Résumé
Canopy phenology of Acacia tortilis ssp. raddiana, a dominant semi-deciduous species of the northern Sahelian zone, was monitored for 39 mature individuals, each month and biweekly in the rainy season, over a 5.5-year period in North Senegal. To investigate the relationships between leaf phenology and environmental variables, soil water availability and several climatic variables were monitored. Over six rainy seasons, annual rainfall ranged between 146 and 367 mm. The full canopy stage lasted between 5 and 8 months, broadly including the rainy season (July-September) and the ''cool'' dry season (November-January). A significant inter-annual variation, up to 2.0 months, affects both the timing of the peaks of leaf flush and leaf fall. The canopy was maintained during the dry season despite low upper soil water availability and tree roots had access to a deep water table (31 m). These results support the current view that in the dry tropics, groundwater availability is the major environmental variable controlling leaf phenology. However, inter-annual variation in the peaks of leaf flush and leaf fall could not be explained by ground water, genetics or day length. In such water-controlled biome, we focused on a comparison between two additional drivers, upper soil water availability and climatic variables which contribute to evaporative demand. Models predicting changes in canopy fullness from environmental variables were investigated by polynomial logistic regression. We considered each tree and pooled all the years, distinguishing periods of leaf flush (April-August) and leaf fall (January-April). Then, the ability of such models to predict inter-annual variation in the timing of peaks of leaf flush and leaf fall was tested. Inter-annual variation in the timing of leaf flush peak was well predicted by models based on air relative humidity or vapour pressure deficit or global radiation (root mean square error = 0.5 month and R 2 = 0.8). Inter-annual variation of leaf fall peak was also significantly predicted by models based on atmospheric variables (temperatures or maximum value of vapour pressure deficit) however with weaker relationships (root mean square error = 0.7 month and R 2 = 0.7). By contrast, models based on upper soil water availability or rainfall did not predict either leaf flush or leaf fall inter-annual variation.