Sandra Plancade, INRA

Statistics seminars
Sandra Plancade, INRA, France and ISBA
''Survival analysis models for plant development ''

Plant development is usually modeled through phases which affect simultaneously various traits of the plant. While some traits are directly observable (heights, number of leaves, flowering...), internal traits require a destructive observation, which by essence can not be repeated. Therefore, a question of interest is to get information from the internal processes based on observation from the external traits.

In order to circumvent the problem of destructive observations, the usual approach consists in agregating measurements from repeated dissections within a set of pre-selected homogeneous plant as if they originate from the same plant. While this approach allows to develop sophisticated mechanistic models, it does not account for plant-level variability and is therefore limited to laboratory analyses.

In the context of the INRA project ITEMAIZE, we aim at modeling at the plant level the joint distribution of two processes, based on a field experiment on maize : on the one hand, the floral transition, characterised by a change of the shape of an internal organ, and on the other hand, the process of leave appearance characterised by the time of appearance of each leaf. The data consist in repeated measurements of the number of leaves, and dissections indicating if the floral transition already occured. From a statistical point of view, this framework corresponds to interval censoring and current status data, respectively.

In this presentation, I will first present the process which led us to the formulation of our problematics and model, in order to illustrate the necessary upstream work in the context of an open applied research question. Then, I will present our approach based on the AFT (Accelerated Failure Time) model, the estimation algorithm and the first results, as well as the extension we are currently developing.