Know more

For 2024, subjects from different disciplines are proposed in Mathematics and its applications, Humanities and Social Sciences as well as Engineering Sciences.

>Apply directly to the person in charge of your chosen subject

• Maths and its applications
• Human and social sciences
• Sciences for the Engineer

Maths and its applications

Algorithm for management zones generation driven by the performance of crop management operations
internship filled

 Keywords: zoning, optimization, spatial constraints, crop management operation

• Second scientific field: Sciences for the Engineer
• Duration of the internship (nb months): 6
• Begins on: 1/03/2023
• Research units: Mistea, Inrae
• Contact: patrice.loisel [AT] inrae.fr
• #DigitAg: Axe 2 : Innovations en agriculture numérique, Axe 6 : Modélisation et simulation (systèmes de production agricole), Challenge 0 : sujet transversal, Challenge 5 : Les services de conseil agricole

Zoning problems (dividing a plot into management zones) are at the heart of digital agriculture, as they enable the implementation of spatially differentiated Crop Management Operations (CMO). However, zoning algorithms are generally not directly determined by the impact of these spatialized CMO.  However, this information becomes accessible through the predictions of crop models or the use of adapted data. The aim of this internship is to develop a new algorithm for generating optimized delineation of crop management zones based on the performance of CMO. The algorithm will output a zoning associated with a CMO recommendation per zone (irrigation level for example). The optimal recommendation will be based on a variable spatially observed on the agricultural plot. The proposed zonings will respect spatial constraints that will be managed using a method already developed at UMR MISTEA [Loisel et al. 2019].
The first case study is that of a crop irrigated using various possible irrigation modalities and under a global quota constraint. The agronomic variable on which the zoning will be based will be the water reserve in the soil, mapped on a plot. The master student will be hosted by INRAE Montpellier's UMR MISTEA and co-supervised by researchers and associate professors from UMR MISTEA (P.Loisel and S.Roux)and ITAP (H.Jones).

Human and social sciences

Multi-criteria evaluation of the outcomes and impacts of digital tools in agriculture : the case of the Decision Support System SoYield®

Keywords: Impact evaluation ; multicriteria evaluation ; digital tools ; Global South ; mango value chain ; Senegal

• Second scientific field: Life and Environment sciences
• Duration of the internship (nb months): 6
• Begins on: 1/04/2024
• Research units: Innovation / Hotsys, Cirad
• Contact: chloe.alexandre [AT] cirad.fr
• #DigitAg: Axe 1 : Impact des technologies de l'information et de la communication sur le monde rural, Axe 2 : Innovations en agriculture numérique, Challenge 8 : Développement agricole au Sud, Challenge 5 : Les services de conseil agricole

Recent technological developments and breakthroughs in data analysis have generated high expectations of the potential of digital tools to support farmers and other value chain actors in their decision-making and in the management of their activities. However, several recent studies suggest that this potential remains under-exploited (Steinke et al. 2020; Klerkx et al. 2019; Alexandre 2023). There is therefore a need to better identify the impacts (positive and negative) of the digitization of agriculture in both Northern and Southern countries.
The SoYield® decision support system (DSS) is a fruit production data collection and management tool designed to 1) inform farmers of their actual yields in order to guide their decision-making through the monitoring and analysis of production data from their orchards; 2) facilitate relations between industry players (producers, local buyers, exporters, processors, etc.) on the basis of objectively measured yields. By contributing to a better structuring of the Mango value chain through the sharing of information between the various players, the use of the SoYield® DSS aims to contribute to two main impacts: (1) poverty reduction for small-scale producers and local retailers; (2) enhanced food security. However, we do not yet know if and how the SoYield® DSS contributes to these impacts. Two other types of impact will also be explored as part of this internship: the environmental impacts generated by the digital technologies supporting the operation of the SoYield®DSS ; and the effects and impacts in terms of gender inequalities. The aim of this internship is therefore to develop and test a multi-criteria evaluation methodology to assess the social, economic and environmental effects and impacts to which SoYield® contributes. In this regard, this internship contributes to scientific and methodological reflections on the evaluation of the impacts of the digitization of agriculture in developing countries.

Sciences for the Engineer

High-throughput spatiotemporal reconstruction of root system architecture from frugal images using topological tracking and deep learning

Keywords: Root architecture, high-throughput phenotyping, frugal imaging, spatiotemporal reconstruction, deep learning, RSML models

• Second scientific field: Life and Environment sciences
• Duration of the internship (nb months): 6
• Begins on: 3/03/2024
• Research units: Agap, Cirad
• Contact: romain.fernandez [AT] cirad.fr
• #DigitAg: Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Axe 3 : Capteurs, acquisition et gestion de données, Challenge 2 : Le phénotypage rapide, Challenge 8 : Développement agricole au Sud

Plants expand their root system to meet their needs for water and nutrients, from germination to harvest. The architecture of the root system, its growth dynamics and its plasticity determine the plant's production and its ability to withstand adverse growing conditions (flooding, drought). Root traits are difficult to measure in the field. Destructive observation (shovelomics) does not allow to study large GxE panels or to observe root growth dynamics.
High-throughput phenotyping platforms have been developed with automated analysis pipelines based on convolutional neural networks [1][2]. These pipelines segment simple architectures. However, they are not reliable for more complex architectures. Until recently, these methods have never been evaluated for dynamic feature estimation. We studied this aspect in [3] with a spatiotemporal reconstruction algorithm. Its reliability for topology and dynamic feature estimation has been demonstrated. These modern phenotyping techniques have been developed for laboratories (Petri dishes, blotters, model species, imaging robots). There is a need to extend their capabilities in order to apply them to agronomic crops grown under conditions close to the field (Rhizotron) and observed using frugal techniques (simple camera). During this internship, we will study the limitations of reconstruction algorithms [3] and deep learning techniques [1][2] on these data in order to combine these approaches and extend their scope. We will work on the formal constraints of the algorithms to integrate more variate topologies and on the conditions (topological cost function [4], network architecture) to be met for a neural network to contribute to the reconstruction of topologically valid spatiotemporal architectures.

Development of semantic resources for field phenotyping and crop modeling

Keywords: Modeling (crop growth models), Ontology, Phenotyping, Semantic resources, Data schema, Thesaurus

• Second scientific field: Life and Environment sciences
• Duration of the internship (nb months): 6
• Begins on: 1/02/2024
• Research units: Mistea, Inrae
• Contact: Pascal.Neveu [AT] inrae.fr
• #DigitAg: Axe 4 : Système d’information, stockage et transfert de données, , Challenge 0 : sujet transversal, Challenge 2 : Le phénotypage rapide

One of the major challenges facing interdisciplinary scientific communities is the effective use and sharing of semantic resources (thesaurus, taxonomies, ontologies). In the context of phenotyping and crop modeling, these semantic resources need to be enriched and adapted to take into account different sources of data in the field, simulation models and, above all, research questions and phenotyping methods that are evolving by proposing new variables to be observed. Standards do exist, but they are not complete in relation to these new research questions and are difficult to evolve. As a result, their use can be an obstacle for researchers. The aim of this proposal is to enable controlled and efficient management of the evolution of semantic resources, while maintaining a link with existing standards. Standards and semantic resources in the field will be enriched, and a solution based on existing environments (AgroPortal) and ontology design patterns will be developed to provide researchers with a user-friendly strategy and tools.

Development of a text annotation and classification protocol for food security monitoring

 Keywords: food security, natural language processing, annotation, classification

• Second scientific field: Life and Environment sciences
• Duration of the internship (nb months): 6
• Begins on: 1/02/2024
• Research units: Tetis, Inrae
• Contact: maguelonne.teisseire [AT] inrae.fr
• #DigitAg: Axe 4 : Système d’information, stockage et transfert de données, Axe 5 : Fouille de données, analyse de données, extraction de connaissances, Challenge 6 : La gestion des territoires agricoles, Challenge 8 : Développement agricole au Sud

This internship aims to develop an annotation method for identifying risk factors linked to food security in West Africa, using data from journalistic sources. The predictive models used to calculate food insecurity indices combine data such as satellite imagery and climate data. The interpretation of these indices could be improved by integrating data from online articles or television news transcripts. This textual data could be used to pinpoint crises at a regional or even local level, in near-real time or retrospectively. In this context, it is becoming essential to deploy machine learning methods that can extract relevant information from vast volumes of textual data. This internship will aim to develop an annotation method to identify the structural and conjectural factors of food crises, as well as their spatio-temporal attributes, from a corpus of articles in French that has already been compiled ( https://agritrop.cirad.fr/602069/). This method will be used to build a corpus from which to evaluate (1) automatic language processing and machine learning models for extracting thematic information and spatio-temporal attributes (for example, via the adaptation of language models such as CamemBERT) and (2) data augmentation approaches to overcome the constraint of the limited size of the training corpus. In particular, we will rely on the ability of extended language models, such as ChatGPT, to generate new textual data that is semantically close to the annotated data. Time permitting, a contribution to the writing of an article describing the annotation approach and the annotated corpus could be envisaged.

Spatial transfer of deep learning models for rapeseed crop mapping
internship filled

 Keywords: Remote Sensing, Agriculture, Machine Learning, Land cover mapping

• Second scientific field: Life and Environment sciences
• Duration of the internship (nb months): 6
• Begins on: 1/02/2024
• Research units: Tetis, Inrae
• Contact: cassio.fraga-dantas [AT] inrae.fr
• #DigitAg: Axe 5 : Fouille de données, analyse de données, extraction de connaissances, , Challenge 6 : La gestion des territoires agricoles,

Nowadays, more and more remote sensing data are available, offering the possibility to follow a geographical area over time. The time series thus generated represent an essential source of information to efficiently manage agriculture on a territorial scale.
To this end, remote sensing data is used as input to machine learning (ML) methods to provide updated land cover maps. To do so, ML methods require a large amount of ground-truth data, which poses challenges for their applicability where little or no reference data is available. Re-using ground-truth data acquired at a particular study site to transfer the learnt model to a different area would avoid (or reduce) new costs and take advantage of previous investments. Unfortunately, directly transferring a model from a geographical zone to another one can be inefficient as the two regions may present different environmental and/or climatic conditions. This results in differences in the distribution of the acquired satellite data. This internship proposal aims at developing an innovative deep learning/transfer learning method with the aim to transfer a model learnt on a particular area (where ground truth data is available) to a different geographical area where no available ground truth data is accessible. In the context of this internship we will exploit freely available multi-temporal Sentinel-1 imagery, less sensitive to cloud occlusions due to the intrinsic nature of the SAR signal, with the aim to build a deep learning classification model for the mapping of the rapeseed crop culture. In addition, the designed deep learning method will exploit recent domain adaptation techniques to cope with the transfer task between three different geographical areas, namely: France, USA and Canada. We will evaluate the capability of the underlying deep learning model to be calibrated over one of the three areas and deployed on the remaining ones.