Abstract
Yam is vital for food security and income in West Africa. However, population growth and climate change are calling into question current production methods, which focus on Dioscorea rotundata. The species D. alata offers a credible alternative because of its superior yield and adaptation to difficult growing conditions. However, despite the improvements made, there has been a low rate of adoption of the new varieties, mainly due to the lower quality of the tubers because of the inability of breeding programmes to assess this quality on a large scale. This thesis focuses on the development of phenotyping methods for assessing the growth and quality of yam (D. alata).Initially, we developed a phenotyping method for emergence vigour and ground cover dynamics. This method extracts the main characteristics of emergence and growth via mathematical modelling and an image analysis pipeline respectively. Validation of this pipeline, based on images from a variety of sensors, confirms the robustness of the methods used.The quality assessment of yam tubers focused on their composition (dry matter, starch content, proteins, amylose and soluble sugars), the structure of the starch granules and functional characteristics such as cooking time and mouldability. While the characterisation of starch granules rely on the analysis of microscopic images, near infrared spectroscopy is used to estimate all the other traits simultaneously. The models developed to predict these traits vary in performances depending on the data available. For protein, soluble sugar and starch content, external validation confirms the robustness of the models and their potential use in yam improvement programmes. On the other hand, the predictions for amylose content, pilability and cooking time lack performance on external data sets. Renewing the reference measurements or using classification models is suggested to improve these predictions. For example, using convolution neural networks to predict whether yam tubers are mouldable or not by classification, we were able to make a correct prediction in 80% of cases.Although mouldability is not the only criterion for acceptability, it is a necessary one. This is why this trait was used as an indicator of overall quality, equated with acceptability. Mouldability appears to be strongly correlated positively with dry matter, starch content, starch granules size and amylose content, and negatively with sugar content.On the basis of measurement quality (estimated by repeatability), improvement potential (estimated by genotypic variability) and heritability (H²), emergence time, duration of growth of ground cover plateau and maximum ground cover are the three agronomic traits that we propose to retain for future studies. With regard to quality, cooking time showed very low heritability and protein content showed insufficient repeatability and low heritability. The other quality traits show good measurement repeatability and significant potential for improvement (H² and minimum and maximum values).This thesis therefore made it possible to develop the phenotyping methods needed to study the determinants of quality. Based on the quality of the measurements and the potential for improvement, we recommend focusing further studies on the following quality traits : starch content, soluble sugars and dry matter, size and shape of starch grains and mouldability.