La pomme, fruit le plus cultivé sur le sol français, est principalement menacée par le pathogène Venturia inaequalis, responsable de la maladie de la tavelure qui génère des pertes considérables si elle n'est pas traitée. La politique actuelle de gestions des risques phytopathologiques en France incite à une forte réduction des traitements phytosanitaires tout en maximisant le rendement et la qualité des productions. Dans ce contexte, il apparaît indispensable de mieux comprendre les interactions entre l'arbre, son pathogène, et leur environnement, qui s'articulent pour le cas de la tavelure du pommier autour de la durée d'humectation des feuilles. Au cours de ce travail nous nous sommes intéressés à ce paramètre pour essayer de mieux comprendre les interactions entre microclimat de l'arbre et durée d'humectation. L'étude s'est déroulée en trois étapes majeures : la modélisation de l'évaporation d'une goutte sur un support végétal, l'étude expérimentale de la mouillabilité des feuilles de pommier, et l'étude expérimentale de la variabilité spatiale de la durée d'humectation sous un couvert de pommiers. Ce travail a permis d’expliciter la forte variabilité intra-couronne de la durée d'humectation via la prise en compte de la structure de l’arbre et de la dynamique horaire du microclimat. Le modèle développé, au delà des liens déjà connus entre l’intensité du flux évaporatif et les variables climatiques, a montré la sensibilité importante du temps d’évaporation à la mouillabilité du support via la forme de la goutte d’eau, mettant en avant la nécessité de quantifier au mieux cette interaction goutte support via l’estimation des angles de contact statiques et dynamiques. / Apples, which are the most cultivated fruit in France, are mainly endangered by the fungal pathogen Venturia inaequalis that cause apple scab disease on apple. This disease can be responsible of major products loss unless orchards are treated against apple scab. Nowadays in France, the phytopathological diseases management policies are encouraging growers to reduce considerably the use of pesticides, while keeping a high quality and yield level. In this context, one must understand better how the plant, the pathogen and their environment, interact with each other: for apple scab, the most important environmental parameter is leaf wetness duration. During this work, we studied leaf wetness duration to understand the interactions that occur between the tree microclimate and the wetness duration. To do that we divided our work in three major steps : the modeling of evaporation of a droplet at rest on a leaf, the experimental study of apple leaves wettability, and the experimental study of wetness duration spatial variability within an apple trees orchard. This study led us to clarify the strong intra-crown variability of leaf wetness duration through the consideration of tree structure and hourly dynamics of microclimate. The model we developed, beyond the known links between the evaporative flux intensity and the climatic parameters, showed a strong sensibility of the evaporation duration to the substrate wettability, highlighting the necessity to quantify at best this interaction, through the estimation of static and dynamic contact angles.

Made available in DSpace on 2014-06-11T19:31:34Z (GMT). No. of bitstreams: 0 Previous issue date: 2006-12-11Bitstream added on 2014-06-13T20:42:01Z : No. of bitstreams: 1 mathias_im_dr_botfca.pdf: 1186171 bytes, checksum: a99a3192900af068caf82ad18c373cfa (MD5) / Coordenação de Aperfeiçoamento de Pessoal de Nível Superior (CAPES) / Universidade Estadual Paulista (UNESP) / O trabalho descrito nesta tese apresenta o desenvolvimento de um sistema computacional denominado PMNeural, baseado em Redes Neurais Artificiais (RNAs). A finalidade do sistema é o tratamento de dados climáticos e de molhamento foliar por orvalho, visando reconhecer padrões de comportamento de variáveis meteorológicas em relação ao molhamento foliar por orvalho. Para determinar as melhores arquiteturas e algoritmos de treinamento de RNAs, bem como, definir quais as variáveis climáticas que influenciam significativamente na ocorrência do molhamento foliar, foram utilizados dois simuladores: o simulador SNNS (Stuttgart Neural Network Simulator) versão 4.2, que utiliza plataforma operacional Linux e o simulador JavaNNS - Java Neural Network Simulator 1.1, com ambiente de execução Windows, o qual é baseado no SNNS. Foram utilizados dados climáticos de três estudos de caso, dois destes referentes à cultura do trigo, oriundos de locais e datas diferentes. Base de Dados 1 - Fazenda Capão do Cipó, em Castro - PR, safra de inverno de 2003. Base de Dados 2 - Campo Demonstrativo e Experimental da Fundação ABC - Fazenda Palmeirinha, em Piraí do Sul - PR., safra de inverno de 2005. Base de Dados 3 - Posto Agrometeorológico ESALQ/USP em Piracicaba - SP, período entre julho e setembro de 2005. Um quarto estudo de caso foi elaborado a partir dos arquivos dos estudos de casos 1, 2 e 3, utilizando-se as variáveis climáticas comuns, juntamente com seus respectivos índices de molhamento. Dentre os algoritmos de treinamento testados nos simuladores, o Resilient 2 Propagation (Rprop) foi o que apresentou as menores taxas de erro em relação aos outros... / The work described in this thesis presents the development of a computational system named PMNeural based on Artificial Neural Networks (ANNs). The system has for purpose the handle of climatic and leaf wetness data, aiming to recognize patterns of behavior of meteorological variables in relation to the wetness from dew. Two simulators were used in order to determine the best architecture and ANNs training algorithms, as well as, to define which the climatic variables that influence significantly in the leaf wetness occurrence: the SNNS (Stuttgart Neural Network Simulator) version 4.2 for Linux platform, and the JavaNNS - Java Neural Network Simulator 1.1, for Windows platform, which is based on the SNNS. Climatic data of three case studies were used, two related to wheat culture, obtained from different places and dates. Dataset 1 - Capão do Cipó Farm, in Castro - PR, 2003. Dataset 2 - Palmeirinha Farm in Piraí do Sul - PR, 2005 winter crop. Dataset 3 - Meteorological Station of ESALQ/USP in Piracicaba - SP, from July to September, 2005. A fourth case study was elaborated from datasets of the case studies 1, 2 and 3, using the common climatic variables together with their respective wetness indexes. After testing the training algorithms in the simulators, the Resilient Propagation (Rprop) presented lower training errors than the others evaluated methods: Backpropagation Standard, Backpropagation for batch training, Backpropagation with momentum term, Backpropagation with chunkwise update, Backpropagation with Weight Decay and Quickprop. It was verified 4 also that, among the climatic variables used for classification of leaf wetness from dew, the inclusion of the schedule had influenced in the obtaining better ANNs results... (Complete abstract, click electronic access below)

This dissertation contains three research chapters conducted on Sclerotinia stem rot (SSR) of canola (Brassica napus L.). This disease is caused by the fungus Sclerotinia sclerotiorum and is considered endemic in canola-producing areas of North Dakota. The first research chapter presents results of a study that evaluated the role of eight phenotyping scoring systems and nine variant calling and filtering methods in detection of QTL associated with response to SSR. The study, conducted on two doubled-haploid mapping populations, showed that using multiple phenotypic data sets derived from lesion length and plant mortality and imputing missing genotypic data increased the number of QTL detected without negatively affecting the effect (R2) of QTL. Nineteen QTL were detected on chromosomes A02, A07, A09, C01, and C03 in this study. The second research chapter presents results of a work that assessed the role of temperature regimes and wetness duration on S. sclerotiorum ascospore germination and ascosporic infection efficiency. This study showed that optimum ascospore germination occurred at 21 °C while it significantly decreased at 10 and 30 °C. Infection efficacy experiments indicated that extreme temperatures and interrupting wet periods were detrimental for the disease development. A logistic regression model with 75% accuracy was developed for the disease perdition. The third research chapter presents results of a study that evaluated the role of temperature on mycelial growth of 19 S. sclerotiorum isolates collected from different geographical regions and on SSR development on plant introduction (PI) lines with different levels of resistance. Mycelial growth and disease development peaked at 25 °C. While lesion expansion on resistant cultivars and the susceptible check was negatively affected at 30 °C, the disease developed significantly on the PI with a high level of susceptibility. Results of these studies provide insights into integrated management strategies of SSR.

Resumo: O trabalho descrito nesta tese apresenta o desenvolvimento de um sistema computacional denominado PMNeural, baseado em Redes Neurais Artificiais (RNAs). A finalidade do sistema é o tratamento de dados climáticos e de molhamento foliar por orvalho, visando reconhecer padrões de comportamento de variáveis meteorológicas em relação ao molhamento foliar por orvalho. Para determinar as melhores arquiteturas e algoritmos de treinamento de RNAs, bem como, definir quais as variáveis climáticas que influenciam significativamente na ocorrência do molhamento foliar, foram utilizados dois simuladores: o simulador SNNS (Stuttgart Neural Network Simulator) versão 4.2, que utiliza plataforma operacional Linux e o simulador JavaNNS - Java Neural Network Simulator 1.1, com ambiente de execução Windows, o qual é baseado no SNNS. Foram utilizados dados climáticos de três estudos de caso, dois destes referentes à cultura do trigo, oriundos de locais e datas diferentes. Base de Dados 1 - Fazenda Capão do Cipó, em Castro - PR, safra de inverno de 2003. Base de Dados 2 - Campo Demonstrativo e Experimental da Fundação ABC - Fazenda Palmeirinha, em Piraí do Sul - PR., safra de inverno de 2005. Base de Dados 3 - Posto Agrometeorológico ESALQ/USP em Piracicaba - SP, período entre julho e setembro de 2005. Um quarto estudo de caso foi elaborado a partir dos arquivos dos estudos de casos 1, 2 e 3, utilizando-se as variáveis climáticas comuns, juntamente com seus respectivos índices de molhamento. Dentre os algoritmos de treinamento testados nos simuladores, o Resilient 2 Propagation (Rprop) foi o que apresentou as menores taxas de erro em relação aos outros... (Resumo completo, clicar acesso eletrônico abaixo) / Abstract: The work described in this thesis presents the development of a computational system named PMNeural based on Artificial Neural Networks (ANNs). The system has for purpose the handle of climatic and leaf wetness data, aiming to recognize patterns of behavior of meteorological variables in relation to the wetness from dew. Two simulators were used in order to determine the best architecture and ANNs training algorithms, as well as, to define which the climatic variables that influence significantly in the leaf wetness occurrence: the SNNS (Stuttgart Neural Network Simulator) version 4.2 for Linux platform, and the JavaNNS - Java Neural Network Simulator 1.1, for Windows platform, which is based on the SNNS. Climatic data of three case studies were used, two related to wheat culture, obtained from different places and dates. Dataset 1 - Capão do Cipó Farm, in Castro - PR, 2003. Dataset 2 - Palmeirinha Farm in Piraí do Sul - PR, 2005 winter crop. Dataset 3 - Meteorological Station of ESALQ/USP in Piracicaba - SP, from July to September, 2005. A fourth case study was elaborated from datasets of the case studies 1, 2 and 3, using the common climatic variables together with their respective wetness indexes. After testing the training algorithms in the simulators, the Resilient Propagation (Rprop) presented lower training errors than the others evaluated methods: Backpropagation Standard, Backpropagation for batch training, Backpropagation with momentum term, Backpropagation with chunkwise update, Backpropagation with Weight Decay and Quickprop. It was verified 4 also that, among the climatic variables used for classification of leaf wetness from dew, the inclusion of the schedule had influenced in the obtaining better ANNs results... (Complete abstract, click electronic access below) / Orientador: Angelo Cataneo / Coorientador: Alaine Margarete Guimarães / Banca: Célia Regina Lopes Zimback / Banca: Marco Antonio Martim Biaggioni / Banca: Manoel Henrique Salgado / Banca: Marcelo Giovanetti Canteri / Doutor

Lhumectation des surfaces végétales, due principalement aux précipitations sous forme de pluie ou de rosée, joue un rôle déterminant lors de la phase de contamination des plantes par de nombreux agents phytopathogènes. La connaissance de la pluie et de la rosée constitue un élément fondamental pour létude et la compréhension du fonctionnement des modèles de simulation des épidémies et des systèmes d'avertissements agricoles. Lobjectif de cette recherche est de contribuer à lamélioration du système davertissement des principales maladies cryptogamiques affectant le blé dhiver au sud de Belgique et au G-D de Luxembourg. Notre démarche a consisté, dans un premier temps à évaluer les potentialités du radar météorologique de Wideumont. Nous avons décrit son fonctionnement général ainsi que son principe de mesure et nous avons détaillé les différentes sources derreur qui affectent les estimations de précipitations dérivées des observations radar. Les mesures radar sont moins précises que les mesures de précipitations par des pluviomètres. Néanmoins, le radar permet dobserver en temps réel les précipitations sur un large domaine avec une très bonne résolution spatiale et temporelle. La comparaison quantitative et qualitative des précipitations mesurées au sol avec celles estimées par le radar a été faite sur une période de trois ans (2003, 2004 et 2005). Les résultats de la validation des cumuls mensuels font apparaître que le radar a tendance à sous-estimer les précipitations. Lerreur calculée pour lensemble des stations varie entre -50% et +12%. La validation qualitative du radar a été réalisée sur des occurrences de cumuls horaires. Les indices calculés à partir des tables de contingence donnent des valeurs de POD (Probability Of Detection) entre 0.44 et 0.80 durant la période étudiée. Limpact des estimations radar sur les périodes dinfection de Septoria tritici simulées par PROCULTURE a été évalué durant trois saisons culturales (2003, 2004 et 2005) par comparaison entre les données de sortie du modèle (alimenté par des estimations radar de précipitations horaires) et les estimations visuelles du développement des symptômes de la maladie sur les trois dernières feuilles. Les outputs de PROCULTURE via les données radar ont montré un grand accord entre la simulation et lobservation. Le radar météorologique devrait dès lors être bénéfique pour des régions où le réseau des pluviomètres est inexistant (ou moins dense) et où lincidence de la septoriose est importante. Dans un deuxième temps, sur base dune recherche bibliographique, un modèle dhumectation a été choisi. Le modèle sélectionné, appelé SWEB, se base sur le bilan énergétique et le bilan hydrique. Il simule la durée dhumectation due à la pluie et à la rosée sur lensemble du couvert végétal à partir des données issues des stations agrométéorologiques. Le modèle a été ensuite testé et validé sur différentes variétés de blé dhiver. Les données de sortie du modèle ont été comparées statistiquement aux mesures des capteurs (préalablement calibrés) et aux données dobservation obtenues sur des parcelles expérimentales et au champ durant les saisons culturales 2006 et 2007. Sur base des résultats obtenus, le modèle SWEB semble sous-estimer la durée dhumectation et plus particulièrement pour les événements de la fin dhumectation (dryoff). Lerreur moyenne en général est inférieure à 90 minutes. Dans un troisième temps, afin dobtenir une relation entre les périodes dhumectation et le développement de la septoriose sur les trois dernières feuilles, les périodes dhumectation simulées par SWEB ont été comparées dune part aux périodes dinfection de Septoria tritici simulées par PROCULTURE et dautre part aux estimations visuelles. Le modèle de la durée dhumectation simule avec succès des périodes dhumectations, dues à la fois à la rosée et à la pluie, qui ont déclenché linfection de la septoriose observée sur des parcelles expérimentales. Une durée minimale dhumectation favorable à linfection des feuilles de blé par Septoria tritici a été déterminée. Il est donc désormais nécessaire délaborer un système opérationnel intégrant le radar météorologique, le modèle de la durée dhumectation et le modèle épidémiologique. Notre travail a permis dacquérir via lanalyse des données agrométéorologiques et des données phytopathologiques, les connaissances nécessaires à lélaboration dun tel système et de participer ainsi à lamélioration des modèles davertissements existants. En effet, nous avons analysé les avantages et les limites du système radar comme données dentrée aux modèles et son aptitude dans la spatialisation des données. Nous avons également testé le modèle dhumectation pour la détermination des périodes dinfection nécessaires au développement de la septoriose. Dans une perspective dune meilleure opérationnalisation du système, lapproche envisagée pourrait facilement être intégrée dans le système existant pour la simulation dautres maladies comme les rouilles, loïdium et la fusariose à léchelle régionale. En définitive, ce travail aura prouvé une fois de plus lintérêt du "mariage" entre lagrométéorologie et la phytopathologie. [en] Summary - Weather-Radar Rainfall Measurement and Simulated Surface Wetness Duration for Septoria Leaf Blotch Risk Assessment. The persistence of free moisture on leaves, mainly as a result of precipitation in the form of rainfall or dew, plays a major role during the process of plant infection by most fungal pathogens. Acquiring rainfall and leaf moisture information is needed for accurate and reliable disease prediction and management. The objective of this research is to contribute to improve forecasting Septoria leaf blotch and other fungal pathogens on winter wheat in Belgium and Luxembourg./In the first part of this work, the potential of weather-radar rainfall estimates for plant disease forecasting is discussed. At first step, we focused on assessing the accuracy and limitations of radar-derived precipitation estimates, compared with rain-gauge data. In a second step, the Septoria leaf blotch prediction model PROCULTURE was used to assess the impact on the simulated infection rate of using, as input data, rainfall estimated by radar instead of rain gauge measurements. When comparing infection events simulated by PROCULTURE using radar-derived estimates and reference rain gauge measurements, the probability of detection (POD) of infection events was high (0.83 on average), and the false alarm ratio (FAR) of infection events was not negligible (0.24 on average). FAR decreased to 0 and POD increased (0.85 on average) for most stations, when the model outputs for both datasets were compared against visual observations of Septoria leaf blotch symptoms. Analysis of 148 infection events observed over three years at four locations showed no significant difference in the number of simulated infection events using either radar assessments or gauge measurements. This suggests that, for a given location, radar estimates are just as reliable for predicting infection events as rain gauges. As radar is able to estimate rainfall occurrence over a continuous space, unlike weather station networks that do observations at only a limited number of points, it has the great advantage of being able to predict the risk of infection at each point within an area of interest with an accuracy equivalent to rain gauge observations. This gives radar an important advantage that could significantly improve existing warning systems. In the second part, a physical model based on the energy balance, known as the Surface Wetness Energy Balance (SWEB), was applied for the simulation of Surface Wetness Duration (SWD) on winter wheat canopy. The model, developed in the United States on grapes canopies, was adapted for the winter wheat cultivars and was applied for use with agrometeorological data easily available from standard weather stations and weather-radar rainfall estimates. The SWEB model simulates surface wetness duration for both dew and rain events. The model was validated with data measured by sensors and with visual observations of SWD conducted in experimental plots during two cropping seasons in 2006 and 2007. The wetness was observed visually by assessing the presence or absence of surface water on leaves. Based on the results, the SWEB model appeared to underestimate surface wetness duration and especially for the dry-off events when compared statistically to visual observations. The error, on average, is generally less than 90 minutes. In order to establish a relationship between the surface wetness periods and Septoria leaf blotch development risk on the top three leaves, the SWEB model SWD outputs were compared with the number of hours of high probability of infection simulated by PROCULTURE as well as with visual plant diseases observations. A minimal surface wetness duration of favourable infection conditions for Septoria tritici was established. It is now required to develop an operational system that would integrate weather radar, surface wetness duration and foliar epidemic model. In this work, we have analyzed the advantages and limitations of the radar system as input to models and its ability for spatial interpolation of rainfall. We also tested the model for the determination of surface wetness periods required for Septoria Leaf Blotch Risk development. The proposed approach could be integrated in the existing system. Finally this approach shows once more the "happy marriage" between agrometeorology and plant disease management.