This video in our Ecological Forecasting series introduces Bayesian hierarchical models as a way of capturing observable, but unexplained, variability in processes by allowing model parameters to vary probabilistically. Considering the simple case of modeling data from multiple observation units (sites, plots, lakes, etc.), the hierarchical approach is contrasted with the traditional alternatives of lumping unit-to-unit variability versus fitting different units independently. From a forecasting perspective, hierarchical models also provide a natural means of formally distinguishing differences in within-unit versus outside-of-sample predictive uncertainty.