In rhythmic tree growth, the alternation between growth and resting phases can be either periodic or irregular, depending on climatic and endogenous factors. The aim of this study was to analyze the growth pattern of yerba-mate, a subtropical South American tree with monopodial and rhythmic growth, over a two-year period. Metamer emission rate of selected axes was followed during two years after a severe pruning in two cultivation systems contrasting for light conditions, agroforestry system and monoculture. A new longitudinal data modeling approach relying on hidden semi Markov chains to identify growth and resting phases and multiple change-point models to relate these phases to climatic factors was used. Despite large variability among individuals, two growth phases separated by a resting phase were most often identified within each year. Contrasting situations regarding the modulation of the polycyclism were observed between the two years and cultivated systems: in the first year of growth, the pattern consisted of a single long growth phase in vigorous individuals and the cultivation system had a major effect, likely due to the induced light environment; in the second year, some individuals did not grow during the second autumn phase likely due to the drought during the warm season. The observed differences in growth patterns between years and cultivation systems were interpreted with respect to ontogenetic and climatic effects, in interaction with endogenous factors resulting from plant reproductive phenology. This study introduces a new longitudinal data analysis approach for investigating the phenology of perennial plants over long follow-up periods, at time scale intermediate between days and years.