A key goal in ecology is to predict where species occur accurately. Accurate predictions not only reflect a good biological understanding, but are also crucial to inform environmental management. Species distribution models, however, are hindered by narrow prediction horizons: they often describe local or snapshot patterns of species abundance well, but transfer very poorly to new locations, new time periods, or new taxa. This transferability problem impedes the sharing of knowledge from one study system to another, thus forcing researchers to collect data and fit models from scratch whenever they confront new species or new environments — essentially reinventing the wheel. We aim to develop a prediction horizon framework for identifying distribution models that explain how traits govern species adaptations to local environments while also projecting well into future, distant conditions by combining one of the largest vegetation datasets in New Zealand with extensive trait and environmental measurements. The developed framework will facilitate knowledge transfer across study systems, and thereby improve the prediction accuracy of the spread of undesired species or the decline of species vulnerable to global change.