Rio Tinto uses a complex hydrological model to make ensemble predictions (ESP) for expected freshet volumes that are vital to planning the management of their hydro-electrical power plants. However, the ESP predictions show an under-dispersion problem that is apparent in Talagrand histograms used for evaluating model performance. This problem has been successfully solved for the winter season but the summer season presents unique challenges. During the workshop, we proposed four different approaches to correcting the under-dispersion problem in summer : optimization of the initial state in summer , transfer function models, Gaussian processes and one dimensional modeling.