Digital plants: Models, Analyses and Data, from organs to ecosytems
Heads : Jean-Baptiste DURAND (CIRAD), Hervé GOEAU (CIRAD)
Challenges
The understanding and construction of answers to research issues and societal questions related to ecology, environment, biodiversity and agronomy rely jointly on mathematical and numerical approaches through the construction of descriptive, mechanistic and data models fed by : (i) the modelling of complex processes involved in the dynamics of these systems; (ii) the rapid development of the massive exploitation of data from these systems (sensors, Big Data, computing capacity, AI, etc.), their processing, analysis and representations. In this context, the ability to develop formal and numerical approaches and methods to enable a better understanding of short and long-term dynamics, to make diagnoses, and even prognoses, is a major challenge. In particular, this capacity must make it possible to: determine, using expert knowledge, or extract the relevant variables by analyzing data acquired in situ or by remote sensing; process measured data using statistical or Artificial Intelligence approaches; cope with data scarcity; use data to feed models (mathematical and numerical) or to inform model "target values".
Objectives
Our objective is to build and validate generic and quantitative approaches based on expert knowledge, field expertise and experiments, and mechanistic processes, to tackle complex issues. Faced with voluminous, distributed, heterogeneous and/or scattered data, to specify, we construct and instantiate a range of approaches adapted to the representation, identification, characterisation and quantification of characteristic markers in the understanding of the plant, from the organ to the stand, and even beyond.
Main topics of interest
- Portability of Machine Learning approaches to unstructured data or metadata
- Constraints of mesoscale approaches -allowing the transition from one scale to another- and identification of emergent properties of systems during the change of scale
- Development, analysis, and implementation of heterogeneous approaches (e.g. discrete/continuous) for modelling complex plant world systems
- Development of hybrid approaches (data models and mechanistic models)
Approach
We mainly mobilize two approaches:
- The development and adaptation of theoretical models, discrete or continuous, possibly spatialized, allowing descriptions of arrangements (patterns) or mimicking processes, on different spatial and temporal scales, in order to take into account the diversity of the systems studied, using generic and theoretical approaches (MAGNET research topics) or more targeted approaches, such as structure-function modelling of plants (FSPM research topics) ;
- The implementation of Machine Learning approaches, with a particular emphasis on the issues of constitution of training datasets and their validation (BIAS research topics).
One of the specificities of AMAP is to address a wide range of scales from organ to landscape (using for example meta-modelling approaches), giving elements of understanding for key processes about the systems studied. We also place emphasis is also placed on the qualification of data, models and outputs via statistical or specific approaches.
Expected results
Our fundamental and applied research in Mathematics and Computer Science for the study of (agro-)ecosystems is expected to provide new insights, tools and methods for the optimisation of systems, especially agronomic (in terms of yields, water efficiency, phytosanitary protection, etc.), and large-scale forecasting of the spatio-temporal dynamics of vegetation. Our innovative mathematical and numerical tools, methods and approaches are expected to be applied to answer the major challenges tackled by the two other research lines in AMAP (Biodiversity and Biomass) and our partners in the North and, in the South.