Search

International audience; L'utilisation des Systèmes Multi-Agents (SMA) pour modéliser, selon une approche ascendante (bottom-up), des systèmes complexes où un grand nombre d'entités interagissent de façon non linéaire (e.g. écosystèmes, insectes sociaux, foules, trafic routier, etc.) est un domaine applicatif qui a largement fait ses preuves, ainsi qu'une source d'inspirations qui permet de faire évoluer ce paradigme [5]. Dans le cadre de l'étude des systèmes complexes, la diversité des approches multi-agents existantes témoigne à la fois de la richesse de ce paradigme mais aussi de la complexité liée à la conception et à l'utilisation des simulations basées sur cette approche. Il existe, en effet, de multiples façons de modéliser la dynamique d'un SMA, sans qu'on puisse pour autant hiérarchiser la qualité des modèles tant leur dépendance au contexte et aux objectifs de l'expérience est forte. Enfin, l'augmentation des capacités de calcul ainsi que la qualité des outils existants ont permis une envolée spectaculaire du nombre de simulations multi-agents, ainsi que la complexité qu'elles prennent en compte en leur sein, de telle sorte qu'on peut véritablement parler d'un foisonnement d'approches.

Introduction Agroecological studies dealing with genotype by environment by management interactions generate heterogeneous datasets difficult to gather, store, share and analyse. Modelling is an essential tool for designing and evaluating innovative sustainable cropping systems. Data integration, sharing and reusing for crop modelling need a good data management. CIRAD has developed the AEGIS (AgroEcological Global Information System) platform, which aims to sustain the data value chain {Curry, 2015) for agroecological studies. This paper present the platform overall organisation and the approach used to ensure the data-model continuum. Materials and Methods Good data management is based on two key principles. The first is Data Lifecycle Management (DLM), which consists of managing data throughout its life cycle from production to use. The second is the FAIR data principles that aim to make data findable, freely accessible, interoperable and increase data reuse. Compliant with these standards, the AEGIS platform is organized into four pilars that focus on the steps of data acquisition, processing, sharing and enhancement. Data acquisition is a process of gathering, describing and harmonizing data. AEGIS integrates a generic ®ECOFI database (Auzoux et al., 2017) using metadata technology that allows any type of data to be easily imported according to a collection process and uses a variable dictionary to facilitate the annotation of data making them understandable to all. Data processing is about making the raw data acquired easily usable for analysis and modelling. Through integrated dashboards, AEGIS offers a real-time overview of all stored data, ranging from raw data to indicators for assessing the sustainability and performance of agroecosystems. It provides homogeneous datasets for crops models simulations and capitalizes processed data, which can be simulation parameters and outputs, analysis results and performance indicators. AEGIS proposes data visualisation tools that highlight patterns and correlations inaccessible from the raw data. In the context of open data, AEGIS ensures data sharing increasing the impact and visibility of agroecological studies, promoting potential data reuse for modelling. Interoperability is considered as a necessity for data sharing and involves four levels of data exchange: system, syntactic, structural and semantic. AEGIS uses ontologies dedicated to the plant, pest, environment and cultural practices to ensure compatibility with data from other platforms that comply with these ontologies. Datasets provided comply with metadata standards such as Darwin Core, MIAPPE and EML. AEGIS is able to export data using the standard open Breeding Application Programming Interface (BrAPI). By ensuring data standardization, optimization and curation, AEGIS enhance data in term of publishing quality, accuracy in decision-making and financial value creation. Results and Discussion Such coupling of cropping system models allows performing three kinds of fundamental works in the present agroecological studies: (i) model validation, (ii) parameters estimation, and (iii) cropping system models comparison. In the first work {Chaput et. al 2019), that aims to test the ability of a crop growth model {STICS) to simulate the sugarcane growth response to different climates, soils and nitrogen management. The STICS model was calibrated using the observed data provided by AEGIS that illustrates more than 10 years of sugarcane trials in Reunion. In the second work {Christina et. al, 2019) that aims to model the annual variability of sugar cane yield in Reunion Island, AEGIS is the scheduler of the estimation process until the RMSE error is minimized. It asks to a simulated annealing algorithm to generate a new set of input parameters. These parameters are used to set up simulations of cane growth MOSICAS model thanks to the generic ®ECOFI database and output variables are compared to observed values stored within AEGIS. In the third work, that aims to study on sugarcane dealing with genotype, environment and management interactions as part of the "International Consortium for Sugarcane Modelling" {ICSM), AEGIS has been used to setup, launch simulations of STICS, MOSICAS, and DSSAT models, and compare simulations regarding observed values stored within AEGIS. Conclusions AEGIS is a platform that participates in the construction of a data repository characterizing agroecosystems for integrated multi-scale analysis and ensures data consistency through harmonized reference systems and procedures. It provides all the features of the data value chain paradigm. It ensures a formal coupling (parametrization and output analyse) with the most well-known and ad hoe cropping systems models. It has been used in European and international projects involved in the agro-ecological transition as a steering and decision-making tool.

The agricultural recycling of organic residues (OR) represents an important level in agro-ecological transitions. ORs are used to provide the nutrients needed for crop growth and to improve soil properties, as an alternative to mineral fertilizers. In Réunion, sugarcane represents 58% of agricultural area and offers opportunities to promote OR recycling. We tested the ability of a crop growth model (STICS) to simulate the sugarcane growth response to different climates, soils and nitrogen management. The STICS model was calibrated using the ECOFI database that included more than 10 years of sugarcane trials in Réunion. After calibration, the model was tested using the TERO network. In this network, four trials were set-up in contrasting soil and climatic conditions to assess the growth response of different cane cultivars to increasing levels of N (as urea and organic fertilisers). Despite being only recently applied in tropical areas and in sugarcane cropping systems, the STICS crop model accurately simulated growth in response to soil, climate and fertiliser variability. In the context of a circular economy, the STICS model could be coupled with a territorial model of OR management. This approach would make it possible to evaluate management strategies for ORs and to identify optimal areas for applying them.

Faced with increasing environmental, economic and social challenges, sugarcane industries are adopting agroecological approaches to design and evaluate systems that use natural resources more efficiently, mobilize plant biodiversity and adopt agroecological practices. In order to set up this agroecological transition, stakeholders of sugarcane industries need to: (i) access and analyze raw data; (ii) capitalize and share knowledge through professional networks; (iii) define performance and impact indicators; and (iv) engage in learning processes to acquire new skills based on successful experiments. CIRAD developed AEGIS (AgroEcological Global Information System), a platform to support digital agriculture and successful agroecological transition. AEGIS can provide standardized, harmonized and organized data that come from various sugarcane agroecosystems. Data stored in AEGIS are collected at different spatial and temporal scales, from different experiment designs and protocols, and in different contexts (agronomy, ecology, sociology, and economy). AEGIS meets the expectations of stakeholders through the development of generic statistical analysis tools and the implementation of ex-ante and ex-post data processing methodologies. It provides datasets for simulation of crop models and complex visualization tools to facilitate the interpretation of data and to highlight indicators, patterns and correlations inaccessible from raw data. AEGIS uses ontologies, metadata standards and web services, which ensure the semantic and technical interoperability of the various components of the information system. These features allowed development of a common language for sharing and exchanging contextualized information between stakeholders, whatever their fields of activity. By integrating dashboards, statistical analysis tools, data processing tools (data mining), simulation and visualization tools (artificial intelligence), our platform is a complete steering and decision support tool in the context of the agroecological transition.

Pollination of young palm trees in Asia is mainly due to an introduced wee- vil, Elaeidobius kamerunicus. The pollinators congregate and multiply only on male inorescences in anthesis (during ower opening). Then, loaded with pollen grains, they may visit female owers and pollinate them, more or less, e_ectively. However, this entomophilous pollination is not always su_cient to have a good fruit set. In particular, because the density of male inorescences per hectare is often low in (young) plantations. That is why it is important to study and understand the mutualistic interactions between the inorescences and the weevil population, despite the fact that we have partial knowledge. The aim of this talk is to present a mathematical model and to compare it with an Individual-Based approach [1]. Using the qualitative analysis of the mathematical model, and numerical simulations, we will discuss the main out- comes of both models that may help us to elaborate new observations, new experiments, or to understand how to sustain this mutualistic system. Finally, based on old published data [2], we try to estimate the mean number of male inorescences per hectare necessary to maintain a population above a certain threshold in order to reach a good fruit set.

Face aux enjeux de compréhension des écosystèmes marins et de gestion des usages marins, les modèles complexes se révèlent des outils pertinents pour tester les modifications induites par le changement global, anticiper des évolutions des socio-écosystèmes marins et aider à la sélection de stratégies de gestion. Construire un modèle numérique et faire des simulations est une chose, mesurer la confiance des sorties du modèle en est une autre. Une étape indispensable dans l'usage des modèles numériques est la confrontation des sorties du modèle aux observations du système modélisé pour caler le modèle. La sélection de stratégies de gestion et la calibration sont deux finalités de l'optimisation. Les problèmes d'optimisation en modélisation halieutique sont le plus souvent complexes avec des caractéristiques mathématiques diverses. La fonction à optimiser peut être déterministe ou stochastique, avec ou sans contraintes, à une ou plusieurs dimensions. Le nombre de paramètres à optimiser peut varier de l'unité à plusieurs centaines et le coût informatique peut induire de fortes restrictions sur le nombre de simulations réalisables avec le modèle, d'une centaine à quelque milliers pour les moins coûteux. Aucun guide pratique n'est disponible dans la littérature pour mettre en oeuvre une optimisation rigoureuse avec un modèle complexe. Nous proposons ici une démarche d'optimisation articulée en 3 étapes (prétraitement, choix de l'algorithme et post-traitement), basée des outils et méthodes existants et dont la réalisation peut être non linéaire. Ce guide inspiré d'une analyse des expériences d'un groupe de modélisateurs ouvre des pistes de recherche pour pallier aux difficultés, aux autocensures et frustrations des modélisateurs.

The sorghum plant can produce great amount of stem biomass due to the elaboration of one main stem made of thick and large internodes possibly associated with few tillers, depending on the genotype and the environment (GxE). Depending on GxE as well, Carbon (C) is differentially allocated to stem internode structural (lignocellulosic) and non-structural (sugar) components, making stem biomass appropriate to diverse end-uses (feed, energy, bioproducts...). Then optimizing C resource acquisition and allocation to stem component sinks is a major challenge for sorghum breeding toward the conception of plant ideotypes. By accounting for traits involved in C source-sink relations in a dynamic way and at the organ level, FSPM can be of major interest to support the analysis of biomass sorghum phenotypes. EcoMeristem model is designed for this purpose. It simulates plant and crop performance (biomass, sugar, grain) as the result of GxE acting on C sink activity (organ size and number, dynamically set-up in plant topology) and their regulation by their competition for a common pool of C resource within the plant, computed in a simple way at crop level (Beer-Lambert for light interception, Monteith for its conversion into C). This model, initially developed for rice, was recently tested for its ability to capture the phenotypic variability met across eight contrasted genotypes of biomass sorghum. Whereas tiller number, stem and leaf biomass at whole plant level were correctly modelled, biomass partitioning between the main stem and tiller(s) was unreliable: nonstructural (C stored) and a minor extent structural biomass of the main stem was reduced to the benefit of tiller stem growth, suggesting some limits in the way C source acquisition and/or C sink rules of dominance among culms. The aim of the present study was to evaluate concepts for representing C source-sink relationships in EcoMeristem and better capturing the phenotypic variability of biomass sorghum. For this purpose, the current concepts used in this model were compared to more detailed approaches enabling to differentiate the culms with respect to C source and/or sink activity. Regarding C source acquisition, a light interception model taking into account canopy closure dynamics (from isolated plant to row and closed canopy) and horizontal canopy layers defined by leaf age, temperature and sunlit/shaded ratio, was implemented. It was combined with a photosynthesis model inspired from Farquhar-von Caemmerer-Berry's model to compute C assimilation at leaf level. These modules together enable to compute C supply either at whole plant or culm level. Regarding C sink activity, an exploratory formalism was implemented that prioritized C sinks (i.e. growing organs and C storage in internodes) for their access to C resource, according to the age of the culm they belong to (proxy of apical dominance). These different C source and sink related concepts are currently compared to a dedicated field data set. This data set deals with two genotypes and tiller pruning treatments to evaluate the competition between main stem and tiller growth. Accordingly, these source-sink concepts are being benchmarked regarding their ecophysiological relevance and computation time efficiency. Results will be presented and discussed with respect to the added value of each approach for analyzing biomass-sorghum phenotypes.

Sorghum genetic diversity offers the opportunity to develop multipurpose genotypes combining high grain, ligno-cellulosic biomass and/or sweet juice productions. It provides a large spectrum of adaptive traits, particularly to drought, essential for ensuring production stability in an increasingly fluctuating climatic context. Sorghum is thus a crucial crop to breed for and contribute to meet future expectations regarding food, feed and bioenergy productions, while minimizing resource and land use competition. Accordingly, ideotypes must be developed combining appropriate developmental, morphogenetic and biochemical traits depending on targeted cropping contexts (eg. dedicated vs. intermediate crop, wateravailability) and end-uses (eg. energy, feed). Trait impact on plant phenotype construction must be first quantified, considering possible linkages and trade-offs among them that can appreciably modulate their respective benefits depending on the genotype and the environment. With this respect, ecophysiological modelling plays an original role since, by formalizing experimental knowledge on individual processes, it helps analyzing their trade-offs and impacts on resulting plant phenotypic variability, across genotypes and environments. Supported by appropriate mathematical tools, it can help quantifying elemental traits within a range of genetic diversity, in away not accessible experimentally, toward their genetic study and the in silico optimization of their combination for a given context. This approach is currently set-up in two companion projects aiming to develop multipurpose sorghum for both Mediterranean and semi-arid, drought prone environments. Field trials were carried out to identify traits controlling vegetative biomass production, composition and expressing variability across genotypes and water situations, at tissue, organ and plant level. Accordingly, the plant growth model Ecomeristem is progressively adapted to capture key processes controlling such phenotypic variability. Recent progresses and future developments will be presented, highlighting how modelling enables to analyze trait compensations and impacts on grain and biomass production and finally define ideotypes depending on targeted contexts. (Texte intégral)

The fit of rice genotypes and copping calendars into an agro-ecological niche is largely a function of hydrology and the thermal environment. Although rice can be grown in very diverse climates, a given genotype has a narrow band of thermal adaptation. The most sensitive processes are development per se and reproduction, namely the microspore stage (chilling) and anthesis (heat). The crop-generated microclimate must be taken into account because organ temperature can differ greatly from ambient temperature. A multi-environment study in Senegal (hot-dry and cool-dry seasons), Philippines and southern France was conducted to observe and model microclimate (water, canopy and panicle temperatures, the latter by IR imagery), phenology, the flowering period, time of day of anthesis (TOA) and spikelet sterility. Water and panicle temperatures differed strongly from air temperature, both being by up to 10°C colder than air at midday under dry atmospheric conditions. TOA occurred in the morning or early afternoon, depending on night temperature and VPD, but had a constant duration of 2 h. The plasticity of TOA was adaptive because it helped anthesis escape from very hot or very cold conditions. Together, transpiration cooling (heat avoidance) and plasticity of TOA (temporal escape) constitute powerful coping mechanisms in rice for extreme temperatures. The simple crop model RIDEV V2 (V1 was a simpler tool in the 90s) was developed to predict phenology and thermal sterility of rice while taking into account microclimate. RIDEV can be used as a predictive tool for agronomy and agro-ecology (e.g. climate-change impacts), but is also equipped with a powerful parameter-optimization tool for reverse modeling approaches (heuristics). The latter was applied to phenotype phenological traits (photothermal constants of genotypes) for 230 diverse rice accessions studied in Senegal and Madagascar in the ORYTAGE (Diversité des caractères d'adaptation aux contraintes hydriques et thermiques chez le riz) project. Future uses of RIDEV in rice genetics and agronomy are discussed in the paper, namely for phenotype- genotype association studies (reverse mode) and mapping of climate-change impacts on the rice crop (forward mode).

International audience; The OpenAlea platform (Pradal et al., 2008) was designed to facilitate the integration and inter-operability of heterogeneous models to get comprehensive FSPMs. It relies on Python gluing capabilities, that allow non intrusive integration of programs written in various languages (Fortran, C, C++, R, L-system); and on the dataflow computing paradigm, that promotes decomposition of applications into independent components that can be recombined dynamically into customized workflows. Still, a plugable collection of components is not by itself a solution to the modularity problem in FSPM modeling. First, heterogeneities between components inputs and outputs can lead to exponential needs for specific adaptors and converters to get functional assemblies. Second, several ways exist to decompose models into independent components. This can lead to incompatibilities or difficulties for re-assembly into comprehensive models. Last, users of the platform may find difficult to build applications, without some knowledge on how a simulation has to be reasoned within the data-flow computing paradigm. Here, we propose a modeling strategy to help for building coherent, yet modular FSPM under OpenAlea. We first define the key concepts of this strategy, illustrate how they can be used under Visualea and how it lead to a first set of reusable components resulting from various ecophysiological studies.

In mature palm plantation, the number of harvested bunches is the main component of yield. In Ivory Coast it varies from 600 to 1600 ha-1 year' with large seasonal variations from 2 to up to 300 bunches ha' month-' for L2T x D10D material. Phenology of adult oil palm is complex. The apical meristem initiates new leaves (20 to 30 per year) with a constant plastochron. Leaf lifespan from initiation to senescence is more than 4 years. The first unopened leaf emerging from the crown is spear-leaf (rank 0). Approximately 60 developing leaves are hidden in the crown (ranks -1 to -60) and 40 autotrophic expanded leaves are visible (Corley and Tinker 2003). Each leaf bears a male or female inflorescence which is first visible at flowering (rank 18 to 20). Bunch is harvested 5 to 6 months after flowering. By dissection of the crown and electronic microscopy, the inflorescence primordia can be seen at rank -40 but the sex cannot be identified before rank 4 (Adam & al. 2005). The determinism of gender is highly dependent of environment and physiological stress but is not explained. The time and duration of sex differentiation are not known with certainty: it can be assumed at spikelet initiation (rank -5 to -10) or at bract initiation (rank -20 to -30) or at any time between rank -10 to -30 (Jones 1997). Abortions occur between ranks 8 to 12, but, because some leaves have no inflorescence, an ealier sensitive phase may occur. Because of weak evidence and of the difficulty of experiments, crop modeling is a way to test hypotheses. For example, in a model simulating seasonal trends at tree level, Jones (1997) introduced a feedback loop of bunch load to induce physiological stress in earlier stages responsible of abortion or sex differentiation.

The ANR DELICAS project will address the problem faced in sugarcane for identifying molecular markers with a strong effect on traits of agronomic interest. The genetic control of cane and sugar yield traits in sugarcane typically involves numerous QTLs with small individual effects (Hoarau et al. 2002; Aitken et al. 2006). The lack of markers has been so far one of the most important constraints to a wider contribution of genomics to sugarcane breeding and for the development of marker assisted selection in this crop. Phenotypic traits for production potential interact strongly with the environment, and represent the outcome of multiple processes difficult to measure and to tag at the genetic level. Plant growth modelling can describe yield formation dynamically as a set of interactive equations using only a small number of genotypic parameters. The different parameters participating in phenotype expression can be considered as synthetic component traits, presumably controlled by fewer genes than the complex agronomic traits and can be expected to be closer to gene or QTL effects, in the sense that Genotype x Environment `noise' is reduced (Hammer et al. 2002, Dingkuhn et al. 2005). In such a `heuristic' approach, parameter variation among genotypes can be interpreted as an expression of allelic diversity, and analyzed accordingly in QTL or association studies. Model assisted phenotyping has been applied to maize (Reymond et al. 2003) or peach tree (Quilot et al. 2004). The DELICAS project, which associates the breeding company eRcane and Cirad, aims at identifying molecular markers associated with genes implied in the elaboration of sugarcane yield by using ecophysiological models. To achieve the objectives of the project, we will: (i) elaborate methods and tools for model assisted ecophysiological phenotyping of sugarcane, and (ii) phenotype a core collection with two ecophysiological models, Mosicas (Martine, 2003) and EcoMeristem (Luquet et al. 2006), and detect associations between genetic markers and model parameters.

Climate change scenarios are predicting, by the end of the century, an increase in air temperature from 1.1 to 6.4 °C and in air [CO 2] from 600 to 1500 ppm, associated with more frequent submergence and drought events in the rice-growing regions. Considering that rice is a staple food consumed by 3 billion people, it is essential to predict the impact of such climates on rice production and select genotypes adapted to future environments. The lack of (i) compatibility between climatic and crop models, (ii) characterization of future target population of environments, and (iii) formalization of interaction between climate factors on plant morphogenesis, gives crop modeling a central role for addressing these challenges. The crop model SARRAH has been successful in identifying relevant sites for rice breeding programs and matching plant types with environment characteristics in Brazil. This model was also applied for predicting crop productivity of distinct cereals in many villages in four Western African countries by determining the part of climate and rainfall involved in grain yield variability. The crop model EcoMeristem, designed to account for the effect of environmental factors on plant morphogenesis at the organ level, is already used as a phenotyping tool for rice under drought. It will soon formalize the effect of the microenvironment on organ temperature and grain yield formation. These two models are already relevant tools to address crop adaptation and response to climate change. In the near future, they will allow conceptualizing and evaluating ideotypes under various climate scenarios.

To sustain the future world demand in sugar, bioenergy and biofuels, sugarcane crops with higher productivity are needed. This relies in first place on the creation of varieties with higher yield potential. New approaches exist that use known linkages between genome regions and agronomic traits of interest (Yin et al., 1999). In this respect, the DELICAS ANR project aims at identifying molecular markers associated with genes involved in the elaboration of sugarcane yield (Nibouche et al. poster, same session). The success of such an approach relies on the identification of simple process based traits constituting yield formation, more simple genetically and less prone to genotype by environment interactions (GxE, Hammer et al., 2005). As growth models mimic dynamically elemental processes of yield formation, related process based parameters could be considered as finer phenotypic traits and used in a phenotyping approach. Since a few years, model assisted phenotyping has been used on various plants with simple models (Yin et al., 1999). On Sugarcane, modelling studies mainly dealt with crop growth model (Canegro, Apsim, Qcane, Mosicas) applied to very few varieties in not too contrasted E. This did not provide to date a clear view of the genetic variability of their crop parameters. Also, O'Leary (2000) and Singels et al. (2005) underlined the need to explore the genetic variability that can be accounted for by the parameters in existing sugarcane crop models, this, to consequently adapt those models to support breeding purposes. The objective of this study is to determine the genetic and environmental variability of the crop parameters of two dynamic sugarcane crop growth models applied to 18 sugarcane genotypes in two contrasted environments in La Réunion Island: 1/ Mosicas [ Martiné & Todoroff, 2002] a population level model used widely for agronomic purposes, 2/ and Ecomeristem [Luquet et al., 2006], simulating rice and sorghum plant growth in its stand and adapted to sugarcane in the context of DELICAS project(see Luquet et al. in this session). This study being underway, this paper gives an overview of applied methodologies and first experimental results regarding elemental processes of canopy development.

Early vigour is essential for rapid crop establishment and resource acquisition (light, water and nutrients), and by consequence for competition with weeds (Caton et al. 2003). Rice is known to be a poor competitor, particularly in poorly controlled systems prone to weeds and abiotic stresses. Crop modellers consider early vigour a source-acquisition driven process, consisting of intercepting light and converting it into biomass, some of which is partitioned to leaf area growth (Brisson et al. 1998). Recent works suggested that sink dynamics may be a driving force of resource acquisition and thus, early vigour. Luquet et al. (2006) presented a rice growth model (EcoMeristem) that simulates two-way interactions between growth and development processes. Organogenesis and morphogenesis thereby drive structural growth and light capture, mobilization of transitory reserves and, to some extent, affect leaf photosynthetic rates by feedback. According to this concept, early vigour of the rice plant depends on development rate (DR, inverse of phyllochron, in °C.d-1), tillering capacity and potential leaf size which together constitute incremental demand for assimilates. The present study aims to test this hypothesis for rice, model plant for branching cereals. Experimental and modelling approaches are combined.

International audience; no abstract

Traits related to crop performance (sugar, grain, biomass) are complex and their genetic bases difficult to study. To take advantage, in the process of selection, of tools now provided by molecular genetics, these traits need to be 'dissected' into elemental processes, more simple genetically and less prone to GxE (genotype X environment) interactions (Dingkuhn et al. 2007). However, evaluating a component trait individually is meaningless as plant and crop performance is the dynamic result of physiological, and sometimes genetic, linkages between traits in a given environment. This represents today a great challenge in the context of plant phenotyping and selection. Modelling enables formalizing biological processes as equations with G-dependent parameters, potentially linked to genetic information (eg. QTL). The use of model parameters as phenotypic traits for genetic analyses already showed its relevance in the case of simple models (Reymond et al. 2003; Yin et al. 1999). More recently, a few studies demonstrated the interest of using more integrative plant models for discriminating genotypes, based on parameter values within component, process based modules (Lecoeur et al. 2008; Luquet et al. 2008). Among them is Ecomeristem (Luquet et al. 2006), model simulating cereals' crop performance based on a simple formalism of plant vegetative morphogenesis, i.e. of meristem activity and its GxE regulation (organ creation, size, etc). Its use for QTL detection in a rice mapping population (based on parameters related to tillering, leaf size, phyllochron) gave even promising preliminary results (Ahmadi et al. 2008). DELICAS is a project that aims to identify, for sugar cane, molecular markers associated with genes of yield formation (Nibouche et al. 2010). Crop modelling is intended to support the phenotyping of component processes. In this context, Ecomeristem is currently adapted to sugar cane in order to take advantage of its plant level approach of morphogenesis and source-sink relationships. This paper aims to present this new model version, before it is tested and compared to other modelling approach(es) for its discriminative value (Martiné et al. 2010).

Tropical japonica rice is frequently exposed to drought in upland ecosystems. Its drought tolerance is poor compared to other crops but the group has great genetic, in part unexploited diversity in adaptations. Exploring the japonica's phenotypic and genetic diversity for drought tolerance is thus crucial to breed rice for drought prone environments. Under such conditions, yield depends on drought timing and intensity along plant phenology. During the vegetative phase, drought affects vigour (leaf number and area, tillering, roots) and therefore resource acquisition. Plant adaptation to drought is a complex mix of physiological tolerance, phenology and morphology, all of which interact with each other and with environment, resulting in phenotypic plasticity. This involves variable regulation of source-sink relationships during morphogenesis (phyllochron, organ expansion, tillering), leaf senescence and transpiration. Studying this system requires dissection into simpler traits involving a smaller number of genes/alleles. However, trait dissection must also account for Genotype*Environment (GxE) interactions and trait plasticity. This is particularly difficult for process based traits that cannot be measured directly. In this context, modelling is relevant if used to dissect a complex system into elemental processes. Each process can be formalized as a response function, with parameters seen as being analogous to genes. The objective of this work is to explore the added value of using dynamic whole plant modelling to assist phenotyping plant response to drought, as a basis for a genetic association study. This work focuses on rice plant transpiration and morphogenesis processes evaluated on seedlings of a diverse sample of tropical japonicas. A greenhouse pot experiment was conducted at Cirad, Montpellier, on 203 japonica accessions with three replications and two treatments (irrigated and drought). Drought was imposed by dry-down from leaf-6 appearance until a targeted stress level was reached, as indicated by Fraction of Transpirable Soil Water (FTSW). FTSW and plant transpiration rates were monitored gravimetrically. At the same time a minimum set of morphological plant descriptors and climate were collected, in order to calibrate, for each genotype and in both well watered and stressed conditions, the corresponding modules of EcoMeristem plant growth model . This paper presents first results and discusses the discriminative power and the added value of model assisted phenotyping for the case of rice drought responses. (Texte intégral)

Climate change and variability (CCV) exposes tropical crops as rice to heat and drought. Increasing atmospheric CO2 is expected to improve plant transpiration efficiency but benefits may be more than offset by reduced transpirational cooling and accelerated phenology. Yield is generally affected by environment effects on morphogenesis, particularly to stress during reproductive sink capacity determination. Exposure of reproductive processes to heat and drought, in turn, depends on plant structura1 development and resulting microclimate and water availability. These highly genotype-dependent interactions make it difficulty to predict CCV impacts and design adaptations. An important step is to model this system, particularly interactions between plant morphogenetic and phenological processes with climate and resources, and resulting microclimate within the crop stand. Models are needed that consider crop structura1 development at organ level, while providing sufficient phenological and physiological detail to situate stress sensitive processes within time and canopy. Such a model must be coupled with a heat balance providing accurate information on soil, floodwater, leaf and panicle temperature. Key physiological processes would thus become predictable, including: tillering and tiller maintenance/abortion, leaf area dynamics including senescence, spike number dimensioning and adjustments, stem carbohydrate accumulation and mobilization to grains, thermal and drought induced spike sterility determined at the sensitive microspore and anthesis stages or by panicle exertion limitation, and finally grain filling process. A new type of functional-structura1 plant models is needed that integrates environment dynamics within soil-water-plant-atmosphere continuum. EcoMeristem model was designed to simulate environment and genotype driven phenotypic plasticity for rice and other cereals. It simulates rice plant morphogenesis at organ, plant and canopy levels in response to drought and climatic (excluding C02) factors. The key concept is the feedback of trophic status (source-sink and competition among sinks) on organ initiation and (pre-)dimensioning processes, through signals to the meristem. A state variable quantifying internal competition for assimilates constitutes this signal (Ic: supply-demand ratio), which also governs resource and stress feedbacks on senescence processes. Water deficit is described by Fraction of Transpirable Soil Water (FTSW) and derived physiological coefficients. The sensitivity of development vs. trophic feedbacks is set by genotypic parameters (threshold and slope parameters, e. g. for tillering response to Ic or leaf expansion, assimilation and transpiration rates to FTSW). EcoMeristem was developed to - explore phenotypic plasticity concepts as affected by abiotic factors (drought, T ... ), - explore ideotype concepts for specific environments, and - measure heuristically hidden (process based) traits/parameters within phenotyping context. A new objective is to study rice varietal response to CCV (CIRAD, NIAES, IRRI, WUR collaboration). EcoMeristem was therefore recently linked with 3D visualization tool (OpenAJea), opening the way to spatial micrometeorological computations within the canopy. Further model developments are needed: - Completing the model for all developmental stages, - Extending the water balance to flooded and mixed flooded/aerobic systems, - Introducing C02 effects on plant gas exchange, - Introducing a stratified heat balance for the soil-water-plant-atmosphere continuum. This paper presents the current state of model development and applications, and future improvements for research on rice crop adaptation to CCV. (Texte intégral)

Seasonal variation of fresh bunch production is difficult to explain and predict. A population scale model, EcoPalm, was developed to explain those using new physiological hypotheses. A central hypothesis was that day length influences time of flowering during a photosensitive period (PSP) at the scale of individual fronds (phytomers). The variable PSP duration between successive phytomers result in a queue of phytomers in "waiting state", which partly explains seasonal variations. The monthly, potential number of harvested bunches, based on a 100% sex ratio, and the absence of abortions and failures of fruit set, was simulated. The second hypothesis was that sex ratio, intensity of bunch abortion and bunch failure, , depend on the trophic status of a given phytomer during sensitive phases of its development, expressed through a plant-scale Index of Competition for assimilates (Ic), which is the ratio between instantaneous assimilate supply and aggregate demand. Annual demand for vegetative organ growth was assumed to be constant (stabilized growth). Bunch dry matter production was assumed to be variable. If Ic < 1 during sensitive stages of bunch development, (high competition among sinks), sex ratio was considered to be low and abortion and bunch failure high. If Ic > 1, sex ratio was high. The 3rd hypothesis was that resulting sink-source imbalances are buffered by stem carbohydrate reserves. If Ic > 1, demand was not down-sized and excess assimilates were stored (overflow model). If Ic < 1 reserves were mobilized. The model was tested on a multi-annual period in La Mé (Ivory Coast) on a control progeny, LM2T X D10D. Sensitive periods to Ic were adjusted by optimization of parameters using evolutionary algorithm method. The PSP occurs at the early stage of development of phytomer, begin 36 months before harvest (mbh) and ends 28 mbh. The sex determination is very complex and occurs during PSP. The initial sex ratio of palm population was determined by mean Ic from nearly 34 to 28 mbh and the last month a second phase increases sensitivity to low Ic. This sexualization phase ends with PSP. A first stage of abortion is founded 17 mbh (frond emergency). A second stage of abortion occurs from 10 to 8 mbh. The model shows good monthly and annual simulations of bunches number harvested by ha at La Mé station. Validations at sites in Indonesia are in progress. Other model features including simulation of oil production and impacts of drought are presented in subsequent papers.