recently, environmental species distribution models based on deep neural networks have begun to emerge. These first experiments showed that they could have a strong predictive power, potentially much better than the models used traditionally in ecology. One of their advantages is that they can learn an environmental representation space common to a very large number of species so that the prediction performance can be stabilized from one species to another. A first objective of the thesis will be to extend such transfer learning principle to the context of agro-ecology. In particular, data characterizing the landscape and the agricultural practices will be integrated for the prediction of crop weeds and/or associated functional traits. The second objective of the thesis will be to remove the lock on the interpretability of these models in order to deduce new tangible knowledge in agro-ecology. This will include qualifying the environmental representation space learned by the neural network in its terminal layers, typically the last layer of description on which the final linear regression or classification is based. The variables (neurons) in this representation space necessarily correspond to deterministic ecological and environmental patterns, but their exact nature is totally unknown. In the case of the deep agri-environmental models targeted in the thesis, these patterns will also integrate information from the landscape and agricultural practices. Their analysis will provide a better understanding of whether or not massive integrative approaches, based on a wide variety of input data, are needed, or whether they should focus on certain key factors.

Engineering Science – Life & Environmental Sciences

Definition, Design and Evaluation of a decision support system for pastoralism

Keywords: Information system, Heterogeneous data, Pastoralism, Agroecological transition

  • Contact: Lucile Sautot – lucile.sautot [AT] agroparistech.fr
  • PhD director(s): Maguelonne Teisseire, Irstea TETIS
  • Supervisor(s): Lucile Sautot, UMR TETIS, AgroParisTech / Magali Jouven, UMR SELMET, Montpellier SupAgro / , ,
  • Research Units: TETIS, SELMET
  • Co-Funding:  AgroParisTech
  • #DigitAg : Axis 4: Information systems, data storage, transfers and sharing, Challenge 4: ICT and sustainable animal production, Axis 5: Data Mining, Data Analysis and Knowledge Discovery

Pastoral systems are widely recognized for their social, environmental and cultural value. They also represent a type of livestock farming system consistent with agroecology. Their sustainability depends on their ability to cope with wide spatio-temporal variations in the availability of pastoral resources. Thus, both animals and farmers need to constantly adapt their strategies to the changing context.

Decision making and diagnosis in pastoral systems rely on the analysis of heterogeneous data from various sources (local and scientific knowledge, direct observations and technical references, embarked sensors such as GPS). Such heterogeneous data is more or less available to the farmer or technical advisor, and its comprehensive analysis is carried out on an informal basis and with varying success.

The digital age offers the opportunity to link data that could not be previously correlated. What are the methodological solutions in Computer Science to manipulate and link such data? How to evaluate and measure the interest and the impact of this information for farmers and technical advisors?

The proposed work aims at addressing these issues by (1) defining a data warehouse model allowing the analysis and cross-referencing of heterogeneous data (digital platform offering new services to farmers according to spatial dimensions, temporal and thematic); and (2) carrying out an analysis of the contribution of these new types of information in terms of lever for innovation for pastoral livestock farming systems, at several levels and with a suitable multicriteria analysis.

Engineering Science – Life & Environmental Sciences

Combining T-LiDAR, deep-learning methods and functional-structural plant models for the quantification of architectural and functional traits in fruit trees

  • Keywords: LiDAR, deep learning, functional-structural plant model, multispectral and thermal imaging, fruit trees
  • Contact: Frédéric Boudon & Benoit Pallas – benoit.pallas [AT] inra.fr
  • PhD director(s): Marc Chaumont, LIRMM, Univ Montpellier / Evelyne Costes, AGAP, Inra
  • Supervisor(s): Frédéric Boudon, Cirad AGAP / Benoît Pallas, Inra AGAP / Emmanuel Faure, CNRS LIRMM
  • Research Units: AGAP, LIRMM
  • Co-Funding:  Inra
  • #DigitAg : Axis 3: Sensors and data acquisition/processing, Challenge 2: Digital solutions to optimize the genotype in changing production systems and markets, Axis 6: Multiscale modelling and simulation, Challenge 1: Agroecology

In the context of climate change and limitation of environmental impacts, the evaluation of genetic resources and breeding of new fruit tree genotypes more adapted to sub-optimal growing conditions become crucial.  Architectural traits should be considered in these new breeding programs to account for the potential genotypic production and the interaction between genotype and environment (light, water resources, diseases). Moreover, for fruit trees, architectural characteristics are of major importance since they determine the labor time dedicated to tree training by pruning or bending procedures. In this project, we propose to use new technologies based on terrestrial LIDAR scans for high-throughput phenotyping of apple tree orchards. If some first methods have been already used in forests and in agronomical fields for estimating plant biomass or leaf area, identifying all individual plant vegetative and reproductive components remains challenging. We thus want to assess the ability of deep-learning methods for improving plant reconstruction methods and make them more accurate. The methodological issue will be to adapt such formalisms to LIDAR point clouds with contrasted resolutions and to constitute database of examples based on simulation or partial reconstructions. A second objective of this project is to apply some sensor fusion by coupling LIDAR data with thermal and multispectral images for evaluating functional-related plant traits. For this, functional structural plant models will be used as integrators of data. The study will be carried out on an apple tree core collection on which phenotypic traits coming from airborne imaging and LIDAR have been already acquired. The final objective will be to evaluate the genotypic variability and heritability of all these traits in order to propose new traits for plant breeding programs.

Engineering Science – Life & Environmental Sciences

Learning of hybrid rules for the analysis of the dynamics of plant diseases and pests according to climatic conditionsdata science, plant health monitoring, machine learning, vines, arable crops

  • Keywords : data science, épidémiosurveillance, machine learning, vigne, grandes cultures
  • Contact: François Brun – francois.brun@acta.asso.fr
  • PhD director(s): Alexandre Termier, LACODAM, Inria / David Makowski, Agronomie & CIRED, Inra
  • Supervisors:Luis Galárraga, LACODAM, Inria / François Brun, Acta
  • Research Unit(s): Acta, Inria / LACODAM
  • Co-funding :  Acta
  • #DigitAg : Axes 5 & 6, Challenges 3 & 8

As the list of different phytosanitary problems on various crops is very long, agricultural Research & Development services cannot devote enough efforts to develop decision support tools (DST) for effective crop protection in all cases. One of the goals of such tools is to limit the use of phytosanitary products in crops. Given the plethora of existing data on the dynamics of diseases and the usage of phytosanitary products in crops, such a goal requires us to exploit data science techniques in order to accelerate our understanding on the dynamics of diseases as well as to automate the processes for decision support.

This thesis aims at assisting crop protection experts by automating the discovery of working hypotheses underlying future DST on plant disease behaviour, using learning techniques based on the mass of data collected in epidemiological surveillance networks. We will mobilize original machine learning methods. It will consist in finding “hybrid rules” : they predict a numerical variable (e. g. the incidence of downy mildew disease) by taking into account the interactions between categorical variables (e. g. the phenological stage of the plant) and statistical models on numerical variables. This original approach seems to us to be a promising way of reconciling predictive modelling and “local” configuration in line with the great diversity of agronomic situations to be considered.

Exchanges with experts in diseases and pests on this work will make it possible to build new decision-making tools to better understand plant protection treatments. This work can also help to improve the effectiveness of plant health monitoring systems by making better use of the observed data collected in this context.

Engineering Science – Life & Environmental Sciences

Animal wellfare: characterizing the diversity between and within livestock farming situations with data mining methods used on information from dairy herds sensorsWelfare ; data mining ; sensors, dairy cows

  • Keywords: Bien-être ; data mining, capteurs, vache laitière
  • Contact: Alexandre Termier – alexandre.termier@irisa.fr
  • PhD director(s): Alexandre Termier, Inria – Lacodam, Université de Rennes 1 / Inria / Yannick Le Cozler, INRA – PEGASE, AGROCAMPUS OUEST
  • Supervisors: Alexandre Termier, Inria – Lacodam, Université de Rennes 1 / Inria / Yannick Le Cozler, INRA – PEGASE, AGROCAMPUS OUEST / Véronique Masson, Inria – Lacodam, Université de Rennes 1 / Inria
  • Research Unit(s): Inria – Lacodam, INRA – PEGASE
  • Co-funding:  Inria
  • #DigitAg: Axis 5, Challenges 4 & 7

Nowadays, the consumer as well as the citizen are expecting herd management strategies that take into account animal welfare. Animal welfare is satisfied if the animal is in good physical and psychological health, feels good and does not suffer. This led to the design of certification systems. Most of them are based on means obligations that are easy to verify (building type, surface per animal, equipment…). However, these are not sufficient to guarantee animal welfare. To guarantee an accountable performance obligation, using monitoring technologies (sensors for reproduction or health,…) can contribute to better evaluate the different dimensions of animal welfare, after a proper processing of the data. To date, monitoring data analysis is mainly focused on the individual animal, and do not allow to understand the diversity and variability of herds. Such an analysis would allow establishing a precise “map” of the behaviors observed in cattle herds and to distinguish behaviors coming from individual animals and those coming from the management strategy. Data Mining techniques are especially well suited for this task. “Discriminative pattern mining” approaches can discover regularities or irregularities in data for groups of animals or groups of herds. Such information can be used to establish a taxonomy of herd behaviors, allowing a quick positioning of each herd depending on its management strategy. The farmer would then have precise information on the welfare status of its herd and the possibilities of improvement. Labelling systems could also better answer to consumer expectations.

Engineering Science – Life & Environmental Sciences

Potential for combining time series of multispectral and radar satellite data to develop  spatialized variables of vine development at the regional scale: application to the estimation of growth and its cessation and integration into a spatialized water status model.time series, remote sensing, water stress, vines

  • Keywords: time series, remote sensing, water stress, vineyard
  • Contact: bruno tisseyre – bruno.tisseyre@supagro.fr
  • PhD director(s):  bruno tisseyre, UMR ITAP, Montpellier SupAgro / James Taylor, UMR ITAP, IRSTEA
  • Supervisor(s): Bruno TISSEYRE/James TAYLOR/Jean-Michel ROGER, UMR ITAP, Supagro/Irstea / Aurélie METAY, UMR SYSTEM, Montpellier SupAgro / Nicolas DEVAUX, UMR LISAH, Montpellier SupAgro
  • Research Unit(s): UMR ITAP, UMR SYSTEM
  • Co-funding :  Montpellier Supagro
  • #DigitAg : Axes 3 & 6, Challenges 5 & 6

This thesis will investigate the potential of new sources of satellite-based earth observations to be used in operational decision support in viticulture (either directly by experts or as a model input). It will primarily focus on imagery acquired from the ESA Sentinel satellites (Sentinel 1 and 2). In effect, these new sources of earth observation present unique characteristics in terms of the improved revisit time (5 days), spatial resolution (10 m), types of imagery available (multispectral and radar), area covered (regional and greater) and the cost (free imagery). As such, they are extremely interesting for spatial viticulture applications at multiple scales ranging from sub-field to regional.

The originality of the thesis will be in the adoption and adaptation of image analysis techniques to extract relevant and pertinent information from the various image types that :

  • are multivariate and heterogenous with the attribute space (optical and radar based imagery)
  • have a temporal dimension (defined by the revisit time and free access)
  • exist in a geographical context, i.e. the data are spatial.

The thesis proposes to explore the data within the images using methods based originally in chemometrics and spectral analysis but integrated with a spatial (geostatistical) and/or temporal (time-series) functionality. The objective will be to extract spatial descriptors (metrics, indices etc…) from these new sources of information that provide information relating to water stress in grapevines. These descriptors could be used to validate zoning and management, as a covariate in the extrapolation and mapping of point data or as an input into a predictive crop model. These potential applications have been identified because ;

  • water stress and its management is an important and historical field of research for the research group and the project will benefit from access to existing databases that will enable and enhance the research.
  • there is a strong social and industry demand for operational dignostics and tools to follow the evolution of water stress in vines at different scales with the intent to support enhanced management of grape quality and vine health.

Life & Environmental Sciences – Engineering Science

Co-design and sustainability of innovative breeding systems using alternative methods to hormonal treatments in the management of the reproduction of small dairy ruminants

Keywords: Automatic estrus detector, small ruminants, co-design, simulation, agent based models, innovative breeding, hormone-free reproduction

  • Contact: Amandine Lurette – amandine.lurette [AT] inra.fr
  • PhD director(s): Eliel Gonzales-Garcia, Inra SELMET
  • Supervisor(s): Amandine Lurette, Inra SELMET / Patrick Taillandier, Inra MIAT / Nathalie Debus, Inra SELMET
  • Research Units: SELMET (Inra Montpellier), MIAT (Inra Toulouse)
  • Co-Funding:  Inra
  • #DigitAg: Axis 2: Innovation in digital agriculture, Challenge 4: ICT and sustainable animal production, Axis 6: Multiscale modelling and simulation, Challenge 1: Agroecology

The management of hormone-free reproduction in small dairy ruminant farming is a lever to be mobilized to respond to the challenges of agro-ecology and the evolution of societal demands. The Alpha-D® automated heat detector system remains to this day the only operational tool in sheep farms that allows artificial insemination without the use of hormonal synchronization. The introduction of such new tools implies change processes for the farms concerned. These changes may include biotechnical processes, at the herd level and/or surface management level. The objective of this thesis is double. On the one hand, it is a question of characterizing the processes by which conventional and organic farmers integrate the Alpha-D® device into their system and the implementation methods, taking into account changes in associated practices. On the other hand, the work aims to assess the sustainability of a diversity of systems (conventional and organic) that no longer use hormone treatments in their reproductive management. The comparison the simulation results with the actors will make it possible to validate the most appropriate scenarios for implementing hormone-free reproduction according to the sectors and breeding systems tested within the Rayon of Roquefort area.