Projected changes in rainfall seasonality and interannual variability are expected to have severe impacts on arid and semi‐arid tropical vegetation, which is characterized by a fine‐tuned adaptation to extreme rainfall seasonality. To study the response of these ecosystems and the related changes in hydrological processes to changes in the amount and seasonality of rainfall, we focused on the caatinga biome, the typical seasonally dry forest in semi‐arid Northeast Brazil. We selected four sites across a gradient of rainfall amount and seasonality and analysed daily rainfall and biweekly Normalized Difference Vegetation Index (NDVI) data for hydrological years 2000 to 2014. Rainfall seasonal and interannual statistics were characterized by recently proposed metrics describing duration, timing and intensity of the wet season and compared to similar metrics of NDVI time series. The results show that the caatinga tends to have a more stable response with longer and less variable growing seasons (3.1 ± 0.1 months) compared to the duration wet seasons (2.0 ± 0.5 months). The ecosystem ability to buffer the interannual variability of rainfall is also evidenced by the stability in the timing of the growing season compared to the wet season, which results in variable delays (ranging from 0 to 2 months) between the peak of the rainfall season and the production of leaves by the ecosystem. The analyses show that the shape and size of the related hysteresis loops in the rainfall–NDVI relations are linked to the buffering effects of soil moisture and plant growth dynamics. Finally, model projections of vegetation response to different rainfall scenarios reveal the existence of a maximum in ecosystem productivity at intermediate levels of rainfall seasonality, suggesting a possible trade‐off in the effects of intensity (i.e. amount) and duration of the wet season on vegetation growth and related soil moisture dynamics and transpiration rates. Copyright © 2016 John Wiley & Sons, Ltd.