Proportional loss models commonly used in disease surveys are based on the assumption that per cent yield loss is the same in all crops, regardless of their yield potential. Estimates of regional crop loss may be inaccurate if the relationship between disease and yield loss is affected by crop yield potential. The importance of crop yield potential in disease: yield loss modelling was investigated and models for more accurate regional crop loss estimates were developed, taking crop yield potential into account. Two spring sown barley (cv. Triumph) experiments were conducted in 1987/88 and 1988/89 in Canterbury, New Zealand, to study the effect of crop yield potential on the relationship between disease and yield loss. Crop yield potentials of 323 to 806gDM/m² were generated in seven crops by varying nitrogen and water inputs, sowing date (mid-spring and early-summer) and season. Leaf rust (Puccinia hordei Otth) epidemics of different severity were generated by applying fungicides at different times, frequencies and rates to control the natural epidemics. Disease was measured as per cent disease severity (%DS), green leaf area, radiation interception and near-infrared radiation (NIR) reflectance from crop canopies. Yield was measured as total and grain dry weight. Epidemics were severe in the fully diseased plots from GS 34 and 46 to maturity in the late and early sown crops respectively. Disease reduced grain yield by 50 to 63% in 1987/88 and 24 to 38% in 1988/89 in the fully diseased plots. Disease: yield loss models were derived by regression analysis for each crop in 1987/88. Single point, multiple point and area under curve models were derived from %DS and GLAI variables, and proportional (%) and actual (gDM/m²) grain yield. The effect of yield potential was determined by comparing regression equation coefficients for each crop with crop yield potential. An area under green leaf area index curve (AUGLAIC): actual yield model was best suited to determining the effect of yield potential on yield loss. This model was selected because AUGLAIC summarised the effect of disease on plant growth over the season and actual yield represented the crop yield potential in the absence of disease and the response of actual yield to disease. Crop yield potential did not affect actual yield loss caused by leaf rust. Disease measured as AUGLAIC explained most of the variation in yield (R²adj=0.93) for all crops in both years. Assessment of GLAI is not suitable for estimation of regional crop loss because of the requirement for a rapid and low cost method. Reflectance of NIR from the crop canopy was investigated as an alternative to GLAI measurements. Reflectance was correlated significantly (P<0.001) with GLAI (r=0.66 to 0.89) and green area index (r=0.76 to 0.92). Reflectance measured at grain-filling (GS 85-87) explained most (R²adj=0.94) of the variation in yield for all crops in both years. The relationship between AUGLAIC and yield was validated with data from independent diseased and healthy barley crops. The AUGLAIC: yield model described the effects of disease on yield accurately but overestimated yield by 49 to 108% in the healthy crops. Models based on accumulated PAR (photosynthetically active radiation) intercepted by green leaves explained the observed deviations in yield of these crops from the AUGLAIC: yield model. Accumulated PAR models accounted for differences in incident radiation, canopy structure, radiation interception by green leaves, radiation use efficiency and harvest index which are important in determining dry matter production and grain yield. Accumulated PAR models described the effects of disease on crop growth which were not represented by GLAI alone. Variation in crop yield potential at the regional scale is important in disease: yield loss modelling and can be accounted for by using either separate equations for each yield potential crop or crop category, robust models, inclusion of a form function for yield potential or choice of disease and yield variables which integrate yield potential.

D’ici la fin du siècle, les prévisions climatiques prévoient une diminution de la quantité et de la régularité des pluies s’accompagnant d’une augmentation du risque de sècheresse en Europe et dans de nombreuses régions du monde. La création de nouvelles variétés de maïs plus tolérantes au stress hydrique est un levier indispensable pour faire face à ces contraintes futures. L’objectif principal de cette thèse est d’approfondir les connaissances des déterminismes génétiques de la tolérance à la sècheresse chez le maïs. Pour ce faire, il est proposé de disséquer ce caractère complexe en caractères physiologiques sous-jacents dont le déterminisme génétique est a priori plus simple. L’évolution de l’indice foliaire vert (GLAI : Green Leaf Area Index) au cours du cycle de la plante, par son rôle majeur dans l’interception lumineuse, la transpiration et les échanges de CO2, est un caractère secondaire prometteur pour identifier les bases génétiques de la tolérance à la sècheresse et en améliorer la compréhension. Au cours de cette thèse, nous avons développé une méthode de phénotypage haut débit permettant d’estimer la cinétique du GLAI au champ. Cette méthode combine la caractérisation multispectrale par drone et l’utilisation d’un modèle physiologique simple de GLAI. Elle permet d’estimer la cinétique du GLAI de manière continue sur l’ensemble du cycle de la plante avec une bonne précision, tout en divisant par vingt le temps nécessaire au phénotypage. Nous avons utilisé cette méthode lors de deux essais en conditions optimales et deux essais en conditions de stress hydrique pour mesurer l’évolution du GLAI au sein d’un panel de 324 lignées issues d’une population MAGIC (Multi-parent Advanced Generation Inter-Cross). Les cinétiques estimées présentent une forte héritabilité et expliquent une part significative du rendement en conditions optimales et stressées. Afin d’identifier les bases génétiques de la cinétique du GLAI, trois approches de génétique d’association longitudinales ont été comparées : une approche univariée en deux étapes, une approche multivariée en deux étapes et une approche de régression aléatoire en une étape. Ces trois approches, couplées à la forte densité des données de génotypage disponibles (près de 8 millions de marqueurs), ont permis de révéler de nombreux QTL (Quantitative Trait Loci), dont certains colocalisent avec des QTL de rendement. Enfin, nous avons démontré que les QTL de GLAI identifiés lors de cette étude pouvaient expliquer près de 20 % de la variabilité du rendement observée dans un large réseau d’expérimentations sous stress hydrique. Ce travail fournit des méthodes qui permettront une meilleure caractérisation et une meilleure compréhension des déterminismes génétiques de la cinétique du GLAI, un caractère jusqu’ici inaccessible pour les populations de taille importante. Ce caractère présente toutes les caractéristiques requises pour améliorer l’efficacité des programmes de sélection en conditions de stress hydrique. / By the end of the century, climate forecasts predict a decrease in the quantity and regularity of rainfall with an increasing risk of drought in Europe and in many regions of the world. Breeding for more tolerant varieties will be an essential lever to face these future constraints. The main objective of this work is to characterize the genetic determinisms of drought tolerance in maize. To this aim, it is proposed to dissect this complex trait into underlying physiological traits whose genetic determinism is supposed to be simpler. Green Leaf Area Index (GLAI) dynamics throughout the plant cycle, through its major role in light interception, transpiration and CO2 exchange, is a promising secondary trait to identify and better understand the genetic basis of drought tolerance. During this thesis, we developed a high-throughput method for phenotyping maize GLAI dynamics in the field. This method combines UAV multispectral imagery and a simple GLAI model. It makes possible the estimation of the dynamics of GLAI continuously throughout the whole plant cycle with good accuracy, while reducing the phenotyping time twentyfold. This method was used in two well-watered and two water-deficient trials to characterize the GLAI dynamics of 324 lines from a MAGIC population (Multi-parent Advanced Generation Inter-Cross). The estimated dynamics have a high heritability and explain a significant part of grain yield under well-watered and water-stressed conditions. To characterize the genetic basis of GLAI dynamics, three longitudinal GWAS (Genome Wide Association Study) approaches were compared: a univariate two-step approach, a multivariate two-step approach and a random regression one-step approach. These three approaches, combined with the high density of available genotyping data (nearly 8 million markers), have revealed many QTL (Quantitative Trait Loci), some of which were co-localized with yield QTL. Finally, we demonstrated that the GLAI QTL identified in this study could explain nearly 20 % of the grain yield variability observed in a large network of water-stressed experiments. This work provides methods that will enable a better characterization and understanding of the genetic determinisms of GLAI dynamics, a trait that was out of reach in large populations until now. This trait presents all the characteristics required to improve the effectiveness of selection programs under water stress conditions.