Plant phenology, the study of seasonal plant activity driven by environmental factors, has found a renewal in the context of global climate change. Phenological events, such as leaf unfolding, exert strong control over seasonal exchanges of matter and energy between the land surface and the atmosphere. Phenological models that simulate the start of the growing season should be efficient tools to predict vegetation responses to climatic changes and related changes in energy balance. Species-specific phenological models developed in the eighties have not been used for global-scale predictions because their predictions were inaccurate in external conditions. Recent advances in phenology modelling at the species level suggest that prediction at a large scale may now be possible. In the present study, we tested the performance of species-specific phenological models in time and space, looking at their ability (i) to predict regional phenology when previously fitted at a local scale, and (ii) to predict phenological trends, linked to climate changes, observed over a long-term. For that task we used an historical phenological dataset from Ohio from the late ninetieth century and an airborne pollen dataset from Ontario, Québec and Maryland from the late twentieth century. The results show that the species-specific phenological models used in this study were able to predict regional phenology even though they were fitted locally. The reconstruction of a phenological time series over the twentieth century showed a significant advancement of 0.2 days per year in the date of flowering of Ulmus americana, but very weak trends for Fraxinus americana and Quercus velutina.