Search

International audience; In a context of pluralization, sophistication, and combination of formal models, it becomes difficult to propose uniform - or even comparable - model comparison practices. This paper outlines a broad and classificatory comparative epistemology of models. The aim of this epistemology is to propose applicable conceptual tools - if necessary rectifiable - that can be useful to modellers as well as to historians and epistemologists. The notion of model characteristic vector - incorporating concepts of function, nature, principle and use of the model - is introduced for this purpose. This proposal has been developed to serve a better equipped writing of a comparative and interdisciplinary history of models and simulations.; Dans un contexte de pluralisation, de sophistication et de combinaison des modèles formels, il devient difficile de proposer des pratiques uniformes - sinon elles-mêmes comparables - de comparaison de modèles. Cet article présente les grandes lignes d’une épistémologie comparative des modèles à la fois large et classificatoire. Le but de cette épistémologie est de proposer des outils conceptuels applicables, développables - au besoin rectifiables - qui puissent être utiles aux modélisateurs comme aux historiens et aux épistémologues. La notion ouverte de vecteur caractéristique de modèle - incorporant en particulier les concepts de fonction, de nature, de principe et d'usage du modèle - est introduite à cette fin. Cette proposition s’est développée pour servir à l’écriture mieux outillée d’une histoire comparative et interdisciplinaire des modèles et des simulations.

UMR AGAP - équipe AFEF - Architecture et fonctionnement des espèces fruitières; This study presents the first evaluation of apple flowering phenology models using data from 14 sites across the globe. The dataset includes large variability in growing climates, a prerequisite to investigate phenology models for use in climate change applications. Two flowering stages, early and full, were investigated allowing for unique model evaluation based on both statistical performance and biological assumptions. Two overarching phenology models (Sequential and Chill Overlap) and two sub-models of chill (Dynamic and Triangular) and heat (GDH and Sigmoidal) were tested. Flowering times from the different sites illustrated the differing effects of contrasting winter and spring temperatures. Sites with similar springtime temperatures, but different winter temperatures, had different flowering patterns (warmer winter sites flowered later). Across all analyses, results from the Chill Overlap model were better than those from the Sequential model. Of the Chill Overlap models, those fitted with the Triangular or Dynamic chill model and the GDH heat sub-model performed well statistically and met the assumptions of the model across both flowering stages. The mild sites in the analysis were least well represented, regardless of model selection. This global evaluation demonstrated that flowering modelling in temperate fruit trees would progress through appropriate choices of overarching model, sub-models and parameters.

The CO2 fertilization effect is a major source of uncertainty in crop models for future yield forecasts, but coordinated efforts to determine the mechanisms of this uncertainty have been lacking. Here, we studied causes of uncertainty among 16 crop models in predicting rice yield in response to elevated [CO2] (E-[CO2]) by comparison to free-air CO2 enrichment (FACE) and chamber experiments. The model ensemble reproduced the experimental results well. However, yield prediction in response to E-[CO2] varied significantly among the rice models. The variation was not random: models that overestimated at one experiment simulated greater yield enhancements at the others. The variation was not associated with model structure or magnitude of photosynthetic response to E-[CO2] but was significantly associated with the predictions of leaf area. This suggests that modelled secondary effects of E-[CO2] on morphological development, primarily leaf area, are the sources of model uncertainty. Rice morphological development is conservative to carbon acquisition. Uncertainty will be reduced by incorporating this conservative nature of the morphological response to E-[CO2] into the models. Nitrogen levels, particularly under limited situations, make the prediction more uncertain. Improving models to account for [CO2] × N interactions is necessary to better evaluate management practices under climate change.

With world population growing quickly, agriculture needs to produce more with fewer inputs while being environmentally friendly. In a context of changing environments, crop models are useful tools to simulate crop yields. Wheat (Triticum spp.) crop models have been evolving since the 1960s to translate processes related to crop growth and development into mathematical equations. These have been used over decades for agronomic purposes, and have more recently incorporated advances in the modeling of environmental footprints, biotic constraints, trait and gene effects, climate change impact, and the upscaling of global change impacts. This review outlines the potential and limitations of modern wheat crop models in assisting agronomists, breeders, and policymakers to address the current and future challenges facing agriculture.

The potential impact of global temperature change on global crop yield has recently been assessed with different methods. Here we show that grid-based and point-based simulations and statistical regressions (from historic records), without deliberate adaptation or CO2 fertilization effects, produce similar estimates of temperature impact on wheat yields at global and national scales. With a 1 °C global temperature increase, global wheat yield is projected to decline between 4.1% and 6.4%. Projected relative temperature impacts from different methods were similar for major wheat-producing countries China, India, USA and France, but less so for Russia. Point-based and grid-based simulations, and to some extent the statistical regressions, were consistent in projecting that warmer regions are likely to suffer more yield loss with increasing temperature than cooler regions. By forming a multi-method ensemble, it was possible to quantify ‘method uncertainty’ in addition to model uncertainty. This significantly improves confidence in estimates of climate impacts on global food security. The potential impact of global temperature change on global crop yield has recently been assessed with different methods. Here we show that grid-based and point-based simulations and statistical regressions (from historic records), without deliberate adaptation or CO2 fertilization effects, produce similar estimates of temperature impact on wheat yields at global and national scales. With a 1 °C global temperature increase, global wheat yield is projected to decline between 4.1% and 6.4%. Projected relative temperature impacts from different methods were similar for major wheat-producing countries China, India, USA and France, but less so for Russia. Point-based and grid-based simulations, and to some extent the statistical regressions, were consistent in projecting that warmer regions are likely to suffer more yield loss with increasing temperature than cooler regions. By forming a multi-method ensemble, it was possible to quantify ‘method uncertainty’ in addition to model uncertainty. This significantly improves confidence in estimates of climate impacts on global food security.

An important aspect that determines the ability of crop growth models to predict growth and yield is their ability to predict the rate of water consumption or evapotranspiration (ET) of the crop, especially for rain-fed crops. If, for example, the predicted ET rate is too high, the simulated crop may exhaust its soil water supply before the next rain event, thereby causing growth and yield predictions that are too low. In a prior inter-comparison among maize growth models, ET predictions varied widely, but no observations of actual ET were available for comparison. Therefore, another study has been initiated under the umbrella of AgMIP (Agricultural Model Inter-Comparison and Improvement Project). This time observations of ET using the eddy covariance technique from an 8-year-long experiment conducted at Ames, IA are being used as the standard. Simulation results from 29 models have been completed. In the first “blind” phase for which only weather, soils, and management information were furnished to the modelers, estimates of seasonal ET varied from about 200 to about 700 mm. A detailed statistical analysis of the daily ET data from 2011, a “typical” rainfall year, showed that, as expected, the median of all the models was more accurate across several criteria (correlation, root mean square error, average difference, regression slope) than any particular model. However, some individual models were better than the median for a particular criteria. Predictions improved somewhat in later stages when the modelers were provided additional leaf area and growth information that allowed them to “calibrate” some of the parameters in their models to account for varietal characteristics, etc.

International audience

International audience

International audience; This study evaluated the performance of alternative green solvents, i.e. D-limonene, dimethylcarbonate (DMC), ethanol, ethyl acetate, ethyl lactate and methyltetrahydrofuran (MeTHF) compared to the petroleum based dichloromethane, for extraction of alpha-mangostin from Garcinia mangostana pericarps. The Hansen solubility parameters (HSPs) were used to explain the dissolution behavior of the solutes and solvents, and the conductor-like screening model for realistic solvation. The (COSMO-RS), a statistical thermodynamic approach based on the results of quantum chemical calculations for comprehending the dissolving mechanisms were used to predict the extraction prediction. On the basis of the Hansen analysis, dichloromethane was the most suitable solvent for extraction of alpha-mangostin. However, COSMO-RS analysis showed a higher solubility of alpha-mangostin in ethyl lactate, DMC, MeTHF, ethyl acetate and ethanol. Moreover, the experimental studies using a classical reflux extraction followed by a quantitative HPLC analysis of alpha-mangostin showed similar results to the predictive values from the COSMO-RS model. The alpha-mangostin levels extracted by ethyl lactate, DMC, MeTHF, ethyl acetate and ethanol were higher than those using dichloromethane and D-limonene. The results support the potential use of ethyl lactate, DMC, MeTHF, and ethanol as alternative green solvents for the preparation of alpha-mangostin extracts.

International audience

International audience

A large number of organic matter degradation, CO2 transport and dissolved organic matter models have been developed during the last decades. However, organic matter degradation models are in many cases strictly hard-coded in terms of organic pools, degradation kinetics and dependency on environmental variables. The scientific input of the model user is typically limited to the adjustment of input parameters. In addition, the coupling with geochemical soil processes including aqueous speciation, pH-dependent sorption and colloid-facilitated transport are not incorporated in many of these models, strongly limiting the scope of their application. Furthermore, the most comprehensive organic matter degradation models are combined with simplified representations of flow and transport processes in the soil system. We illustrate the capability of generic reactive transport codes to overcome these shortcomings. The formulations of reactive transport codes include a physics-based continuum representation of flow and transport processes, while biogeochemical reactions can be described as equilibrium processes constrained by thermodynamic principles and/or kinetic reaction networks. The flexibility of these type of codes allows for straight-forward extension of reaction networks, permits the inclusion of new model components (e.g.: organic matter pools, rate equations, parameter dependency on environmental conditions) and in such a way facilitates an application-tailored implementation of organic matter degradation models and related processes. A numerical benchmark involving two reactive transport codes (HPx and MIN3P) demonstrates how the process-based simulation of transient variably saturated water flow (Richards equation), solute transport (advection-dispersion equation), heat transfer and diffusion in the gas phase can be combined with a flexible implementation of a soil organic matter degradation model. The benchmark includes the production of leachable organic matter and inorganic carbon in the aqueous and gaseous phases, as well as different decomposition functions with first-order, linear dependence or nonlinear dependence on a biomass pool. In addition, we show how processes such as local bioturbation (bio-diffusion) can be included implicitly through a Fickian formulation of transport of soil organic matter. Coupling soil organic matter models with generic and flexible reactive transport codes offers a valuable tool to enhance insights into coupled physico-chemical processes at different scales within the scope of C-biogeochemical cycles, possibly linked with other chemical elements such as plant nutrients and pollutants.

International audience

Maintaining or improving biodiversity is an important goal of sustainable forest management.Ground flora, which is responsible for most floristic diversity in temperate forests, plays multiple important roles in biodiversity but may be impacted by the increasing mechanisation of forest practices. At stand scale, we investigated in Montargis forest the individual and combined effects of tree stand attributes and skid trail area on ground flora diversity. Tree stand attributes (stand type or basal area) were the best indicators of ground flora diversity, depending on the successional traits or light preference of the species group. The effects of skid trail area were negligible. At finer scale, we studied plant response to skid trail disturbance (represented by subplot on and off skid trails), micro-environmental factors (soil moisture, soil compaction, light) and stand attribute (stand type, basal area). The best models for ecological groups included subplot location, soil moisture or soil compaction, depending on which ecological groups (classified by life form, seed bank persistence, light and moisture requirements) the species belonged to. Stand type as a covariate played a significantly important role in fine-scale diversity pattern. Subplot location was the dominant factor at species level. In conclusion, skid trails had either no impact or a positive impact on ground flora diversity. These results are dependent on the context of Montargis forest (ecological and historical), especially that mechanized harvesting is relatively recent. The employment of heavier machines and increased number of passages is likely to happen. This might induce greater soil compaction and negative effects on plant.; Le maintien ou l'amélioration de la biodiversité est un des objectifs importants de la gestion forestière durable. La flore du sous-bois, qui représente la partie la plus diversifiée de la flore dans les forêts tempérées, joue des rôles écologiques importants. Pourtant, elle pourrait être impactée par l'augmentation de la mécanisation de la gestion forestière. A l'échelle de la parcelle, nous avons étudié en forêt de Montargis les effets simples et combinés de caractéristiques du peuplement et de la surface en cloisonnement sur la diversité floristique du sous-bois (richesse et abondance). Les caractéristiques du peuplement (type de peuplement ou surface terrière des essences à étaient les meilleurs indicateurs de la diversité du sous-bois. La surface des cloisonnements avait un effet négligeable. A plus petite échelle – à l’intérieur du cloisonnement – nous avons étudié la réponse statistique de la diversité du sous-bois à la position dans ou hors du cloisonnement, à des facteurs micro-environnementaux (humidité du sol, compaction du sol, lumière) et aux caractéristiques du peuplement. A cette échelle, les meilleurs modèles incluaient pour les groupes écologiques la position par rapport au cloisonnement, l’humidité du sol et/ou la compaction du sol, selon le groupe écologique considéré. Au niveau espèce, la position par rapport au cloisonnement était le facteur dominant. Globalement, les cloisonnements avaient soit pas d’effet soit un impact positif sur la diversité floristique de sous-bois. Ces résultats ont dépendants du contexte écologique et historique de la forêt de Montargis. L’utilisation d’engins plus lourds ou des passages répétés sur une plus longue période pourraient changer ces conclusions.

The goal of task 4.4 of AnimalChange project is to improve a range of models used to assess climate change impacts on the productivity of pastures, feeding crops and animals.

Modeling nitrous oxide (N2O) emissions from agricultural soils is still a challenge due to influences of artificial management practices on the complex interactions between soil factors and microbial activities. The aims of this study were to evaluate the process-based DeNitrification-DeComposition (DNDC, version 9.5) model and modified non-linear empirical Nitrous Oxide Emission (NOEV2) model with weekly N2O flux measurements at eight sites cropped with winter wheat across a tile-drained landscape (around 30-km(2)) in Central France. Adjustments of the model default field capacity and wilting point and the optimum crop production were necessary for DNDC95 to better match soil water content and crop biomass yields, respectively. Multiple effects of varying soil water and nitrate contents on the fraction of N2O emitted through denitrification were added in NOEV2. DNDC95 and NOEV2 successfully predicted background N2O emissions and fertilizer-induced emission peaks at all sites during the experimental period but overestimated the daily fluxes on the sampling dates by 54 and 25 % on average, respectively. Cumulative emissions were slightly overestimated by DNDC95 (4 %) and underestimated by NOEV2 (15 %). The differences between evaluations of both models for daily and cumulative emissions indicate that low frequency measurements induced uncertainty in model validation. Nonetheless, our validations for soil water content with daily resolution suggest that DNDC95 well represented the effect of tile drainage on soil hydrology. The model overestimated soil ammonium and nitrate contents mostly due to incorrect nitrogen partitioning when urea ammonium nitrate solution was applied. The performance of the model would be improved if DNDC included the canopy interception and foliar nitrogen uptake when liquid fertilizer was sprayed over the crops.

International audience; This paper describes the scientific validation of the first version of global biophysical products (i.e., leaf area index, fraction of absorbed photosynthetically active radiation and fraction of vegetation cover), namely GEOV1, developed in the framework of the geoland-2/BioPar core mapping service at 1 km spatial resolution and 10-days temporal frequency. The strategy follows the recommendations of the CEOS/WGCV Land Product Validation for LAI global products validation. Several criteria of performance were evaluated, including continuity, spatial and temporal consistency, dynamic range of retrievals, statistical analysis per biome type, precision and accuracy. The spatial and temporal consistencies of GEOV1 products were assessed by intercomparison with reference global products (MODIS c5, CYCLOPES v3.1, GLOBCARBON v2 LAI, and JRC SeaWIFS FAPAR) over a global network of homogeneous sites (BELMANIP-2) during the 2003–2005 period. The accuracy of GEOV1 was evaluated against a number of available ground reference maps. Our results show that GEOV1 products present reliable spatial distribution, smooth temporal profiles which are stable from year to year, good dynamic range with reliable magnitude for bare areas and dense forests, and optimal performances with ground-based maps. GEOV1 outperforms the quality of reference global products in most of the examined criteria, and constitutes a step forward in the development of consistent and accurate global biophysical variables within the context of the land monitoring core service of GMES.

Les jeux de données écologiques présentent fréquemment des dépendances de type spatiales, consistant en ce que les observations géographiquement proches les unes des autres ont plus tendance à se ressembler que les observations plus distantes. Cet état de dépendance, appelé autocorrélation spatiale, s’il n’est pas correctement pris en compte dans la modélisation statistique, est connu pour produire des tests non fiables ainsi que des estimations imprécises des effets fixes. Bien que les chercheurs disposent de nombreuses méthodes permettant de modéliser l’autocorrélation spatiale, ils restent souvent perplexes face aux nombreux débats présents dans la littérature d’écologie statistique, où s’opposent les partisans des méthodes spatiales à ceux qui s’en méfient. Beale et al. (2010) apportent des éclaircissements dans le contexte de données continues réparties sur grille, mais les chercheurs ont besoin de conseils supplémentaires pour d’autres types de données dans des contextes spatiaux plus flexibles. Dans cette étude, nous étendons les travaux de Beale et al. (2010) au cas de données de comptage irrégulièrement réparties dans l’espace. Nous avons sélectionné quatre méthodes spatiales faciles à implémenter dans la famille des modèles linéaires généralisés mixtes et des modèles additifs généralisés mixtes, incorporant des effets aléatoires spatialement structurés qui permettent de modéliser les dépendances spatiales, soit à partir d’une notion de distance, soit par la spécification de relations de voisinages. A l’aide d’une approche par simulation, on évalue les performances relatives des méthodes testées en termes d’erreur de type I et de précision réelle des estimations. On s’intéresse, de plus, à leur robustesse face à des phénomènes de non-stationnarité spatiale et de sur-dispersion, qui sont des propriétés typiques des données de comptage spatialement autocorrélées en écologie. Parmi les méthodes fréquentistes et bayésiennes implémentées, les méthodes bayésiennes spatialement structurées estiment les effets fixes de la manière la plus précise avec des erreurs de type I correctes. La méthode spatiale basée sur une inférence MCMC avec mises à jours de Langevin-Hastings s’adapte le mieux à toutes les complexités simulées.

Times Cited: 1Zhao, Y. Ciais, P. Peylin, P. Viovy, N. Longdoz, B. Bonnefond, J. M. Rambal, S. Klumpp, K. Olioso, A. Cellier, P. Maignan, F. Eglin, T. Calvet, J. C.; We analyze how biases of meteorological drivers impact the calculation of ecosystem CO2, water and energy fluxes by models. To do so, we drive the same ecosystem model by meteorology from gridded products and by meteorology from local observation at eddy-covariance flux sites. The study is focused on six flux tower sites in France spanning across a climate gradient of 7-14 A degrees C annual mean surface air temperature and 600-1040 mm mean annual rainfall, with forest, grassland and cropland ecosystems. We evaluate the results of the ORCHIDEE process-based model driven by meteorology from four different analysis data sets against the same model driven by site-observed meteorology. The evaluation is decomposed into characteristic time scales. The main result is that there are significant differences in meteorology between analysis data sets and local observation. The phase of seasonal cycle of air temperature, humidity and shortwave downward radiation is reproduced correctly by all meteorological models (average R-2 = 0.90). At sites located in altitude, the misfit of meteorological drivers from analysis data sets and tower meteorology is the largest. We show that day-to-day variations in weather are not completely well reproduced by meteorological models, with R-2 between analysis data sets and measured local meteorology going from 0.35 to 0.70. The bias of meteorological driver impacts the flux simulation by ORCHIDEE, and thus would have an effect on regional and global budgets. The forcing error, defined by the simulated flux difference resulting from prescribing modeled instead of observed local meteorology drivers to ORCHIDEE, is quantified for the six studied sites at different time scales. The magnitude of this forcing error is compared to that of the model error defined as the modeled-minus-observed flux, thus containing uncertain parameterizations, parameter values, and initialization. The forcing error is on average smaller than but still comparable to model error, with the ratio of forcing error to model error being the largest on daily time scale (86%) and annual time scales (80%). The forcing error incurred from using a gridded meteorological data set to drive vegetation models is therefore an important component of the uncertainty budget of regional CO2, water and energy fluxes simulations, and should be taken into consideration in up-scaling studies.

Appropriate tools and models are needed for the management of mixed forests. The aim of this thesis is to show how the construction and the comparison of models with different levels of detail can help us to choose the most appropriate level to model the growth of a mixed stand. We developed a family of models at different levels of detail from data collected in mixed stands of sessile oak (Quercus petraea L.) and Scots pine (Pinus sylvestris L.) : a tree distance independent model (MAID), a tree distance dependent model (MADD), three stand models and an intermediate model bridging the MAID and the MADD. We ensured consistency between models using several approaches in order to make relevant comparisons. These models have given us some knowledge on the growth and dynamics of these forests, in particular on the spatial and temporal interactions between oaks and pines. Thus, we showed a compensatory growth phenomenon between the two species using the MAID. The MADD made it possible to show that, in these stands, the intraspecific competition was stronger than interspecific competition. A stand model developed from the MADD helped us to study the influence of mixing rate on production. To assess the quality of models predictions, we used an independent data set obtained by splitting our data. For example, we have shown that the MAID was more efficient than the MADD to predict individual increments. The models were also compared on examples of applications with short or medium term simulations. The proposed approach is of interest for both understanding the studied phenomenon and developing predictive tools. The different results of this work, allowed us to assess the relevance of a type of model for different contexts of use. This very general approach could be applied to the modeling of other processes such as mortality or regeneration.; Face à l'intérêt grandissant pour les forêts mélangées, des d'outils et des modèles adaptés à leur gestion sont nécessaires. L'objectif de cette thèse est de montrer comment la construction et la comparaison de modèles ayant différents niveaux de détail peuvent nous aider à choisir le niveau le plus approprié pour modéliser la croissance d'un peuplement mélangé dans un contexte d'application donné. A partir de données récoltées dans des peuplements mélangés chêne sessile (Quercus petraea L.) pin sylvestre (Pinus sylvestris L.), nous avons construit une famille de modèles à différents niveaux de détail : un modèle arbre indépendant des distances (MAID), un modèle arbre dépendant des distances (MADD), trois modèles peuplement et un modèle intermédiaire entre le MAID et le MADD utilisant des distributions de voisinage. Pour que la comparaison de ces modèles soit pertinente, nous avons assuré une cohérence entre les modèles en utilisant plusieurs approches. Ces modèles nous ont permis d'acquérir des connaissances sur la croissance et la dynamique de ces forêts en particulier sur les interactions spatiales et temporelles entre le chêne et le pin. Ainsi le MAID a permis de montrer qu'il peut y avoir des phénomènes de compensation de croissance entre les deux espèces. Le MADD a permis de montrer que dans ces peuplements la compétition intraspécifique est supérieure à la compétition interspécifique. Un modèle peuplement obtenu à partir du MADD a permis d'étudier l'influence du taux de mélange sur la production. Pour évaluer la qualité prédictive des modèles, nous avons utilisé un jeu de données indépendant obtenu en partageant nos données avant la construction des modèles. Nous avons ainsi montré que le MAID était plus performant que le MADD pour prédire les accroissements individuels. Les modèles ont aussi été comparés sur des exemples d'applications mettant en œuvre des simulations à court ou moyen terme. La démarche proposée présente un intérêt aussi bien pour la compréhension du phénomène étudié que pour sa modélisation dans un but prédictif. En regroupant l'ensemble des résultats acquis, ce travail nous a permis d'apprécier la pertinence d'un type de modèle en fonction du contexte d'utilisation. Cette démarche très générale pourrait être appliquée à la modélisation d'autres processus comme la mortalité ou la régénération.

Deux indicateurs de biodiversité associés à la diversité du peuplement arboré sont utilisés en France et en Europe, en l'absence de validation scientifique forte: (1) la richesse en essences ou en genres d'essences, comme un indicateur positif; et (2) l'abondance relative de l'essence dominante ("dominance") comme un indicateur négatif. Nous avons testé le caractère indicateur de ces indicateurs vis-à-vis de la biodiversité floristique du sous-bois, en les comparant à d'autres modèles écologiques, principalement associés à la composition et l'abondance des essences. Nous avons utilisé des modèles statistiques Bayésiens des données d'abondance et de richesse de groupes écologiques, définis sur la base du statut successionnel des espèces ou de leur tolérance à l'ombrage. Les distributions de probabilités utilisées dans les modèles sont nouvelles en écologie. Nous avons étudié la magnitude des effets avec des tests d'équivalence. 49 placettes adultes situées dans la région IFN de la Brie Francilienne ont été étudiées. Sur notre jeu de données, les indicateurs basés sur la richesse en genres ou la dominance étaient plus mauvais que les indicateurs basés sur l'abondance des genres d'essences. La magnitude des effets et l'identité du meilleur modèle indicateur variaient d'un groupe écologique à l'autre. Nos résultats montrent par ailleurs l'influence négative de la surface terrière des essences de demi-ombre sur le recouvrement de tous les groupes écologiques d'herbacées et de ligneux et sur la richesse des groupes non-forestiers et péri-forestiers de ligneux et d'herbacées. Comparé aux études du même genre, notre échantillonnage contrôlait fortement le type de station, éliminant ainsi partiellement la confusion potentielle entre les influences de la gestion et du type de station sur la biodiversité. / Two different biodiversity indicators based on tree species diversity are being used, in Europe and France respectively, without strong prior scientific validation: (1) tree species or genus richness as a positive indicator, and (2) relative abundance of the main species (dominance') as a negative indicator. We tested the relevance of these ecological models as indicators of understory vegetation biodiversity by comparing them to other ecological models, mainly related to tree species composition and abundance. We developed Bayesian statistical models for richness and abundance of ecological groups of understory vegetation species, classified according to successional status or shade tolerance. The count data probability distributions in the models were new to ecology. These models were fitted using data from 49 plots in mature lowland forests in the center of France (Bassin Parisien) with similar site conditions. We used equivalence and inequivalence tests to detect negligible and non-negligible effects. Tree genus richness and dominance resulted in models that were worse than ones based on the abundance of tree genus groups. Furthermore, the only significant results for dominance and tree genus richness were opposite to the ones implicitly assumed in the indicator system. However, the magnitude of the effects and which indicator provided the best statistical model varied among ecological groups of plants. Our results show the negative non-negligible effect of the basal area of undergrowth tree species on the cover of all ecological groups of herbaceous and woody species, and on the species richness of non-forest and peri-forest herbaceous and woody species. Compared to the literature, our sampling design strongly controlled forest and site type, thus removing to some degree the potential confusion between influences on biodiversity of management specific variables and other ecological variables. We discuss our results from both an ecological perspective and in terms of the value of these models as indicators of sustainable management. For example, the best-performing model was a multivariate model, which may be more difficult to explain to forest managers or policy-makers than an indicator simply based on tree genus richness.

International audience; Tree species is one of the main attributes of forest ecosystems, that modifies resources levels (light, nutrients, water), and so, indirectly, understory vegetation composition and diversity. Furthermore, forest managers often apprehend biodiversity through the diversity of tree species. This may explain why two biodiversity indicators based on tree species diversity are used, without strong prior scientific validation, in Europe and France respectively: (1) tree species richness as a positive indicator, and (2) relative abundance of the main species ("dominance") as a negative indicator. We developed Bayesian statistical models to test these indicators for the richness of ecological groups of understory vegetation species, classified according to their successional status and/or their shade tolerance. We compared these models with other ones, based on tree species composition and abundance. We worked in 49 stands of adult lowland forests in the center of France (Bassin Parisien), in a controlled site type. Tree species richness and dominance of the main tree species were not good indicators of understory vegetation species richness. There were better indicators than tree species richness or dominance, such as the abundance of tree species successional groups (pioneers, oaks, shade tolerant trees). The magnitude of the effects however varied among ecological groups of vascular plants. We discuss the meaning and limitation of our results in the light of possible extensions of this approach to bigger data sets. First, we discuss the magnitude and significance of the effects and their relation to the sampling scheme. Second, the best model is a multivariate model which is more difficult to communicate to forest managers or policy-makers than an indicator simply based on tree species richness. Lastly, our approach deals only with plant biodiversity; other taxa should also be considered.; La diversification des essences arborées en forêt est préconisée pour plusieurs raisons, parmi lesquelles l'augmentation de la diversité végétale. Cet effet a priori positif sur la flore du sous-bois n'a cependant pas été beaucoup étudié. La diversité en essences a plusieurs composantes. Parmi les quelques études disponibles qui se sont intéressées à l'effet de la diversité (richesse ou indice de shannon) en essences sur la diversité des strates herbacées ou muscinales, certaines montrent un effet positif. L'effet négatif d'essences de sous-étage telles que le charme a été montré, mais on manque de données sur l'effet du degré de mélange d'essences arborées, en particulier en France. Les principaux indicateurs de biodiversité basés sur la diversité en essences sont la richesse en essences en elle-même et le niveau de pureté de l'essence dominante (indicateurs de gestion durable 4.1 et 4.1.1, édition 2006, Ministère de l'Agriculture et de la Pêche). Nous avons testé dans des forêts de Brie (peuplements adultes à base de chêne, charme, tilleul, bouleau et tremble) l'effet de la richesse en essences et l'effet de la pureté. Nous distinguons les espèces de la flore (vasculaires et bryophytes) en plusieurs groupes selon leur caractère forestier. Nous avons comparé ces deux indicateurs à d'autres indicateurs potentiels de biodiversité. Nos résultats montrent que ces deux indicateurs ont des effets non significatifs, souvent faibles. Un modèle indicateur, basé sur l'abondance de différents groupes d'essences, apparaît être un meilleur indicateur de biodiversité floristique. Attention, ces résultats ne sont pas forcément généralisables ; des études doivent être menées sur d'autres forêts françaises. Pour le moment, trop peu d'études se sont intéressées à l'effet de la diversité en essences sur la flore. Enfin, d'autres groupes que la flore devraient être étudiés.

soutenue le 02/07/2008 Diplôme : Dr. d'Université

Thèse dirigée par Jean-Claude Rameau jusqu'en octobre 2005.Thesis advisor was Jean-Claude Rameau untill october 2005.; Tree species is one of the mains attributes of a forest. It is a factor that modifies resources levels (light, nutrients, water), and thus it potentially acts on understory vegetation composition and diversity. However, few studies deal on tree species effect on understory vegetation, at least in temperate forests. These studies show a lower vegetation diversity in coniferous stands compared to hardwoods, and generally a lower diversity in two-species mixed stands compared to pure ones. Even so, tree species richness and relative abundance of the main species (pureness) are used in France and Europe as biodiversity indicators (respectively positive and negative), without previous scientific validation.We have developed statistical models to test these indicators for the richness of ecological groups of understory vegetation species. We have compared these models with other ones, based on tree species composition and abundance. We worked in lowland forests in the center of France (Bassin Parisien). Site type was either constant, either incorporated into models. According to our results, tree species richness is not a good indicator of understory vegetation species richness, and pureness of the main tree species has positive or null effects. Positive responses are probably due to tree species stratification in the forests studied: they are mainly oak forests managed as standards or coppice (hornbeam) with standards, in which the decrease of pureness often corresponds to an increase in coppice density. Coppice species have negative or null effects (or sometimes positive effects, for woodland species).There are better indicators than tree species richness or pureness, such as main tree species identity and abundance of tree species groups. Main tree species identity is a useful indicator: it shows for example that coniferous stands are less rich in shad-tolerant species, ash stands have a higher richness for many understory groups, and sessile oak stands are less rich than common oak stands for most of understory groups. The abundance of tree species groups (oaks, pioneers, shad tolerant trees, other hardwoods, post-pioneer conifers and late-successional conifers) is the best descriptor for the richness of quite a few ecological groups. However, this multivariate model is not appropriated to be communicated as an indicator based on tree species.To avoid site type bias, we have studied tree species effect at the individual tree scale. Soil water variations between tree pairs (pineoak; beech-oak) were rare, available light for vegetation sometimes highly varied with tree species (mainly near the trunk), but there were few understory differences between tree species at this scale (except for bryophytes). Stand structure seems to be more determining than characteristics peculiar to each individual tree. Tree species composition is a synthetic variable that partially informs about factors of understory diversity to which it is bound: site type, management practices, trees structure and historical factors. These factors are inherent and closely linked to tree species composition in managed forests. If future changes occurred in management practices (tree species choice according to site type, cutting frequency and intensity), relations between tree species and understory diversity may change; thus our indicators should be tested again.; La composition en essences est une des principales caractéristiques des forêts. Elle a notamment des effets sur les niveaux de ressources (lumière, nutriments, eau) et donc potentiellement sur la flore. Cependant, seules quelques études se sont intéressées aux relations entre composition en essences et diversité floristique dans les forêts tempérées. Elles montrent que les peuplements de résineux sont moins diversifiés que les peuplements de feuillus, et que les peuplements mélangés ne sont en général pas plus diversifiés que les peuplements purs. Pourtant, la richesse en essences et la pureté de l'essence principale sont utilisées en France et en Europe comme indicateurs (respectivement positif et négatif) de biodiversité, sans que leur validité ait été vérifiée.Nous avons développé des modèles statistiques pour tester ces indicateurs sur la richesse de groupes écologiques de la flore. Nous les avons comparés avec des modèles basés sur la composition et l'abondance des essences, dans les forêts de plaine du centre de la France (Bassin Parisien). Les caractéristiques stationnelles ont été soit fixées, soit prises en compte dans les modèles. Il apparaît que la richesse en essences n'est pas un bon indicateur de diversité floristique, et que la pureté de l'essence principale a des effets positifs ou nuls sur la richesse floristique ; ces réponses positives s'expliquent probablement par la structure des peuplements étudiés qui sont principalement des chênaies et chênaies charmaies, dans lesquelles la diminution de la pureté de l'essence principale (souvent le chêne) correspond souvent à une augmentation des essences de taillis dans le sous-étage (comme le charme). Ces essences de taillis ont des effets négatifs ou nuls (voire positifs pour les herbacées forestières).Il existe de meilleurs indicateurs de biodiversité que la pureté et la richesse, en particulier l'identité de l'essence principale et l'abondance des essences classées par groupes. Le premier est un indicateur pratique ; il nous montre par exemple que les peuplements de résineux sont moins riches en espèces sciaphiles ; les peuplements de frênes sont plus riches pour de nombreux groupes d'espèces vasculaires ; les peuplements de chênes pédonculés sont plus riches que ceux de chênes sessiles pour de nombreux groupes d'espèces. L'abondance des groupes d'essences (chênes, pionnières, essences de demi-ombre, autres essences feuillues, résineuses dryades et résineuses post-pionnières) est le meilleur descripteur de la richesse pour plusieurs groupes écologiques. Cependant, c'est un modèle multivarié qui n'est pas facile à utiliser en tant que tel comme indicateur.Nous avons étudié l'effet essence à l'échelle de l'arbre pour s'affranchir des biais stationnels ; les variations de l'eau du sol entre couples d'essences (pin–chêne ; hêtre–chêne) sont faibles, l'éclairement est parfois différent entre essences (surtout à proximité du tronc), mais il y a peu de différences floristiques entre essences à cette échelle (sauf pour les bryophytes). L'effet de la structure des peuplements semble plus important à considérer que les caractéristiques propres à chaque arbre individuel. La composition en essences est une variable synthétique qui nous renseigne en partie sur de nombreux facteurs de diversité végétale auxquels elle est liée : des caractéristiques stationnelles, sylviculturales, de structure du peuplement et historiques. Ces facteurs sont inhérents à la composition en essences dans les forêts existantes. Si la sylviculture changeait (choix d'essences par station et fréquence et intensité des perturbations), les relations entre composition en essences et diversité floristique changeraient probablement aussi ; nos indicateurs devraient alors être testés de nouveau.

Knowledge diffusion models typically involve two main features: an underlying social network topology on one side, and a particular design of interaction rules driving knowledge transmission on the other side. Acknowledging the need for realistic topologies and adoption behaviors backed by empirical measurements, it becomes unclear how accurately existing models render real-world phenomena: if indeed both topology and transmission mechanisms have a key impact on these phenomena, to which extent does the use of more or less stylized assumptions affect modeling results? In order to evaluate various classical topologies and mechanisms, we push the comparison to more empirical benchmarks: real-world network structures and empirically measured mechanisms. Special attention is paid to appraising the discrepancy between diffusion phenomena (i) on some real network topologies vs. various kinds of scale-free networks, and (ii) using an empirically-measured transmission mechanism, compared with canonical appropriate models such as threshold models. We find very sensible differences between the more realistic settings and their traditional stylized counterparts. On the whole, our point is thus also epistemological by insisting that models should be tested against simulation-based empirical benchmarks.

In this poster, we present how to build growth models for mixed stands. First we describe the different type of growth models in forest research, and then we present two types of models we have built for oak-pine mixed stands. These models will be compared to see their advantages and their drawbacks. Finally we present the main perspective of this works which is to build a simulator for forest managers to describe the growth and the dynamic of oak-pine mixed stands and to evaluate forest strategies., Mettre en ½uvre une gestion durable de nos forêts nécessite de pouvoir prédire la production de bois future et d'évaluer l'impact des interventions humaines sur l'évolution de la forêt. Pour répondre à ces besoins la recherche forestière a construit un grand nombre de modèles forestiers que l'on peut caractériser par rapport aux informations qu'ils utilisent pour prédire la croissance : informations à l'échelle du peuplement, de l'arbre (ou individu), infra arbre, etc. Dans de nombreux domaines, après une période de développement intense de modèles à l'échelle de l'individu, les modélisateurs se sont rendu compte que ces modèles répondaient parfois moins bien aux objectifs des utilisateurs que des modèles plus frustres. L'objectif de ma thèse est donc de définir l'échelle de modélisation la plus pertinente pour des peuplements mélangés en utilisant les méthodes de comparaison de modèles. Dans des peuplements mélangés chêne sessile pin sylvestre nous avons récolté des données de croissance en réalisant des sondages dans les arbres et en mesurant les largeurs cernes. Ces données nous permettront de déterminer les paramètres des modèles de croissance que nous souhaitons tester : un modèle à l'échelle du peuplement, un modèle à l'échelle de l'arbre, et des modèles correspondant à des étapes intermédiaires d'agrégation. Nous intégrerons ces modèles dans un programme informatique permettant de simuler la croissance de la forêt ainsi que des coupes. Ce simulateur sera utilisé pour tester et évaluer les modèles selon des critères de précision et d'utilisation. Ce travail permettra par la suite de développer un modèle plus complet à destination des gestionnaires forestiers.

International audience; The vertical water flow, heat flow and transport of the herbicide methabenzthiazuron were monitored for 627 days in lysimeters sampled at a field site close to the research centre Jülich, Germany. During this period the lysimeters were cropped with winter wheat, winter barley and oat. The models TRACE, MARTHE, ANSWERS and MACRO were applied to the lysimeter data with the scope of upscaling local scale process understanding for regional scale. MARTHE and TRACE solve the 3-d Richards' equation for variably saturated water flow. MACRO is a 1-d model based on the Richards' Equation and accounting for preferential flow in the unsaturated zone, while ANSWERS is a regional scale capacity based watershed model. Measurements of soil moisture, evapotranspiration, drainage, soil temperature, pesticide residues and leaching are used for comparison with model results. Although the adopted models differ in terms of model concepts, the use of model performance indices proved a proper simulation of water flow for all models. The heat flow is also well described with ANSWERS, MARTHE and MACRO. Larger deviations were found between model results and measured pesticide transport. An inadequate reproduction of the measured MBT degradation was found for the available model input parameters. A very small amount of MBT leaching, observed in the measurements, was only reproduced with MACRO after strong calibration. In other respects only plant parameters were calibrated. Calibration of the crop conversion factor used for scaling of the potential evapotranspiration was found to be a crucial parameter for the adequate description of the water balance by the models.

Biodiversity is a complex matter related to a developing scientific field. In these conditions, we think both simple and more complex, research related indicators should be tested. We propose that both sets of indicators be dynamically related, and especially insist here on the way complicated indicators, found to be more accurate than currently preferred simple indicators, could be incorporated to a list of simple indicators. We apply the method to a case study. We find that models of species abundance of the herbaceous layer vegetation based on ecological gradients of tree species richness and tree species evenness behave poorer than models based on gradients of tree abundance. We propose two ways to implement these more complicated models into new lists of simple indicators., Dans le cadre de la stratégie pour diminuer l'érosion de biodiversité d'ici 2010, des approches sectorielles sont mises en oeuvre. L'un des outils de ces approches sont des indicateurs de biodiversité, assez souvent basés sur des concepts scientifiques ou des idées généraux. Nous proposons de soumettre ces indicateurs de biodiversité sectoriels à des tests de qualité sur des données réelles. Nous présentons un tel test sur des indicateurs de richesse et de mélange d'essences en milieu forestier pour la biodiversité floristique. Nous trouvons que ces indicateurs expliquent moins bien la biodiversité floristique que des indicateurs d'abondance des essences. Nous discutons comment de telles approches peuvent être intégrées dans les dispositifs d'indicateurs sectoriels.

Cet article est consacré à l'évaluation statistique des descriptions de tables de contingence fournies par les arbres d'induction. On se limite au cas particulier de données catégorielles. Trois aspects sont successivement abordés. i) La nature de l'ajustement en apprentissage supervisé, où l'on souligne la distinction entre prédiction de valeurs individuelles et prédiction de leur représentation sous forme de table de contingence. ii) La description de tables fournies par les arbres d'induction que l'on compare notamment à la modélisation log-linéaire utilisée en statistique. iii) L'adaptation au cas des arbres d'induction des mesures et statistiques de qualité d'ajustement utilisées en modélisation log-linéaire. La discussion est complétée par une illustration sur les données du Titanic. - The paper is concerned with the statistical assessment of the description of contingency tables by induction trees. It focuses on the special case of categorical data. Three topics are successively considered. i) The nature of the fit in supervised learning where we stress the distinction between fitting individual values and fitting their cross-tabulated synthetic representation. ii) The description of contingency tables provided by induction trees which is compared with the log-linear modeling used in statistics. iii) The adaptation of the goodness-of-fit measures and statistics used in log-linear modeling to the case of induction trees. The discussion is completed with an application to the Titanic data set.

Methane (CH4) is a major greenhouse gas whose global warming potential is 23 times more important than carbon dioxide (CO2). CH4 concentrations have steadily increased since the beginning of the industrial era, reaching an unprecedent level (almost 1800 ppb at present times). CH4 is currently considered as a one of the major driver of the climate change and its warming potential has leaded the Kyoto protocol to plan a significative reduction of its emissions. The objective of this thesis is to examine the factors that contribute to the CH4 global budget as well as its year-to-year variations. To that purpose, we have implemented a methane simulation in a global model of chemistry and transport. Conducting a CH4 simulation requires a comprehensive set of interannual varying CH4 emission inventories as well as a year-to-year varying 3-D fields of concentrations of hydroxyl radicals (OH) that is the main sink for CH4. The first part of this thesis examines the OH interannual variation, using results from a "full-chemistry" simulation that accounts for interannual variations in emissions of the main O3 precursors (including carbon monoxide (CO), nitrogen oxides (NOx), and hydrocarbons) as well as the year-to-year variation in meteorology (e.g. atmospheric water vapor content), lightning emissions, and overhead O3 column. We calculated a global OH concentration of 1.22×106 molec cm-3 averaged over the period. We find that global OH concentrations steadily increased during the early 90's by 8% and reached a maximum in 1997. Afterward, concentrations decrease by 5%. We find that changes in water vapor concentrations, anthropogenic emissions of CO and NOx, lightning NOx emissions, and overhead ozone column all play a role in driving the OH year-to-year variations. The relative contributions of these parameters depend on the latitudinal and altitudinal regions of the troposphere. We examine in particular the influence of the 1997-1998 ENSO on the global OH concentrations. We find that the 1997-1998 ENSO resulted in a large increase in OH in the tropical areas and in the extra-tropical areas of the northern hemisphere (largely driven by an increase in water vapor) while it leads to a decrease in OH in the extra-tropical region of the southern hemisphere (that results from changes in the transport pathways that bring CO in the most southern latitudes and deplete OH). Our results (in terms of OH variability) are in contradiction to those found by other methods, especially those using inverse modeling approaches that are based on methyl chloroform (CH3CCl3) observations. This may results from poorly constrained sources. We then seek to implement a comprehensive set of CH4 emissions in our global model. We used anthropogenic CH4 emission data set of anthropogenic emissions from the International Institute of Technology (IIASA) whose emissions vary between 250 and 290 Tg/year between 1990 and 2000. We estimated biomass burning emissions, derived from an inventory of the total annual biomass burned area and emission factors that we constrained with measurements of the isotopic composition of atmospheric methane. This results in a global emissions rate of 60 Tg/year, which is much stronger than most of previous studies. We also developed a wetland scheme that accounts for soil carbon content, wetland fraction areas, soil temperature and humidity (with the three latter parameters varying interannually) and we evaluated the wetland areas with satellite-based estimates. Our global wetland emissions amount to about 150 Tg/year. We find a global CH4 lifetime of 10.6 years, which is in the range of values reported in previous studies. Comparing our results with different set of measurements gives promising results. The model reproduces well the year-to-year variation even if it slightly overestimates concentrations after 2000. By conducting different set of sensitivity simulations, we investigated the role of different parameters on the CH4 variations. We find that the long-term trend in CH4 is driven by a competition between anthropogenic emissions and tropospheric decay. Peaks of growth rate are driven by biomass burning emissions and to a lesser extend, wetland emissions. We find that even if we used increasing anthropogenic emissions during the 90s (that is in contradiction with some previous studies), reconstructing the observed year-to-year variation of CH4 concentrations was possible if one considers an increasing in OH during the 90s.

Even brief periods of high temperatures occurring around flowering and during grain filling can severely reduce grain yield in cereals. Recently, ecophysiological and crop models have begun to represent such phenomena. Most models use air temperature (T-air) in their heat stress responses despite evidence that crop canopy temperature (T-c) better explains grain yield losses. T-c can deviate significantly from T-air based on climatic factors and the crop water status. The broad objective of this study was to evaluate whether simulation of T-c improves the ability of crop models to simulate heat stress impacts on wheat under irrigated conditions. Nine process-based models, each using one of three broad approaches (empirical, EMP; energy balance assuming neutral atmospheric stability, EBN; and energy balance correcting for the atmospheric stability conditions, EBSC) to simulate To simulated grain yield under a range of temperature conditions. The models varied widely in their ability to reproduce the measured T-c with the commonly used EBN models performing much worse than either EMP or EBSC. Use of T-c to account for heat stress effects did improve simulations compared to using only T-air to a relatively minor extent, but the models that additionally use T-c on various other processes as well did not have better yield simulations. Models that simulated yield well under heat stress had varying skill in simulating T-c For example, the EBN models had very poor simulations of T-c but performed very well in simulating grain yield. These results highlight the need to more systematically understand and model heat stress events in wheat.

Knowledge of the soil water retention curve (SWRC) is essential for understanding and modeling hydraulic processes in the soil. However, direct determination of the SWRC is time consuming and costly. In addition, it requires a large number of samples, due to the high spatial and temporal variability of soil hydraulic properties. An alternative is the use of models, called pedotransfer functions (PTFs), which estimate the SWRC from easy-to-measure properties. The aim of this paper was to test the accuracy of 16 point or parametric PTFs reported in the literature on different soils from the south and southeast of the State of Para, Brazil. The PTFs tested were proposed by Pidgeon (1972), Lal (1979), Aina & Periaswamy (1985), Arruda et al. (1987), Dijkerman (1988), Vereecken et al. (1989), Batjes (1996), van den Berg et al. (1997), Tomasella et al. (2000), Hodnett & Tomasella (2002), Oliveira et al. (2002), and Barros (2010). We used a database that includes soil texture (sand, silt, and clay), bulk density, soil organic carbon, soil pH, cation exchange capacity, and the SWRC. Most of the PTFs tested did not show good performance in estimating the SWRC. The parametric PTFs, however, performed better than the point PTFs in assessing the SWRC in the tested region. Among the parametric PTFs, those proposed by Tomasella et al. (2000) achieved the best accuracy in estimating the empirical parameters of the van Genuchten (1980) model, especially when tested in the top soil layer., O conhecimento da curva de retenção de água (CRA) é essencial para compreender e modelar os processos hidráulicos no solo. No entanto, a determinação direta do CRA consome tempo, e o custo é alto. Além disso, é necessário grande número de amostras, em razão da elevada variabilidade espacial e temporal das propriedades hidráulicas do solo. Uma alternativa é o uso de modelos, que são chamados de funções de pedotransferência (FPT), que estimam a CRA por meio de propriedades do solo de fácil determinação. O objetivo deste estudo foi testar a acurácia de 16 FPT, pontuais ou paramétricas, existentes na literatura, em diferentes solos do sul e sudeste do Estado do Pará, Brasil. As FPT testadas foram propostas por Pidgeon (1972), Lal (1979), Aina & Periaswamy (1985), Arruda et al. (1987), Dijkerman (1988), Vereecken et al. (1989), Batjes (1996), van den Berg et al. (1997), Tomasella et al. (2000), Hodnett & Tomasella (2002), Oliveira et al. (2002) e Barros (2010). Utilizou-se um banco de dados contendo textura (areia, silte e argila), densidade do solo, carbono orgânico, pH do solo, capacidade de troca catiônica e CRA. A maioria das FPT testadas não demonstrou boa acurácia para estimar as CRA. As FPT paramétricas apresentaram melhor desempenho do que as FPT pontuais em estimar a CRA dos solos na região. Entre as FPT paramétricas, as propostas por Tomasella et al. (2000) obtiveram melhor acurácia em estimar os parâmetros empíricos do modelo de van Genuchten (1980), principalmente, quando testadas na primeira camada do solo.

In this paper we compare a version of the individual-based 'threshold model' of innovation diffusion with an aggregate deterministic model that we constructed from it. The classical threshold model supposes that an individual adopts a behaviour according to a trade-off between a social pressure (the number of neighbours adopting the behaviour) and a personal interest or resistance to change (the threshold). The aggregate model makes approximations in order to estimate the evolution of groups of individuals with the same number of neighbours of similar behaviour. We compare both models at different points of the parameter space. We find that the aggregate model gives a good approximation of the individual in some cases; however in other cases the behaviour of the aggregate approximation differs. Using theoretical interpretation of this difference based on a study of the attractors of the aggregate model, we hypothesise that the two models have the same behaviour when the aggregate model has only one attractor and that differences can occur when it has two., Nous comparons un modèle individu-centré à seuil de diffusion de l'innovation (Valente 1995) dans la version étudiée par P. Young, et un modèle agrégé déterministe que nous avons construit en cohérence. Le modèle classique de diffusion de l'innovation à seuil suppose qu'un individu adopte un comportement suivant un compromis entre la pression sociale (le nombre de ses voisins ayant adopté ce comportement) et un intérêt personnel de résistance au changement (le seuil). Le modèle agrégé fait appel à des approximations pour estimer l'évolution de groupes d'individus ayant le même nombre de voisins suivant un certain comportement. Nous comparons ces deux modèles en différents points de l'espace des paramètres. Nous trouvons que le modèle agrégé approxime bien le modèle individu-centré dans certains cas ; toutefois, il s'avère également parfois que le comportement du modèle agrégé diffère. En nous appuyant sur une interprétation théorique de cette différence, fondée sur une étude des attracteurs du modèle agrégé, nous proposons l'hypothèse que les deux modèles suivent le même comportement lorsque le modèle agrégé n'a qu'un seul attracteur, et que des différences peuvent apparaître quand il en a deux.