The management of tropical rain forests in Central Africa is an essential issue because of the economic importance of the production of timber for the country and of the resource supply to local populations (non wood forest products, proteins via hunting). The sustainable management of these forests often relies on population dynamics models for size-structured tree populations (also called matrix model) that describe forest dynamics and consequently can be used to predict the temporal evolution of the timber stock. The uncertainty on model predictions is directly related to the precision of estimation of the transition parameters of the model. These parameters, also called vital rates, include growth, recruitment and mortality rates. There are two main sources of variability in parameter estimates : sampling variability, and environmental variability. Sampling variability depends on the amount of available data. As tropical rain forests have a high number of species with many rare species, most species-specific parameter estimates have huge errors. One way to solve this problem is to group species with common behaviour to increase the number of available observations. Environmental variability is related to the spatial and temporal variations of transition parameters due to environmental fluctuations (such as climate or soil). This kind of variability is not yet considered in the models used by forest managers. In this study, we address climate variability (rainfall) in forest dynamic predictions and group species according to their response to rainfall. First, the species classification and the relation between growth, mortality and recruitment rates, and climatic covariates for each species group were simultaneously fitted using finite mixture of regression models. Data come from permanent sample plots (40 ha, 25-year monitoring) located at M'Baïki, in the Central African Republic. The plots are located in a semi-deciduous mixed forest, where the climate has a pronounced dry season and where the average annual rainfall is 1739mm. The climatic covariates used are the length of the dry season, the average rainfall during the dry season, and the annual average soil water content. The response of growth, mortality, and recruitment to climatic covariates varied among species. Nine response groups were identified for growth, three for mortality, and five for recruitment. The response groups based on growth showed a correlation between response to drought and species shade-tolerance. Second, we predicted stand dynamics which incorporates rainfall variability and variability in species response to rainfall. Stand dynamics was predicted for three climate scenarios : increase of drought, increase of rainfall, or no change in precipitation. The response of the forest stand was analyzed in terms of changes in stand structure (basal area and density), and relative composition in the species groups previously defined. The analyses showed a gradient in species drought tolerance, which opposed 9 predominantly pioneer species that responded negatively to drought, to 60 predominantly shade-tolerant species that responded positively to drought. The M'Baïki forest stand seems to be relatively drought resistant. Moreover, the response to drought seems to be driven by the mortality response to the climatic covariates, with shade-tolerant species having a lower mortality during the dry season, probably due to an increase in light availability in the understory as a consequence of a longer period of defoliation of canopy deciduous trees during drought. In this study, we also showed that logging increased the proportion of pioneer species in the stand. Consequently, an increase in logging pressure would result in a more drought-sensitive forest stand.
