Résumé
The problem of analyzing spatial structures in forest tree architectures is addressed. Data samples consist of sequences of homogeneous botanical entities (node or annual shoot) where attributes are recorded for each successive entity (type of axillary production at the node level or length of the annual shoot). The data analysis methodology proposed includes various tools for exploratory analysis and the building of stochastic processes (hidden semi-Markov chains) from samples of discrete multivariate sequences whose aim is to capture the structure embedded in these sequences. This methodology is illustrated at the node level on growth units of red oak main stems and, at the annual shoot level, on corsican pine main stems.