Climate change is expected to increase the frequency and severity of precipitation extremes, causing droughts and flooding. Consequently, grassland communities are forecasted to become increasingly unstable. To predict grassland responses, we need empirical information together with models that reliably extrapolate community dynamics from those observations. However, such prediction is challenging because community models typically simulate long-term (asymptotic) performance, and thus potentially neglect their short-term (transient) performance. Here, we use data from a precipitation experiment performed over eight years to model both short- and long-term responses of three functional groups (grasses, legumes, and non-leguminous forbs) to precipitation extremes. We use multi-functional-group Integral Projection Models and pseudospectral theory, to track grassland community dynamics. We show that the percentage-cover-stage-structure of functional groups shapes their transient instability, and that inter-functional-group interactions are competitive under increased precipitation but facilitative under decreased precipitation. IPMs and pseudospectra enable forecasting of how functional-group-stage-structure drives responses to climatic extremes.