#DigitAg offers subjects for co-funded interdisciplinary theses
#DigitAg activities (in particular, doctoral student days and the #DigitAgora) and contributes to the exchange of information between his/her host laboratory and the #DigitAg community. The PhD student may also be invited to teach in the #Graduate School's masters programs or in the research schools organized by the Institute.
Publications resulting from thesis work must also mention that the work was supported by the Convergences #DigitAg Institute and France 2030 (ANR-16-CONV-0004).
The topics displayed have been selected by #DigitAg, which is co-funding them.
Want to know more? Get in touch with the contact indicated for the topic you are interested in
Selected topics:
Life and environmental sciences - Sciences for the Engineer
SemCrop – Semantic and modular representation of crop growth models
Keywords: Agroecology, Data assimilation, Crop2ML, Crop models, Hybrid models, Modularity, Semantics, Workflows
- Contact: Pierre Martre - pierrre.martre [AT] inrae.fr
- Thesis supervisors: Pierre Martre, Lepse, Inrae et Christophe Pradal, UMR Agap, Cirad
- Managers: Pierre Martre, Lepse, Inrae - Christophe Pradal, UMR Agap, Cirad
- Units: Lepse - Agap
- Cofunding: Inrae
- #DigitAg: Axe 6 : Modélisation et simulation (systèmes de production agricole), Axe 4 : Système d’information, stockage et transfert de données, Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Challenge 0 : sujet transversal, Challenge 1 : Le challenge agroécologique, Challenge 2 : Le phénotypage rapide, Challenge 3 : La protection des cultures, Challenge 5 : Les services de conseil agricole, Challenge 8 : Développement agricole au Sud
The use of crop models to predict the performance and environmental impact of crops is widespread at all levels of the value chain. Their use to reduce the use of inputs, to adapt agriculture to climate change, to diversify agro-systems, to preserve biodiversity, and thus to meet the objectives of the Green Deal, leads to constantly review their formalisms and to model new processes.
We have recently developed the Crop Modelling Meta Language (Crop2ML) model representation and transformation system, which allows the development of model components in accordance with FAIR principles. A current limitation of Crop2ML is the lack of semantics to search for components and facilitate their composition in operational modeling solutions.
The SemCrop project aims to address this limitation. The operational objectives concern the interoperability of the modeling tools and the links with the information systems collecting very large data sets. A modularity of the models at the process level is aimed at allowing a better integration at different scales, facilitating the link with the data, and the feedback between data and models (digital twins).
By proposing a modular modeling system, SemCrop will bring an original contribution to address the challenges of ecological, climate, and digital transitions. It will provide #DigitAg and regional AgTech companies (ITK, SMAG, FruitionSciences,...) with innovative tools to develop digital solutions for agriculture. SemCrop will increase the international scope of #DigitAg's research via its insertion in the AMEI (Agriculture Model Exchange Initiative) initiative coordinated by the supervisors.
Data-driven Methods for Modeling the 3D Structure and Growth Patterns of Plants: From Simulation to Real-World Applications. Application to Pests and Disease attacks detection
Mots-clé: Science des données, Apprentissage profond, Analyse de la structure des plantes, Séries temporelles, Coffea Robusta
- Contact: Maguelonne Teisseire - maguelonne.teisseire [AT] inrae.fr
- Directeurs de thèse: Teisseire Maguelonne, UMR Tetis, Inrae et Laga Hamid, , Université de Murdoch
- Encadrants: Maguelonne Teisseire, UMR Tetis, Inrae - Marc Jaeger, UMR Amap, Cirad
- Unités d’accueil: Tetis - Amap
- Cofinancement: Université de Murdoch
- #DigitAg: Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Axe 3 : Capteurs, acquisition et gestion de données, Challenge 3 : La protection des cultures, Challenge 8 : Développement agricole au Sud
Understanding plant growth and modelling how various environmental, genetic and management parameters affect growth are important for assessing its health and its potential yield. Traditional methods however use sparse plant measurements ignoring its 3D structure and the various correlations that exist between various plant parts. In this thesis, we propose to develop novel data-driven methods for the analysis of the 3D structure of plants and for modeling its growth patterns. The underlying idea is that instead of explicitly enumerating all the rules that govern the growth process, we propose to develop novel machine learning techniques that are capable to discover these rules from data. This PhD research work will involve two main steps. First, we will define a novel representation where plant deformations can be seen as trajectories. We plan to explore the latest development in generative deep learning and propose a novel network architecture that is capable to learn such representation from 3D images. Second, since deformations can be treated as trajectories in the novel representation, we propose to statistically model growth patterns of plants by analyzing trajectories in the representation space. Since different plants have different growth patterns and growth rates, a key component for building such statistical growth model is the dynamic time warping (DTW), which will enable us to perform the spatio-temporal registration of plants growing over time (either following a normal growth pattern or a growth pattern affected by diseases). The computational tools that will be developed in this PhD thesis will be tested on 3D images of different plants showing various structural complexities, including views from 3D simulated mockups. On the benefit of data gained from a EU Desira project, a specific validation is planned on Coffea Robusta plant captured at different growth stages.
Life and environmental sciences - Maths and its applications
Identifying unexpected observations in territorial crowdsourcing projects for agriculture: the case of vine water status monitoring at regional scale
Keywords: crowdsourcing, vine water status, outliers, unexpected observations, bayesian approach, geostatistics, conformal prediction
- Contact: Léo Pichon - leo.pichon [AT] supagro.fr
- Managers: Bruno Tisseyre, Itap, Institut Agro Montpellier
- Units: Itap - Mistea
- Cofunding : Agro (DGER) et/ou de la Région Occitanie.
- #DigitAg: Axe 5 : Fouille de données, analyse de données, extraction de connaissancesAxe 3 : Capteurs, acquisition et gestion de donnéesChallenge 5 : Les services de conseil agricoleChallenge 6 : La gestion des territoires agricoles
In the context of climate change, monitoring vine water status at the regional scale is of paramount importance for short- and long-term decisions. One of the promising approaches for this monitoring is the collaborative collection of observations by stakeholders of the wine industry, known as crowdsourcing. Crowdsourcing has already demonstrated its ability to collect a large amount of observations, particularly with the ApeX-Vigne application initiated by the project team. For the democratization of this approach, the missing link is the development of methods for analyzing the collected data. The identification of unexpected observations is a particularly important challenge, as these may be either aberrant observations that need to be eliminated to improve the overall quality of the dataset or, on the contrary, interesting observations that may reflect an original growing system or atypical soil and climate conditions. The phenomena under study are seasonal and generally follow a known and expected temporal dynamic. They also often depend on the environment (soil, climate, etc.), as for drought, and are therefore also spatially structured. The approaches that will be explored in this thesis will seek to use this knowledge about the phenomena under study to define expected behavior and identify observations deviating from it. The formalization of this knowledge may be based on historical data (e.g. historical time series of vine water status monitoring on reference plots) or auxiliary data (e.g. time series of remote sensing images). Spatio-temporal statistical methods will be used, and a Bayesian framework will be favored. Other approaches, such as the use of conformal prediction, may also be tested.
Towards a prototype for on-farm experimentation of biocontrol products on wheat, using a digital model approach
Keywords: Biosolutions, numerical model, wheat, septoria, On Farm Experimentation, agro-ecology
- Contact: Bénédicte Fontez - benedicte.fontez [AT] supagro.fr
- Managers: Bénédicte Fontez, Mistea, Institut Agro Montpellier - Elsa Ballini, PHIM, Institut Agro Montpellier
- Units: Mistea - PHIM
- Cofunding: Région, Institut Agro Montpellier, Métaprogramme Digit-Bio (Mathnum, Inrae)
- #DigitAg: Axe 5 : Fouille de données, analyse de données, extraction de connaissancesAxe 6 : Modélisation et simulation (systèmes de production agricole)Challenge 3 : La protection des culturesChallenge 1 : Le challenge agroécologique
Validating the effectiveness of biocontrol/biostimulation products is difficult in the field, which partly explains the lack of solutions or their low level of acceptance by the profession. As the factors influencing crop response and efficacity of biosolutions are numerous and interact with each other, it seems unrealistic to make a complete multifactorial experimentation even in the field. A digital model is a tool to manage creation and analysis of experimentation in an open system to evaluate the impact of biosolutions. The digital model will simulate reference or expected values and compare them with actual data. Several existing projects provide data that will be used to shape and test this numerical model. The core of the thesis involves formulating an initial theoretical model and adapting it to a more open system (in the field). This will involve 1) Establishing an initial mathematical model based on laboratory data 2) Proposing a numerical model (Combining machine learning and mathematical model) 3) Prefiguring a protocol for validating the effects of bio-solutions in a multi-plot 'On Farm' system. The thesis benefits from collaboration between the UMRs MISTEA and PHIM and a a partnership with the Avalis technical institute, the Vegenov technological resource centre (biosolutions experimenters), the Frayssinet company (biosolutions producer) and the BeStim network, to work towards rapid transfer of the numerical model to the profession and farmers.
Human and social sciences - Life and environmental sciences
Study of the economic impact of digital tools among West African farmers
Keywords: Digital – Productivity – Income – Agriculture – West Africa
- Contact: Serena FERRARI - serena.ferrari [AT] cirad.fr
- Managers: Catherine Araujo Bonjean, CNRS
- Units: Selmet
- Cofunding : Cirad
- #DigitAg: Axe 1 : Impact des technologies de l'information et de la communication sur le monde ruralAxe 5 : Fouille de données, analyse de données, extraction de connaissancesChallenge 0 : sujet transversalChallenge 8 : Développement agricole au Sud
In West Africa, rural poverty affects the majority of the population while agriculture in the region faces challenges such as income seasonality, lack of technical support, etc. The study will examine how digital tools, such as mobile phones and the Internet, are used in three different agricultural sectors (market gardening, cocoa farming and milk). The methodology of the study will involve surveys of farmers and the assessment of the impact of the use of digital tools on their incomes. It is expected that there will be a significant difference in income between agricultural households using and not using digital tools, an increase in income linked to the use of these tools, and the identification of inequalities related to gender, generations, geographical areas and agricultural value chains. The study will aim to provide crucial information for policymakers and development actors to optimize the adoption of digital tools while mitigating inequalities in the West African agricultural sector.
