Digital agriculture is unquestionably demanding interdisciplinarity. Indeed, designing new tools and services for digital agriculture with the ambition to transfer them to users requires not only knowledge in agricultural and digital sciences, but also in social, management and economic sciences for successful innovation and a good understanding of societal impacts. Scientific communities are organized in 6 main scientific areas related to:
Understanding ICT influence on rural societies
As ICT is currently profoundly changing our society, it is expected to also revolutionize the agricultural sector. Digital agriculture may open new avenues for improving management of farms, territories and food supply chains. The way ICT will contribute to these dynamics, from farm to territory and global value chain, is still to be investigated. In particular, new services will be created and it is essential to analyse how these innovations may contribute to organisational, economic and institutional changes. The economic impact of such new tools must be assessed, regarding productivity, competitiveness, governance processes in territories, public policy efficiency.
Farms targeting productivity gains may adopt precision agriculture technologies (e.g. sensors, automatic data processing, decision aid systems, etc.) whereas non-conventional farming may use other digital collaborative tools to communicate, access information or trade. #DigitAg will concentrate ICT adoption issues in agriculture.
ICT tools make it possible build more horizontal collaborations and new networks. New services are created and delivered by organisations of various natures and sizes even with newcomers extraneous to agriculture. It is therefore essential to analyse to what extent these innovations contribute to organisational, economic and institutional changes.
ICT may also induce new relationships between farmers and cooperatives, as well as new organisations and regulatory processes in territory or in agricultural supply chains. Lastly, the impact of such new tools must be assessed regarding productivity, competitiveness and sustainability criteria, as well as changes of governance processes at the rural territory level, and efficiency of public policies related to accompaniment, transparency, control and mobilisation of the agricultural sector.
What are the impacts of ICT-based innovation on the evolution of practices, of resource management processes, on farms performance, sustainability, resilience, exclusion, occupation? Do these new services induce new collaboration schemes? Are there differences in farmer - advisory service relationships, due to digital technologies? Is there a danger that farmers may lose some autonomy) in decision-making? Do ICT make private-public relationships evolve?
|Leaders: Leïla Temri (Montpellier SupAgro), Jean-Marc Touzard (INRAE)|
|Units: INNOVATION, MOISA, MRM, GREEN, LAMETA, AGIR (Toulouse)|
|Disciplines: Primary Economics, management and social sciences; Secondary, Agronomy (prospective knowledge on cropping systems), Digital sciences|
Construction of ICT-based innovation: technological, social and legal issues
This axis deals with social, legal, and management issues related to innovation in digital agriculture; a first scientific question is to understand the way digital tools and services should be built to be adopted by farmers. This raises topics dealing with design sciences, group learning processes, innovation diffusion… In addition to the technological and social issues, this axis deals with ethical issues and legal problems related to data and knowledge property. #DigitAg will contribute to build this knowledge.
Digital agriculture relies on the development of innovations that could be described as managerial, and their construction, dissemination and uses should be studied.
First, the issue of innovation building arises regarding design processes and the integration of users in the early stages of design. Co-design of decision support tools with farmers and regional actors, collective exploration activities, are important factors in product and service adaptation to actual needs and in their broader diffusion. Supporting different modes of design and enhancing participatory design implies studying the variety of uses and applications of such tools.
In a context where innovation is increasingly open, new forms of organizations supporting innovation have to be explored (innovation ecosystems, living labs…). The elaboration of this variety of new services raises questions on the development of operating procedures, governance, and funding (business models).
The next step is to analyze the dissemination of these innovations, including forms and determinants of this dissemination to farmers, companies, cooperatives and actors of territories.
Extraction, analysis and dissemination of the data finally poses ethical issues and legal problems related to intellectual property. Who has access to the data? Who controls and owns the data (private? public?), and to which ends? If intellectual property law does apply to the databases, the question will be then: will the databases be open and free for exploitation and will the data be considered susceptible for-meant for a large dissemination?
|Leaders: Sophie Mignon (University of Montpellier), Danielle Galliano (INRAE)|
|Units: MRM, DYNAMIQUES DU DROIT, MOISA, AGIR (Toulouse), INNOVATION, GREEN|
|Disciplines: Primary Management, law and social sciences; Secondary Agronomy (knowledge on crop management); Digital sciences|
Fostering the development of appropriate sensors and data acquisition systems, including crowdsourcing
Despite 30 years of research on sensors, agriculture still suffers from a lack of appropriate measurement techniques to satisfy certain parameters (e.g. disease, plant & soil composition...) at the right spatial and temporal resolution. There is a need to improve access to data and information on agricultural processes, at high resolution, through a better exploitation of satellite images, and by boosting innovation of low-cost, robust and easy-to-use sensors, including connected objects and smartphone-based devices. The optimal implementation of these technologies (sensor networks, portable sensors, embedded sensor, with energy and cost issues) in real situations is a key issue for a successful innovation, and requires innovation specialists. In #DigitAg, any research on sensor/data acquisition devices will include innovation specialists. Data processing - to turn data into information - and associated issues (e.g. uncertainty, output validation...) will also be dealt with in this axis.
Agriculture is characterized by several strict constraints with regard to data acquisition and sensor development: small markets, low margins, large and variable areas to cover, a high variability of objects of interest, rough conditions of use, and a low-to-medium level of ICT understanding from users. In accordance, sensors/ data acquisition systems must be robust, low maintenance, easy-to-use and with satisfactory metrological properties. Sensors can have different configurations: portable, in field (on animals, plants, soil, etc.), embedded in tractors and other agricultural tools or in aerial drivers (UAV, planes), or in satellites (earth observation). In addition to these classical measurement devices, data acquisition is to be carried out by farmers, though devices such as smartphones and tablet computers. Today, sensor/ data acquisition/processing techniques that match agricultural needs are still lacking.
#DigitAg aims at addressing this gap, especially regarding:
- assessment of plant, animal and soil state (physiology, composition, health…);
- sensor networks (e.g. weather sensors such as temperature, hygrometry, rain, hail…);
- embedded sensors in tractors, UAV;
- and multi-resolution satellite image processing.
The scientific questions linked to these issues are:
- a better understanding of light/matter interaction of optical and radar sensors,
- the application of IoT concepts to agriculture in context-aware approaches,
- the design of microsystems for measuring/transferring data on physical parameters (temperature, hygrometry, droplets…) at low costs, low energy (inclding energy harvesting and for sustainable agriculture,
- the ergonomy of smartphone based data-acquisition, to ensure good adoption by farmers.
|Responsables: Ryad Bendoula (INRAE), Philippe Combette (University of Montpellier)|
|Units: ITAP, TETIS, IES, LIRMM, LEPSE|
|Disciplines: Primary Physics, Engineering sciences, Optics, Electronics, Digital sciences; Secondary Agronomy (knowledge on physiology, crop modelling, indicators); Social and management sciences (user-centered design).|
Making progress in agricultural information system design
With the spread of connected objects (including on farm machinery tools), satellite imaging, high-throughput phenotyping, agriculture faces issues of data & knowledge management that industrial sectors have already started to address. However, in addition to classical big data topics (scalability, complexity, heterogeneity, privacy…), agricultural big data is also characterized by their time and spatial dimensions, which have to be dealt with. There is a need of specific data/ knowledge management systems (information systems, data storage, transfer, sharing) and procedures adapted to meet the new big data challenge in agriculture. Moreover, due to the complexity of interactions of agricultural processes and their very contextual characteristics, it is necessary to add intelligence to agricultural data to reuse them (common vocabulary, ontologies, rules …).
The above-mentioned embedded sensors, Internet of Things, crowd sourcing systems, etc., produce huge data that are potentially available for new services in agriculture. Up to recently, the use of information system in agriculture has been relatively behind the ones in the industrial sector. Information systems in the area of agriculture have some particularities that are not seen in information systems designed for business, as they have to take into consideration additional elements such as space and time. The current advanced agricultural data management processes, which can be used to share data and knowledge across disciplines, sectors and countries, falls within several scientific topics in Computer Science. Data must be retrievable, accessible, interoperable, and re-usable in order to produce datasets that will support interdisciplinary cutting-edge research aiming at meeting the present and future challenges of agriculture, food security and market needs.
The main challenges of data management are:
- Scalability (Big data, big applications),
- Complexity (relevance, uncertainly, confidence, multi-scale, etc),
- Heterogeneity of data sources and semantics,
- Privacy and Ethic (sensitive farm data, fundings, surveys, etc),
- Data flows and reproducibility (scientific workflows, provenance).
Moreover, due to interdisciplinary requirements and context-aware approaches, data management in the agriculture domain requires adding intelligence to data (common vocabulary, ontologies, rules).
|Leaders: Maguelonne Teisseire (INRAE, Tetis), Clément Jonquet (Université de Montpellier, Lirmm)|
|Units: ZENITH, GRAPHIK , MISTEA, TETIS, IATE, SELMET|
|Disciplines: Primary Computer sciences; Secondary Agronomy; Management sciences|
Designing new data mining methods, appropriate to agricultural data, to extract actionable knowledge
The real value of the growing volume of agricultural data (see Axis 3, 4) lies in its potential to be converted into actionable knowledge to provide agriculture with high-added value advisory services. Currently, no application efficiently exploits all the richness hidden in such data, that possess specific characteristics (time/ spatial dimension). There is a need for an integrated approach, combining data from monitoring systems in farms (crops, animals, buildings), from farmers, from scientists and technicians about technical itineraries as well as crowd-testing made by consumers.
Agricultural data are,at the same time, continuously growing in volume, while at the same time possessing multi-scale (time, space), uncertain, dynamic, and heterogenous features. They can also be sensitive in nature and may require solutions that preserve privacy. Unfortunately, there is no data mining approach capable of simultaneously handling all the features characterizing agricultural data. The work hypothesis is to build a methodology based on the broad spectrum of expertise and skills in statistical data analysis, data mining and integration of imperfect knowledge that the #DigitAg teams have developed for other sectors (e.g. medical sector). The proposed methods will be co-designed with domain actors (data scientists, advisor experts, farmers) and interaction with actors will take place at different levels, either to receive domain knowledge, or help specify the problem at hand or to present results. An important challenge will be the definition of interactive methods specifically tailored for non-specialists (ex: visual analytics, justification and explanation). These methods will lead to a major breakthrough in the practical exploitation of agricultural Big Data. A precise framework for the evaluation of results will be set up along the project duration, based on statistical model validation, the agronomist expertise (INRA, ACTA) and collaboration with other domain actors. Such a framework represents a resolutely innovative aspect in the data-mining and statistical analysis domain. Assessing current results on agricultural time scales (seasons, years) will allow us to continuously improve the proposed methods.
|Leaders: Alexandre Termier (INRIA, on the right in the pic), Pascal Poncelet (University of Montpellier)|
|Units: LACODAM (Rennes), ZENITH & GRAPHIK , LIRMM, IATE, MISTEA , PEGASE|
|Disciplines: Primary Data sciences, Computing sciences; Secondary Agronomy (knowledge on data quality issues; knowledge extraction utputs); Social sciences (human-centered design of processing/visualisation methods).|
Exploring new ways for model integration/qualification
Over the years, tens of agronomic models have been developed by agronomists, and some are already exploited in Decision Support System tools (DSS). However, new tools, eg. high-resolution sensing (Axis 3) and new data-driven knowledge (Axis 5), are opening new opportunities for major evolutions in model construction approaches. The scientific objective of this theme is therefore to improve the applicability of dynamic models by coupling models of different natures and different scales, and by using real time data assimilation. The originality here lies in the multiscale (from the plant organ to the small region) and the interdisciplinary approach, which is central, eg (i) in the coupling of “agronomy/ ICT/ innovation sciences” for the elaboration of expert-based models (based on participatory sciences), for the collaboration of expert-based / data-driven / deterministic models, to create agronomy-economic models for decision making, or (ii) in the “agronomy/ ICT” coupling for phenotyping integration in models.
Models are an essential component of digital agriculture, as they can be usede to turn “data” into “information”, “diagnosis” and “agricultural advices”. They are the basis of precision agriculture/ livestock farming and Decision Support Systems (DSS). An overhaul of our current production and impact models is urgently required: models need to better represent and take into account the plant interaction with its environment, multi-stress effects:impacts and extreme climatic events, the impact of alternative field and farm management strategies... Today the main bottleneck to modelling/ simulation is the lack of applicability of dynamic models. Our objective is to improve it by coupling different types of models, and by using real time data assimilation, at various spatial (from farm plot to regions) and time scales. First, the genetic asset has to be considered: genotype-to-phenotype models (ie models who link the genes with plant behaviour in real environment) will be informed by the outputs of high-throughput phenotyping methods. To this end, modular modelling solutions are favoured to easily add and share new models with explicit genetic information. Second, the operational use of next generation sensors in agriculture (IoT, UAV…) creates the need for new methodologies (e.g. spatio-temporal data analysis, including advanced statistical inference) for data integration in models (including seasonal weather projections) to produce more accurate and predictive information in real time. There is also a need to enhance the development of management/decision models based on such data. Production models should be coupled with economic models for strategic decision-making, Fuzzy logic, multi-objective optimization, argumentation etc…) methods are needed for multi-criteria evaluation. Third, for production system assessment, models should predict a larger array of responses (services) and trade-offs than they currently do. Last, modelling research should also address more transversal issues, e.g. quantification of model uncertainty and model integration. Uncertainty quantification is demanded by tactical decision-makers and by policymakers. Finally, models that integrate different sources of information (both qualitative and quantitative) and of knowledge (expert-based knowledge, data-driven knowledge…) are required for decision-making.
|Leaders: Patrick Taillandier (INRAE, MIAT), Pierre Martre (INRAE, LEPSE)|
|Units: SYSTEM, MISTEA, LEPSE, ITAP, ACTA, AGIR (Toulouse), MIAT (Toulouse), PEGASE (Rennes), TETIS, AIDA, AMAP, HORTSYS, GECO, GRAPHIK|
|Disciplines: Higly interdisciplinary with Agricultural sciences, Mathematics, Artificial Intelligence, Computer sciences|